[RISCV] Ignore addend in R_RISCV_GOT_HI20

[Jonathon Penix (jonathonpenix)]

This was recently changed in the psABI to require the addend be non-zero [1].

Handle this similarly to other relocations that must have zero addends by just
ignoring the addend (setting it to zero) and producing a warning.

Note that the calculation in applyGOT is left unchanged (the + A is still
there). applyGOT is shared with some other GOT-relative relocations that have
no explicit zero addend requirement and eld/lld/bfd all seem to a) accept those
addends b) honor those addends and c) not produce a diagnostic.

So for now, I think it is a bit clearer to keep R_RISCV_GOT_HI20 using
applyGOT and just make sure we zero the addend ahead of time.

Since we're 0-ing the addend ahead of time, switch the GOT relaxations
to asserting on non-0 addends rather than returning. It's a bit unrelated,
but the assert was also extended to cover both the got_hi and pcrel_lo
(as both must have 0 addends).

[1] riscv-non-isa/riscv-elf-psabi-doc@eb794e8

Fixes: #1413

[Sam Elliott (lenary)]

I did a bit of background research and it's likely all the other applyGOT calls should also ignore their addends, but we didn't cover those in the psABI update yet, so I think it's reasonable to limit this to those covered by the psABI update (just GOT_HI20)

[Shankar Easwaran (quic-seaswara)]

I did a bit of background research and it's likely all the other applyGOT calls should also ignore their addends, but we didn't cover those in the psABI update yet, so I think it's reasonable to limit this to those covered by the psABI update (just GOT_HI20)

Thanks Sam Elliott (@lenary). Can we raise an issue to update the ABI ?

[Sam Elliott (lenary)]

I did a bit of background research and it's likely all the other applyGOT calls should also ignore their addends, but we didn't cover those in the psABI update yet, so I think it's reasonable to limit this to those covered by the psABI update (just GOT_HI20)

Thanks Sam Elliott (Sam Elliott (@lenary)). Can we raise an issue to update the ABI ?

Yes, I intend to. I just need to finish reading about how tls gd works in LLD to make sure my understanding is right.

[Sam Elliott (lenary)]

riscv-non-isa/riscv-elf-psabi-doc#510

[Jonathon Penix (jonathonpenix)]

Thanks Sam for filing the issue in the psABI!

As I don't think there's been movement in the psABI, does anyone have concerns about keeping this patch limited to just R_RISCV_GOT_HI20?

I'll update this PR as the GOT relaxations have since been merged (which, there's some interaction here I think), but waiting on #1428 to save some rebasing.

But if we'd rather wait for further changes in the psABI, etc. I can put this on hold.

[Jonathon Penix (jonathonpenix)]

Rebased to unstack after merging #1428 + added some fixups of some of the GOT relaxation stuff.