Functions or variables with this attribute will not have local_unnamed_addr/unnamed_addr in IR. So, they will show up in address-significant table.
I'm not opposed to the attribute per se, but I'd like to understand the motivation behind it a bit more. It seems like it solves a special edge case for a specific linker and I'm wondering if that's sufficient motivation for the community to support the attribute or not.
You should also have a test that the attribute does not accept any arguments.