Re: [PATCH] x86-64: Allocate input section memory if needed

["H.J. Lu" <hjl.tools@gmail.com>]

On Wed, Dec 28, 2022 at 2:07 PM Alan Modra <amodra@gmail.com> wrote:
>
> On Wed, Dec 28, 2022 at 09:23:21AM -0800, H.J. Lu wrote:
> > On Tue, Dec 27, 2022 at 5:16 PM Alan Modra <amodra@gmail.com> wrote:
> > >
> > > On Tue, Dec 27, 2022 at 11:47:56AM -0800, H.J. Lu via Binutils wrote:
> > > > +                       /* Cache the section contents for
> > > > +                          elf_link_input_bfd.  */
> > > > +                       elf_section_data (sec)->this_hdr.contents
> > > > +                         = contents;
> > >
> > > You shouldn't really be caching unaltered section contents when
> > > !info->keep_memory.  I'm not saying the patch is wrong, but please fix
> > > this when you have some time.
> > >
> > > --
> > > Alan Modra
> > > Australia Development Lab, IBM
> >
> > elf_x86_64_scan_relocs has
> >
> >   if (elf_section_data (sec)->this_hdr.contents != contents)
> >     {
> >       if (!converted && !_bfd_link_keep_memory (info))
> >         free (contents);
> >       else
> >         {
> >           /* Cache the section contents for elf_link_input_bfd if any
> >              load is converted or --no-keep-memory isn't used.  */
> >           elf_section_data (sec)->this_hdr.contents = contents;
> >           info->cache_size += sec->size;
> >         }
> >     }
> >
> > Are you suggesting that it should be
>
> No, the code above is correct and how it should be in
> elf_x86_size_or_finish_relative_reloc.  Of course it is a little more

But elf_x86_size_or_finish_relative_reloc modifies the section
contents by storing r_addend in it.

> complicated there due to handling multiple sections.
>
> >
> >   if (elf_section_data (sec)->this_hdr.contents != contents)
> >     {
> >       if (!converted)
> >         free (contents);
> >       else
> >         {
> >           /* Cache the section contents for elf_link_input_bfd if any
> >              load is converted.  */
> >           elf_section_data (sec)->this_hdr.contents = contents;
> >           info->cache_size += sec->size;
> >         }
> >     }
>
> This would be wrong as it ignores info->keep_memory.
>
> --
> Alan Modra
> Australia Development Lab, IBM



-- 
H.J.