Re: [PATCH 6/9] objtool: Add elf_create_undef_symbol()

[Sami Tolvanen]

On Wed, Mar 17, 2021 at 7:40 AM Miroslav Benes <mbenes@xxxxxxx> wrote:
>
> On Wed, 17 Mar 2021, Peter Zijlstra wrote:
>
> > On Wed, Mar 17, 2021 at 02:52:23PM +0100, Miroslav Benes wrote:
> >
> > > > + if (!elf_symbol_add(elf, sym, SHN_XINDEX)) {
> > > > +         WARN("elf_symbol_add");
> > > > +         return NULL;
> > > > + }
> > >
> > > SHN_XINDEX means that the extended section index is used. Above you seem
> > > to use it in the opposite sense too (assigning to shndx when shndx_data is
> > > NULL). While it makes the code easier to handle, it is a bit confusing
> > > (and maybe I am just confused now). Could you add a comment about that,
> > > please? elf_symbol_add() seems like a good place.
> >
> > Yes, that was a horrible thing to do :/ And you understood it right.
> >
> > Looking at it again, I'm not sure it is actually correct tho; shouldn't
> > elf_create_undef_symbol() also look at gelf_getsymshndx() of symtab ?
>
> Probably yes.
>
> > What toolchain generates these extended sections and how? That is, how
> > do I test this crud..
>
> Sami might know.

Clang generates these with LTO for vmlinux.o, but I'm guessing gcc
will do the same with -ffunction-sections -fdata-sections.

Sami