[MC] Put the pending fixups into relocation offset symbol's fragment

[StephenFan]

Summary

For .reloc directive, if the relocation offset is a symbol, then the relocation entry should be placed into relocation section and its sh_info should point to the section where relocation offset symbol at.

Test case:

.text
  ret
  nop
  nop
  .reloc foo, R_RISCV_32, 6

.data
.globl foo
foo:
  .word 0
  .word 0
  .word 0

The relocation entry should be placed into .reloc.data. And it is also what gnu as does.

[MaskRay]

llvm/lib/MC/MCObjectStreamer.cpp

131

Several cases are not covered by tests. They need tests.

llvm/test/MC/RISCV/reloc-directive.s

26

[StephenFan]

Add tests for FT_Dwarf and FT_PseudoProbe.

[StephenFan]

llvm/lib/MC/MCObjectStreamer.cpp

129

Hi @MaskRay , The test case of FT_Relaxable is not added. Because I don't know how to define a symbol which its fragment is FT_Relaxable. Do you have some ideas ?

137

I found that FT_CVDefRange is only used by COFF. Does COFF support .reloc directive?

[MaskRay]

LGTM.

llvm/test/MC/RISCV/reloc-directive.s

34

# CHECK:      Section ({{.*}}) .rela.pseudoprobe {
# CHECK-NEXT:   0x0 R_RISCV_32 - 0x6
# CHECK-NEXT: }