[compiler-rt] [RTSan][Apple] Disable AccessingALargeAtomicVariableDiesWhenRealtime (PR #129309)

[Chris Apple via llvm-commits]

cjappl wrote:

Alright, here is what I have for you:

https://gist.github.com/cjappl/be6544731e973497b58cd6af4a0e318f

This reproduces the problem in the simplest way I can think of, basically a re-hash of the unit test but standalone. 


>From my poking around on my machine, I think the possibly fastest way to finding the culprit is to look at the internal Apple implementation of `__atomic_load` in `libcompiler_rt`
`    #1 0x00019a27cb78 in __atomic_load+0xb4 (libcompiler_rt.dylib:arm64+0xb78)`.

On my machine, this eventually goes to `OSSpinLockLock`, one of our intercepted calls. On the problematic machine, it inserts something else that we seemingly do not intercept.  I am hopeful looking at the source points us to the function we are missing, or at least gives us more info. I was not able to get satisfying answers out of Instruments for this one.

> but I think ideally the people working on this should also try to get set up so that they can reproduce failures that pop up in the Apple CI. Happy to troubleshoot that with you as well.

I'd be more than happy to get this instruction! I also assume it is off the table, but if I were to get SSH access to this machine (or a machine that repros this problem) I would be more than happy to spend some time and see if I could figure it out.

Some possible hints on where this may be:
* Is it an STL version thing? A newer update, or an older version?
* Is it ARM vs X86?


Shout if you get any good leads, happy to help.

https://github.com/llvm/llvm-project/pull/129309