[PATCH] D18056: Compute value of local symbol with getVA.

[Rafael Espíndola via llvm-commits]

On 10 March 2016 at 15:19, Rui Ueyama <ruiu at google.com> wrote:
> ruiu added a comment.
>
> This is very similar to a change that I have locally.
>
>
> ================
> Comment at: ELF/Symbols.cpp:79-82
> @@ +78,6 @@
> +
> +    // PPC64 has a special relocation representing the TOC base pointer
> +    // that does not have a corresponding symbol.
> +    if (Config->EMachine == EM_PPC64 && !SC)
> +      return getPPC64TocBase();
> +
> ----------------
> This covers a special case, and I don't think absence of the input section always means that we want TocBase. Can you move this InputSectionBase::relocate like this?
>
>     // PPC64 has a special relocation representing the TOC base pointer
>     // that does not have a corresponding symbol.
>     if (Config->EMachine == EM_PPC64 && RI.getType(false) == R_PPC64_TOC) {
>       uintX_t SymVA = getPPC64TocBase() + A;
>       Target->relocateOne(BufLoc, BufEnd, Type, AddrLoc, SymVA, 0);
>       continue;
>     }

Will do.

> ================
> Comment at: ELF/Symbols.cpp:95-98
> @@ +94,6 @@
> +             Out<ELFT>::TlsPhdr->p_vaddr;
> +    if (Sym.getType() == STT_SECTION) {
> +      Offset += Addend;
> +      Addend = 0;
> +    }
> +    return SC->OutSec->getVA() + SC->getOffset(Offset);
> ----------------
> In this case Addend is taken into consideration.
>
> ================
> Comment at: ELF/Symbols.cpp:108
> @@ +107,3 @@
> +SymbolBody::getVA(typename ELFFile<ELFT>::uintX_t Addend) const {
> +  return getSymVA<ELFT>(*this, Addend) + Addend;
> +}
> ----------------
> ... and adding an addend again? Is this correct?
>

It is, since it is set to 0. STT_SECTIONS symbols are an annoying
special case since you need to use the addend before figuring out
where that position was mapped to the output.

Cheers,
Rafael