TRACECLK and CPU clock

[eblot]

Hi,

I'm running a nRF52832 connected through TRACE port to a Segger J-Trace Pro.

Some traces seem to be lost, and Segger support team tells me it is possible as nRF52832 runs @ 64MHz, while the traceport runs @ 16MHz, which can therefore only output data @ 32MHz. Segger suggest to decrese the CPU clock speed @ 32 MHz so that no overflow occurs.

* Is is possible and how to do that with nRF52832 running SDK 15.2 and S132 6.1.1?

* If if is not possible, how to make the TRACE port reliable?

Thanks,

Emmanuel.

[Kenneth]

Afaik:

1. No, not possible.

 2. The speed of the processor will heavily depend on the cache hit/miss, and the flash accessing. In certain cases, where code is being executed from RAM, there can be potentials for trace overflow.

 

[eblot]

Ok, so that means there is no way to get a full valid trace output out of nRF52 devices, it all depends on parameters that cannot be tweaked or even guessed. Am I correct?

[Kenneth]

eblot said:

Am I correct?

Yes, though I expect if you disable cache that should help, so just call this before ble_stack_init()::

NRF_NVMC->ICACHECNF = 0x00;

[eblot]

Ok.

Is it something that will be fixed in the next iteration of the nRF52, because the ETM may be used, among other needs, to debug performance issues *without* altering the normal behaviour of the CPU, and it looks like the implementation of the ETM port by Nordic defeats this very purpose. It seems quite paradoxal to me.

Thanks

[Kenneth]

Yes, I expect this will be addressed, but I can't comment on when and how etc.

Best regards,
Kenneth