AArch64: Fix passing function pointers #5896
Conversation
agu-z
changed the title
| @@ -920,6 +920,8 @@ fn build_proc<'a, B: Backend<'a>>( | |||
| } | |||
| } | |||
| Relocation::LinkedData { offset, name } => { | |||
| add_undefined_rc_proc(output, name, &rc_proc_names); | |||
There was a problem hiding this comment.
isn't it weird that an RC proc ends up here? I'm happy it works but would like to understand why a bit beter
There was a problem hiding this comment.
@folkertdev Yeah, it took me a while to figure out what was going on here, but I think it's expected.
Mono IR does not include these specific helper/wrappers as Procs, but they are instead generated by the dev backend when it encounters CallType::HigherOrder here.
Since we don't pass a mutable Object reference to backend.build_proc, it can't add any extra Procs itself, but it instead returns them so we can add them in object_builder as it already did for LinkedFunction.
I agree it's a little subtle, but I didn't want to make too invasive of a change. Let me know if you think we should, though!
There was a problem hiding this comment.
I think the missing piece for me was that we use the data_pointer function to create function pointers in this case. Then things make more sense. I think this is fine then
|
Thank you, commits included in #5839 |
In the ARM dev backend, we use
LinkedDatafor function pointers, butobject_builderonly added undefined RC procs forLinkedFunctionrelocs, leading to a crash later when looking up its symbol.This fixes 52 tests 🎉