[Lldb-commits] [lldb] Reapply "[lldb/aarch64] Fix unwinding when signal interrupts a leaf f… (PR #92503)

[Jason Molenda via lldb-commits]

jasonmolenda wrote:

btw @labath I was thinking about "sigtramp routines" and fault / trap / interrupt handlers in general, and how lldb has this list of function names that it treats as fault handlers (`target.trap-handler-names`).  And in the unwinder we have the same idea of "the frame above a designed sigtramp/fault/trap/interrupt function can retrieve all registers".  

But what if the unwinder had a method saying "Do you have an unwind rule to give me a value for register n, without iterating down the stack?"   e.g. frame 5 is interrupted and frame 4 is sigtramp, with full eh_frame.  From frame 5, I can say "does frame 4 have an unwind rule to provide x0, without iterating down to frame 3, etc."   This also means if we have a sigtramp which is missing its eh_frame, we won't apply our "all registers available" rules to the frame above.

An interesting case for a return-address target like aarch64 where normally when we ask for a caller frame's pc, we fetch the link register.  But above a fault/trap/interrupt frame, we can retrieve both the pc and the link register and they are different values.

Just something I started kicking around in my head, I don't have concrete plans to implement an overhaul like this but the more I think about it, the more I like it.

https://github.com/llvm/llvm-project/pull/92503