[PATCH] [PowerPC]Adapt fast stack unwinding to work for Power.
eugenis at google.com
Tue Jun 16 11:28:23 PDT 2015
I agree that the earlier solution is better than relaxing tests to meet the broken unwinder behavior. Still, it papers over the problem instead of fixing it. As I explained in earlier comments, this code does not make sense; unwinding starts from exactly the frame we want it to start and the should be no extra frames to remove. This means that something else is broken, maybe the way we obtain current frame address? Also note that this extra frame means we made one step of frame pointer based unwind through sanitizer library code, which is intentionally built w/o frame pointers. This can break at any time.
I accept that this could be the easiest way to work around the (unknown) problem, and a platform-specific #ifdef is not the end of the world. But I'd like to understand (and leave a comment in the code) what is actually going on there.
More information about the llvm-commits