J-link session teardown breaks modm::delay/DWT->CYCCNT

[WasabiFan]

By default, J-link will clear DWT->CTRL.CYCCNTENA when disconnecting the debug session. It's documented here: https://kb.segger.com/J-Link_Cortex-M_application_uses_cycle_counter . This means in practice that if I disconnect a J-link session (including e.g. a script which flashes, un-halts, and immediately disconnects) the next modm::delay will hang forever. I have found the SetSkipDebugDeInit setting works as advertised to avoid this behavior.

I don't have a particular opinion about whether it's right for J-link to mess with the DWT setting vs. modm depending on DWT, but I figured I'd share and ask for thoughts: is there anything to be done? Is this a known issue? The behavior is inconvenient and can consume some debug time trying to figure out what went wrong.

Some ideas:

• Have modm::delay automatically set CYCCNTENA before beginning
• Have modm::delay assert that CYCCNTENA is set

Neither will help if the bit gets cleared during a sleep (unless done on every loop) but at least would minimize the chances of burning an afternoon investigating. Documenting it in the modm::delay module docs would be good to although I doubt people will go there to look.

[salkinium]

Oh, I didn't know that. I think the best way would be to put mon SetSkipDebugDeInit = 1 into this file for the jlink case and see if that circumvents the problem.

[WasabiFan]

Potentially. I'm not sure whether most users run those scripts. At least personally, since it's so easy to pass the elf to whatever probe software I'm using that day, I haven't used them much. It would be an improvement, in any case.