Re: Patch for ppc _SDA_BASE_ value

[Momchil Velikov <velco at fadata dot bg>]

Ian Lance Taylor wrote:
>   Date: Thu, 23 Mar 2000 01:22:28 +0200
>   From: Momchil Velikov <velco@fadata.bg>
>
>    >    +  /* Make sure _SDA_BASE_ and _SDA2_BASE_ are at offset <= 32768 from the
>    >    +     beginning of the respective output section.  */
>    > 
>    > That isn't the right check, though.  What you really want to ensure is
>    > that _SDA_BASE_ (_SDA2_BASE_) is at an offset <= 32768 from the start
>    > of where the .sdata (.sdata2) section is in the output file.
>    > 
>    > Consider
>    >     .data : { *(.data) *(.sdata) *(.sdata2) }
>    > where the input .data sections are larger than 65536 bytes in total.
>
>    I see. How about adding a call to a new backend function, say 
>    something like find_suitable_defaults_for_special_symbols(),
>    in elf_bfd_final_link(), somewhere before the loop which calls
>    elf_link_input_bfd(), which function in the PPC case would
>    traverse all the .sdata sections in all the input bfd's,
>    get the range [first small data item address, last small data item
>    address]
>    and put _SDA_BASE_ somewhere in between ?
>
> I think you could do this in size_dynamic_sections.

But neither output_section->vma nor output_offset are initialized
at the time size_dynamic_sections is called. How about
ldemul_after_allocation() ?

Regards,
-velco