[llvm] [BOLT][AArch64] Support for pointer authentication (v2) (PR #120064)

[Gergely Bálint via llvm-commits]

bgergely0 wrote:

@paschalis-mpeis 

> Since this is a new feature, do you think it might be worth guarding it with a flag to make it more explicit to users? For example, fail with some advice when the flag is not present, otherwise process Negate-RA CFIs.

Good point, will do. 

> Regardless, I think it would be helpful to emit a BOLT-INFO stat summarizing what was done. So users are aware that this feature was used and applied to N places.

I am not sure if this would be very useful for users. 

We expect BOLT users to optimize big, complex programs (compilers, database engines, etc), so the generated amount would be huge. Almost all functions are affected. 

Because of this, it would be way easier for users to switch off the flag, and use a binary with `-mbranch-protection=none` if they face issues, than to debug which exact function's missing/extra RA state negation causes the unwinder to fail.

https://github.com/llvm/llvm-project/pull/120064