tests: enable helgrind and drd checkers for 'threads' tests

[lukaszstolarczuk]

I believe there might be some issue in threads mover, reported by helgrind/drd... or just a missing suppress.

---

This change is 

[igchor]

I think this load should be atomic: https://github.com/pmem/miniasync/blob/master/src/core/ringbuf.c#L228

and there should be some happens before/after annotation for storing/loading completion flag: https://github.com/pmem/miniasync/blob/master/src/data_mover_threads.c#L168

This is what thread sanitizer reported.

[lukaszstolarczuk]

huh, ok, so your proposed fixed is required... (or at least a part of it)?
// pmem/pmemstream#235 (comment)

[lukaszstolarczuk]

we may also need to add suppressions/annotations for thread-sanitizer.

See TSAN support in pmemstream: https://github.com/pmem/pmemstream/pull/234/files
See our annotations: https://github.com/pmem/pmemstream/pull/236/files#diff-04996d3777968a283c7157ae594beacf8a4642d699a2c6e21a558213e867ab37

[pbalcer]

I fixed the ringbuf stuff in #103. Can you show me an example of how to do the annotations?

[igchor]

@pbalcer I think you only need to add __tsan_acquire/release (https://github.com/llvm/llvm-project/blob/main/compiler-rt/include/sanitizer/tsan_interface.h#L24) where you would normally put memory fences. I also belive that you could just add __tsan_acquire as part of VALGRIND_ANNOTATE_HAPPENS_AFTER (and similarly for release).

E.g. for data_mover_threads.c you would need to add __tsan_release before

miniasync/src/data_mover_threads.c

Line 129 in 2490b0c

util_atomic_store_explicit64(&data->complete, 1, memory_order_release);

and __tsan_acquire after

miniasync/src/data_mover_threads.c

Line 168 in 2490b0c

util_atomic_load_explicit64(&tdata->complete,

For ringbuf.c, I'm actually not sure how many extra annotations are needed (I did not run any miniasync tests with tsan). In general, the safest approach would probably be to annotate every atomic store with memory order different than relaxed