10:28 <karlp> PaulFertser: strace helped thanks, found I had a hookspath configured to git-hooks, instead of .git/hooks :|

13:23 <PaulFertser> karlp: heh, tricky

17:28 <drath42> Hi everyone. Is anyone of you aware of any efforts of adding ARM tracing features to OpenOCD? I'm mainly looking at "larger" SoCs like NXP i.MX8*, TI AM6*, that come with ETBs/ETRs and can trace without dedicated trace hardware.

17:31 <drath42> I found some patches on Gerrit from Jeremy Garff that add support for trace replicator, trace funnel and trace memory controller, but nothing for the ETM itself

17:32 <drath42> (apart from the very old etm.[ch] code)

17:45 <PaulFertser> drath42: hey. Current tracing is possible with orbtrace project. But you normally need dedicated hardware to capture it.

17:45 <PaulFertser> With ETB, hm, I'd still ask them.

18:05 <drath42> PaulFertser: thanks for the pointer. It looks like they're focusing on Cortex-M and SWO/TPIU (1-4 bit parallel). On most of the real-world hardware designs I've seen these pins just aren't available, and there's the question about signal integrity, once you clock these interfaces reasonably fast.

18:09 <drath42> PaulFertser: On the i.MX 8M Plus there's hardware that should be able to trace directly into DDR memory - that's large and easily accessible for post-processing. Decoding can be done with OpenCSD - I've used that with trace data collected from an AM6* ETB already, because TI's own software is rather restricted, but the process was cumbersome, involved many manual steps, and left most of the hardware trace capabilities unused... I

18:09 <drath42> was thinking that OpenOCD could do a better job

18:12 <PaulFertser> drath42: I see. Unfortunately, nobody has contributed the relevant code yet afaict.

18:26 <drath42> PaulFertser: yeah, I'm looking into that, i.e. just how much is missing, and what existing projects might be reused. I just learned about CSAL (Coresight access library). That along with OpenCSD looks promising...

21:38 <borneoa_> drath42: https://github.com/ARM-software/CSAL/blob/master/LICENSE

21:39 <borneoa_> drath42: unfortunately apache-2.0 license is not compatible with GPLv2 used by OpenOCD

21:42 <borneoa_> drath42: anyway it should not be a blocking point. The difficult should be the decoding of the instructions trace. The trace is compressed and you need to parse both the trace and the binary assembly code in execution to get the sequence of execution

21:44 <borneoa_> Let me say better. The license is a blocking point and we cannot mix apache and GPL. But it's possible to rewrite the part of code we need