[PATCH] gas: add --reloc-use-section-sym={all, temporary, none} option for ELF

[Fangrui Song]

On Mon, Feb 9, 2026 at 12:42 AM Jan Beulich <jbeulich@suse.com> wrote:
>
> On 08.02.2026 22:03, maskray@sourceware.org wrote:
> > From: Fangrui Song <maskray@sourceware.org>
> >
> > When generating relocations for non-ifunc local symbols that satisfies
> > several conditions, GAS converts them to reference the section symbol
> > (STT_SECTION) instead, folding the original symbol's offset into the
> > addend.  This allows the original local symbol to be omitted from
> > .symtab, but the STT_SECTION symbol itself must be present, so the
> > conversion saves .symtab entries only when a section has more than one
> > local symbol referenced by relocations.
> >
> > Add --reloc-use-section-sym to control this conversion:
> >
> > - all (default): convert all eligible local symbols
> > - temporary: only convert compiler-generated locals (.L prefix)
> > - none: never convert; keep all symbols as-is in relocations
> >
> > This is useful for debugging and for tools that benefit from preserve
> > symbol names.
> >
> > PR gas/33885
> > ---
> >  gas/NEWS                                      |  4 ++++
> >  gas/as.c                                      | 20 +++++++++++++++-
> >  gas/as.h                                      | 10 ++++++++
> >  gas/doc/as.texi                               | 11 +++++++++
> >  .../gas/i386/reloc-use-section-sym-all.d      | 23 +++++++++++++++++++
> >  .../i386/reloc-use-section-sym-local-label.d  | 23 +++++++++++++++++++
> >  .../gas/i386/reloc-use-section-sym-none.d     | 23 +++++++++++++++++++
> >  .../gas/i386/reloc-use-section-sym.s          | 14 +++++++++++
> >  gas/testsuite/gas/i386/x86-64.exp             |  4 ++++
> >  gas/write.c                                   | 12 +++++++++-
> >  10 files changed, 142 insertions(+), 2 deletions(-)
> >  create mode 100644 gas/testsuite/gas/i386/reloc-use-section-sym-all.d
> >  create mode 100644 gas/testsuite/gas/i386/reloc-use-section-sym-local-label.d
> >  create mode 100644 gas/testsuite/gas/i386/reloc-use-section-sym-none.d
> >  create mode 100644 gas/testsuite/gas/i386/reloc-use-section-sym.s
> >
> > diff --git a/gas/NEWS b/gas/NEWS
> > index e384d1135c0..1d07f909b8f 100644
> > --- a/gas/NEWS
> > +++ b/gas/NEWS
> > @@ -1,5 +1,9 @@
> >  -*- text -*-
> >
> > +* New command line option --reloc-use-section-sym=[all|local-label|none]
> > +  controls whether relocations referencing local binding symbols are adjusted
> > +  to use section symbols.
>
> A number of targets set tc_fix_adjustable() to constant-false. Such targets
> won't be affected at all, which I think wants expressing here by adding some
> for of conditional. Along these lines I'm also not overly happy to see the
> new testcases to all be x86_64-only.

Understood. I wasn't aware that tc_fix_adjustable is defined to 0 for
many targets.

Does this warrant a test in gas/testsuite/gas/elf/ ? But it's unclear
how to write such a generic test that works on all targets.
I want to avoid enumerating the target list (which I don't even know)
and run `configure --target=$target && make all-gas && check-gas
RUNTESTFLAGS=elf.exp`.

I can add an aarch64 test.

"all eligible local binding symbols are converted"

> I also have to admit that the command line option is pretty long. My minimal
> suggestion would be to drop the "use" part.


> > @@ -956,6 +957,16 @@ Ignored.  Supported for compatibility with tools that pass the same option to
> >  both the assembler and the linker.
> >
> >  @ifset ELF
> > +@item --reloc-use-section-sym=all
> > +@itemx --reloc-use-section-sym=local-label
> > +@itemx --reloc-use-section-sym=none
> > +Control whether relocations referencing local binding symbols are adjusted to
> > +use section symbols instead.  With @code{all} (the default), relocations
> > +against all eligible local binding symbols are converted.  With
> > +@code{local-label}, only relocations against local labels (symbols matching the
> > +@code{.L} prefix) are converted.  With @code{none}, no conversions are
> > +performed and all relocations retain their original symbols.
>
> Up to "and" I agree, but what follows is too broad imo. Yes, the context here
> is conversion to section symbols, but still - see in particular the use of
> symbol_equated_reloc_p() in adjust_reloc_syms().
>
> > --- a/gas/write.c
> > +++ b/gas/write.c
> > @@ -804,6 +804,10 @@ adjust_reloc_syms (bfd *abfd ATTRIBUTE_UNUSED,
> >
> >    dump_section_relocs (abfd, sec, stderr);
> >
> > +  /* Disable all conversion for --reloc-use-section-sym=none.  */
> > +  if (flag_reloc_use_section_sym == reloc_use_section_sym_none)
> > +    return;
> > +
> >    for (fixp = seginfo->fix_root; fixp; fixp = fixp->fx_next)
> >      if (fixp->fx_done)
> >        /* Ignore it.  */
> > @@ -909,6 +913,12 @@ adjust_reloc_syms (bfd *abfd ATTRIBUTE_UNUSED,
> >       if ((symsec->flags & SEC_THREAD_LOCAL) != 0)
> >         continue;
> >
> > +     /* If --reloc-use-section-sym=local-label, don't adjust a reloc against
> > +        a local label.  */
> > +        if (flag_reloc_use_section_sym == reloc_use_section_sym_local_label
> > +         && !bfd_is_local_label (stdoutput, symbol_get_bfdsym (sym)))
> > +       continue;
>
> Comment and code look to not match up. I think the comment is inverted and
> means to say "..., adjust only if the reloc is against a local label" or
> "don't adjust a reloc if it isn't against a local label" or some such.
>
> Jan