[compiler-rt] [libcxx] [libcxxabi] [libunwind] [runtimes][PAC] Harden unwinding when possible (PR #143230)

[Oliver Hunt via cfe-commits]

ojhunt wrote:

> The patch looks fine to me but you should finish addressing comments by @kovdan01 .
> 
> I wonder, however, what kind of test coverage we have for your changes. Do we have any? In the current CI, are we even exercising these changes?

These changes do not effect behavior if there is no pointer authentication - if there is pointer authentication then these have to be consistent with the host platform so I'm not sure how to test conformance without adding what are in essence "expected fail" tests where we construct intentionally incorrect frames and attempt to unwind through them. Having this upstreamed would permitted arm64e apple silicon test bots to run which might be useful?

https://github.com/llvm/llvm-project/pull/143230