[lld] [llvm] Reapply "[RISCV] Support RISCV Atomics ABI attributes (#84597)" (PR #90266)

[Alex Bradbury via llvm-commits]

asb wrote:

> I don't think this is very practical. CL opts are probably not enough, and tend to cause issues, particularly when they don't get forwarded to the linker correctly in LTO builds.

I think one reason we'd go with this route is it would make it easy for people to have a downstream patch that flips the default, while having the logic upstream and at least minimally tested. Not ideal, but it's a pattern we've followed before.

Anyway, let's discuss a path forwards in the call today. I agree we could go overboard on maintaining strict compatibility at all times, but the segfault with what was until recently the most current ld release is fairly user hostile. There aren't any great solutions here I think, we just have to find the least-worst :/

https://github.com/llvm/llvm-project/pull/90266