[SKIP CI] Minor docker-run.sh fixes

[marc-hb]

See commit messages.

Fixes recent #6232

[marc-hb]

I'm really puzzled by the many (and unrelated) testbench memory corruptions reported by valgrind in https://github.com/thesofproject/sof/runs/8141544041

I tried both Ubuntu 20 and 22 and I could not reproduce the ... @singalsu, @pjdobrowolski any idea? How multithreaded and non-deterministic is this?

Only recent change I see that could be related is library manager

• library manager: initial implementation #5796

==19983== Thread 3:
==19983== Invalid read of size 1
==19983==    at 0x5FD5E9C: fprintf (stdio2.h:105)
==19983==    by 0x5FD5E9C: module_adapter_copy (module_adapter.c:599)
==19983==    by 0x486BF09: pipeline_for_each_comp (pipeline-graph.c:453)
==19983==    by 0x486CB4F: pipeline_comp_copy (pipeline-stream.c:101)
==19983==    by 0x486CB4F: pipeline_copy (pipeline-stream.c:139)
==19983==    by 0x486E487: pipeline_task (pipeline-schedule.c:203)
==19983==    by 0x4861C45: ll_thread (ll_schedule.c:131)
==19983==    by 0x4951B42: start_thread (pthread_create.c:442)
==19983==    by 0x49E2BB3: clone (clone.S:100)

==19983== Invalid read of size 8
==19983==    at 0x5FD53F8: module_adapter_copy (module_adapter.c:606)
==19983==    by 0x486BF09: pipeline_for_each_comp (pipeline-graph.c:453)
==19983==    by 0x486CB4F: pipeline_comp_copy (pipeline-stream.c:101)
==19983==    by 0x486CB4F: pipeline_copy (pipeline-stream.c:139)
==19983==    by 0x486E487: pipeline_task (pipeline-schedule.c:203)
==19983==    by 0x4861C45: ll_thread (ll_schedule.c:131)
==19983==    by 0x4951B42: start_thread (pthread_create.c:442)
==19983==    by 0x49E2BB3: clone (clone.S:100)

\+ many more.

[marc-hb]

And now re-run https://github.com/thesofproject/sof/runs/8142429253 is passing. Scary.

[pjdobrowolski]

@marc-hb library manager won't work unless it is turned on. I will look at this...

[lgirdwood]

@ranj063 please see the valgrind warning above, could be a race as it's occasional. It may need a if() check in the code.

[pjdobrowolski]

@marc-hb after short analysis failure is caused by not freeing comp https://github.com/thesofproject/sof/runs/8141544041#step:6:585 what leads to memory corruption. Issue is not related to library_manager, similar problem with freeing modules was being fixed on UpDownMixer however I'm not sure if it is connected. From my perspective issue is on TGL machine. FYI @jxstelter

[lgirdwood]

@wszypelt @lrudyX good to merge ? Not expecting docker update to fail a DMIC test ?

[wszypelt]

@lgirdwood To be sure, I turned on the tests again

[marc-hb]

@marc-hb after short analysis failure is caused by not freeing comp https://github.com/thesofproject/sof/runs/8141544041#step:6:585 what leads to memory corruption.

Thanks @pjdobrowolski for the preliminary analysis. Would you be able to go one extra step and git blame the corresponding code?

Also, does it look like a race to you?