Handling of collectible types in tailcall dispatcher

[jkotas]

The tailcall dispatcher operates with raw function pointers that are not tracked by the GC. In theory, it is possible to get into a situation when there is nothing keeping the tailcall target alive and the target gets unloaded by accident.

[RamType0]

Why is this not marked as a bug?

[jakobbotsch]

Do I understand correctly that this can only be an issue for static functions (otherwise 'this' will keep it live)? What is the appropriate way to keep a static function live after taking its address?

[jkotas]

otherwise 'this' will keep it live

This is not strictly guaranteed. this can be null.

What is the appropriate way to keep a static function live after taking its address?

The stackwalker needs to convert the entrypoint back to MethodDesc* and report loader allocator associated with the MethodDesc*.

[jakobbotsch]

The stackwalker needs to convert the entrypoint back to MethodDesc* and report loader allocator associated with the MethodDesc*.

What function can be used to convert the function pointer to MethodDesc*?

I am wondering if we should report loader allocators for both IL_STUB_CallTailCallTarget and the target, since direct calls are sometimes done through calli in the stub when generics are involved. I suppose reporting both can't hurt?

[jkotas]

What function can be used to convert the function pointer to MethodDesc*?

Entry2MethodDesc

I suppose reporting both can't hurt?

Agree.