Re: RFC: syntax for a section ordering file

[Fangrui Song <i@maskray.me>]

On 2024-04-24, Nick Clifton wrote:
>Hi H.J.  Hi Fangrui,
>
>  Is there a defined syntax for the contents of a section ordering file ?
>  IE one passed to the linker via the --section-ordering= option ?
>
>  I am attempting to take H.J.'s bfd linker text section ordering file
>  patch and extend it to cover other sections.  In doing so, I need to
>  be sure that I have the syntax right.

Hi Nick,

* Apple ld -order_file. lld's MachO port ld64.lld has ported the option.
   The feature is like ld.lld --section-ordering-file='s superset with
   filename support. The syntax also supports "x86_64:" prefix, but this
   design seems quite unusual in linker features.

   This option is used by iOS mobile applications.

   example:
   https://github.com/llvm/llvm-project/blob/main/lld/test/MachO/order-file.s

* gold --section-ordering-file=: which might be most similar to this patch.
   I believe this option is effectively unused in the wild.
   People find the section-based naming approach too inconvenient.
   This is incompatible with sections that are not suffixed and clang
   -fno-unique-section-names.
* ld.lld --symbol-ordering-file=:

   This option is used by Android and regular Linux folks focusing on server performance.

   example:
   https://github.com/llvm/llvm-project/blob/main/lld/test/ELF/symbol-ordering-file.s

   I have some notes at
   https://maskray.me/blog/2020-11-15-explain-gnu-linker-options#symbol-ordering-filefile

The default linker script contains an input section description.

   *(.text .stub .text.* .gnu.linkonce.t.*)

I'm still curious whether it is feasible to implement --symbol-ordering-file that decides
how to sort the .text.* pattern.

>  To my mind if the section ordering file contains the following:
>
>     # A comment
>    .text.hot .text.cold,.text.warm
>    .data.big
>    .data.small
>    .text.foo*

If we are going to implement --section-ordering-file=, perhaps be rigid
and disallow space or comma separated patterns?

I believe the aforementioned implementations do not split the line.

>  Then this should be roughly equivalent to:
>
>  SECTIONS
>  {
>    .text : {
>      *(.text.hot)
>      *(.text.cold)
>      *(.text.warm)
>      *(.text.foo*)
>      *(.text)
>      }
>    .data : {
>      *(.data.big)
>      *(.data.small)
>      *(.data)
>      }
>  }
>
>  So all of the .text.<something> entries in the section ordering
>  file are placed at the start of the output .text section (even
>  if some of them occur after entries for other output sections)
>  and all of the .data.<something> entries are placed at the start
>  of the .data section.
>
>  This will require co-operation from the linker script to have
>  the "INCLUDE config.section-ordering-file" statements at the
>  correct places, but I think that it could work.

Hmm.  I am curious why the first INCLUDE (in .text) does not append
.data.big/.data.small (as requested).

>  But obviously I want the option to be compatible with GOLD and
>  LLD and I could not find any clear definitions or examples of
>  the syntax used.
>
>Cheers
>  Nick
>