Re: [GOLD] PowerPC64 TOC indirect to TOC relative code editing

[Cary Coutant <ccoutant at gmail dot com>]

> Let me think about this. I'm thinking of something a bit more
> ambitious (but just a bit).

Here's what I'm thinking:

- Move the loop from do_relocate_sections() to a new protected, but
non-virtual method, relocate_section_range(). This new method will
take a start_shndx and end_shndx.

- The default implementation of do_relocate_sections() will simply
call relocate_section_range() with start_shndx = 1 and end_shndx =
this->shnum() - 1.

- The overrides in the Arm targets will now call the new method
instead of the overridden base implementation.

- Your override for PPC can call relocate_section_range() three times,
for these section index ranges: [toc_shndx, toc_shndx], [1, toc_shndx
- 1], [toc_shndx + 1, shnum - 1].

I can imagine targets having even more complex needs for controlling
the order of section relocation. This gives the target more control
than simply naming a start index.

A proposed patch is attached for your thoughts.

I trust that it's never the case that a candidate for this TOC
optimization might refer to a TOC entry from a different relocatable
object.

-cary

Attachment: reloc-sect-range.patch
Description: Binary data