[PATCH v3] elf: Keep unused section symbols for Solaris binary

[Jan Beulich]

On 09.12.2025 08:18, H.J. Lu wrote:
> On Mon, Dec 8, 2025 at 3:52 PM Jan Beulich <jbeulich@suse.com> wrote:
>>
>> On 06.12.2025 04:04, H.J. Lu wrote:
>>> After
>>>
>>> commit d1bcae833b32f1408485ce69f844dcd7ded093a8
>>> Author: H.J. Lu <hjl.tools@gmail.com>
>>> Date:   Thu Jan 7 06:42:00 2021 -0800
>>>
>>>     ELF: Don't generate unused section symbols
>>>
>>> x86 ELF binutils tools no longer generate unused section symbols.  But
>>> Solaris binary generated by Solaris linker may include symbol sort
>>> sections which hold indexes of data and function symbols, sorted by name
>>> or by address.  When objcopy removes section symbols, it corrupts Solaris
>>> symbol sort sections since symbol indexes are changed.  Keep section
>>> symbols for Solaris binary.
>>>
>>> PR binutils/33684
>>> * objcopy.c (copy_object): Keep section symbols if input is a
>>> Solaris binary and output target is the same as input target.
>>
>> Hmm, an open-coded OSABI check in objcopy. I was expecting change in libbfd
>> to e.g. make bfd_keep_unused_section_symbols() on ibfd true for Solaris, to
>> then copy that property to obfd. Is there a reason to not try to somewhat
>> abstract what is wanted / needed?
> 
> Here is the v3 patch with the updated bfd_keep_unused_section_symbols.
> 
> After
> 
> commit d1bcae833b32f1408485ce69f844dcd7ded093a8
> Author: H.J. Lu <hjl.tools@gmail.com>
> Date:   Thu Jan 7 06:42:00 2021 -0800
> 
>     ELF: Don't generate unused section symbols
> 
> x86 ELF binutils tools no longer generate unused section symbols.  But
> Solaris binary generated by Solaris linker may include symbol sort
> sections which hold indexes of data and function symbols, sorted by name
> or by address.  When objcopy removes section symbols, it corrupts Solaris
> symbol sort sections since symbol indexes are changed.
> 
> 1. Update bfd_keep_unused_section_symbols to return true for Solaris
> binary.
> 2. Keep section symbols if input section symbols should be kept and
> output target is the same as input target.

This is okay afaic, but please give Rainer and/or Ali a chance to comment.

Irrespective I have a question here, though (perhaps also to Alan and Nick):
Isn't it wrong for objcopy to remove _any_ symbols (including section ones)
without an explicit request to do so? IOW shouldn't objcopy without any
specific options largely generate an output identical to the input?

Jan