[PATCH] RISC-V: PR32499, Fix PR18841 segfault caused by ifunc relocation ordering

[Nelson Chu]

Thanks, committed after passing the riscv-gnu-toolchain gcc/binutils
regressions at least.

Nelson

On Fri, Jan 17, 2025 at 2:03 PM Alan Modra <amodra@gmail.com> wrote:

> On Fri, Jan 17, 2025 at 12:47:16PM +0800, Nelson Chu wrote:
> > Hi Guys,
> >
> > Would it be too rushed to do this at 2.44, or would it be more
> appropriate
> > to do it at 2.45?  Both are fine to me.
>
> Looks reasonable for 2.44 to me.
>
> > Thanks
> > Nelson
> >
> > On Tue, Jan 14, 2025 at 3:19 PM Nelson Chu <nelson@rivosinc.com> wrote:
> >
> > > Even though the relocation isn't IRELATIVE, it still should be come
> last if
> > > refering to ifunc symbol.  In order to get the ifunc relocs properly
> sorted
> > > the correct class needs to be returned.  The code mimics what has been
> done
> > > for x86, sparc, aarch64 and arm32.
> > >
> > > bfd/
> > >         PR 18841
> > >         PR 32499
> > >         * elfnn-riscv.c (riscv_reloc_type_class): Handle ifunc
> relocation
> > >         ordering, even though it's not IRELATIVE, it still should be
> come
> > >         last if refering ifunc symbol.
> > > ---
> > >  bfd/elfnn-riscv.c | 36 +++++++++++++++++++++++++++++++++++-
> > >  1 file changed, 35 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
> > > index 2122aa36c7c..c3881bb9edd 100644
> > > --- a/bfd/elfnn-riscv.c
> > > +++ b/bfd/elfnn-riscv.c
> > > @@ -3658,13 +3658,47 @@ riscv_elf_plt_sym_val (bfd_vma i, const
> asection
> > > *plt,
> > >    return plt->vma + PLT_HEADER_SIZE + i * PLT_ENTRY_SIZE;
> > >  }
> > >
> > > +/* Used to decide how to sort relocs in an optimal manner for the
> > > +   dynamic linker, before writing them out.  */
> > > +
> > >  static enum elf_reloc_type_class
> > > -riscv_reloc_type_class (const struct bfd_link_info *info
> ATTRIBUTE_UNUSED,
> > > +riscv_reloc_type_class (const struct bfd_link_info *info,
> > >                         const asection *rel_sec ATTRIBUTE_UNUSED,
> > >                         const Elf_Internal_Rela *rela)
> > >  {
> > > +  struct riscv_elf_link_hash_table *htab = riscv_elf_hash_table
> (info);
> > > +
> > > +  if (htab->elf.dynsym != NULL
> > > +      && htab->elf.dynsym->contents != NULL)
> > > +    {
> > > +      /* Check relocation against STT_GNU_IFUNC symbol if there are
> > > +        dynamic symbols.  */
> > > +      bfd *abfd = info->output_bfd;
> > > +      const struct elf_backend_data *bed = get_elf_backend_data
> (abfd);
> > > +      unsigned long r_symndx = ELFNN_R_SYM (rela->r_info);
> > > +      if (r_symndx != STN_UNDEF)
> > > +       {
> > > +         Elf_Internal_Sym sym;
> > > +         if (!bed->s->swap_symbol_in (abfd,
> > > +                                      (htab->elf.dynsym->contents
> > > +                                       + r_symndx *
> bed->s->sizeof_sym),
> > > +                                      0, &sym))
> > > +           {
> > > +             /* xgettext:c-format */
> > > +             _bfd_error_handler (_("%pB symbol number %lu references"
> > > +                                   " nonexistent SHT_SYMTAB_SHNDX
> > > section"),
> > > +                                 abfd, r_symndx);
> > > +             /* Ideally an error class should be returned here.  */
> > > +           }
> > > +         else if (ELF_ST_TYPE (sym.st_info) == STT_GNU_IFUNC)
> > > +           return reloc_class_ifunc;
> > > +       }
> > > +    }
> > > +
> > >    switch (ELFNN_R_TYPE (rela->r_info))
> > >      {
> > > +    case R_RISCV_IRELATIVE:
> > > +      return reloc_class_ifunc;
> > >      case R_RISCV_RELATIVE:
> > >        return reloc_class_relative;
> > >      case R_RISCV_JUMP_SLOT:
> > > --
> > > 2.39.3 (Apple Git-146)
> > >
> > >
>
> --
> Alan Modra
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://sourceware.org/pipermail/binutils/attachments/20250117/7cef020a/attachment.htm>