riscv: Enable linker relaxation with lld 15 and later

[rick_sloservers.com]

Summary

ld.lld gained RISC-V linker relaxation in LLVM 15, but bsd.linker.mk
only reports riscv-relaxations for GNU ld, so we still build with
-mno-relax.

Report it for lld 15 and later, and pass -mrelax explicitly since our
clang defaults to -mno-relax (389f88cffd0d).

Test Plan

Built world and a QEMU kernel for riscv64 with lld 19 and compared them with an unrelaxed build of the same tree, verified relaxation took effect: every call in libc.so.7 is now a jal, and so are about half of the calls in the kernel.

The kernel boots to multi-user in qemu-system-riscv64, with and without loader.efi. Modules work too, including a VNET jail with epair and pf, which is what relaxation broke on arm64 (PR 264094).

kyua results for lib/libc, lib/msun, bin/sh and sys/net match the unrelaxed build.

Tested only in QEMU, I lack RISCV hardware for testing on real hw.

[jrtc27]

Did you use LLMs to generate this?

[rick_sloservers.com]

In D60278#1383212, @jrtc27 wrote:

Did you use LLMs to generate this?

Yes

[jrtc27]

If you’re to use them, at least have the decency to remove all the slop and write a proper commit message. As it stands this is a very low-effort change that feels extractive.

[rick_sloservers.com]

Yes, no problem. Sorry, it's not lack of effort on my part, just ignorance. I'm new to this system and didn't know that the summary becomes the commit message, I can understand now why that would be irritating. Give a few moments, please, and I will rewrite.

[mhorne]

stand/defs.mk

186–188

What's going on here that you didn't add the -mrelax flag? Are we opting to never enable relaxations for the bootloader?

[rick_sloservers.com]

stand/defs.mk

186–188

No, the loader gets relaxed too. stand/efi/loader and libsa include bsd.prog.mk and bsd.lib.mk, which now add -mrelax when the linker supports relaxations, so this block only matters for a linker that can't relax. With lld 15+ it doesn't apply. I only changed its comment, since it said lld can't relax.

I can add a .else with -mrelax here too if you'd rather it be explicit.

[mhorne]

stand/defs.mk

186–188

Yes, please add the .else case, for consistency.

Given the inclusion of bsd.prog.mk or bsd.lib.mk, I wonder about the necessity of this block at all (it seems I am the original culprit...). I don't think it's necessary to chase it down.

[rick_sloservers.com]

Looking at this again, I think your instinct is right and the block in defs.mk can just go. It came in with the riscv EFI loader (2192efc03bc4), but 79f5991559ce had already moved -mno-relax into bsd.prog.mk and bsd.lib.mk a few months before, so the loader gets the flag from there. I removed it and checked every riscv dir under stand, they all still get -mrelax. I think an .else would also have to go under a riscv check, or else the amd64 and arm64 loaders would get -mrelax. Removing the whole block seems cleaner. What do you think?