QNX MIPS Support Patch

[Thiemo Seufer]

Jeff Baker wrote:
> The attached files comprise a patch to add QNX MIPS support to the 
> binutils.  The changes to 'bfd/elfxx-mips.c' as they appear here are 
> obviously not supposed to be complete, though they are functional. 
> Everything else looks (to me) to be reasonable.
> 
> Please look it over and let me know what I have to change to get it 
> accepted.  As for 'bfd/elfxx-mips.c' I'm still trying to figure out how 
> to piece it together proplerly as I have time.  I just want to get the 
> ball rolling on the rest.

[snip]
> --- bfd/config.bfd	29 Jul 2004 03:48:12 -0000	1.168
> +++ bfd/config.bfd	17 Aug 2004 17:16:21 -0000
> @@ -828,20 +828,35 @@ case "${targ}" in
>      targ_selvecs=ecoff_little_vec
>      ;;
>    mips*el-*-elf* | mips*el-*-rtems* | mips*el-*-vxworks* | mips*-*-chorus*)
>      targ_defvec=bfd_elf32_littlemips_vec
>      targ_selvecs="bfd_elf32_bigmips_vec bfd_elf64_bigmips_vec bfd_elf64_littlemips_vec"
>      ;;
>    mips*-*-elf* | mips*-*-rtems* | mips*-*-vxworks | mips*-*-windiss)
>      targ_defvec=bfd_elf32_bigmips_vec
>      targ_selvecs="bfd_elf32_littlemips_vec bfd_elf64_bigmips_vec bfd_elf64_littlemips_vec"
>      ;;
> +  mips*eb-*-nto*)
> +    targ_defvec=bfd_elf32_bigmips_vec
> +    targ_selvecs="bfd_elf32_littlemips_vec"
> +    targ_cflags=-D__QNXTARGET__
> +    ;;
> +  mips*el-*-nto*)
> +    targ_defvec=bfd_elf32_littlemips_vec
> +    targ_selvecs="bfd_elf32_bigmips_vec"
> +    targ_cflags=-D__QNXTARGET__
> +    ;;
> +  mips*-*-nto*)
> +    targ_defvec=bfd_elf32_littlemips_vec
> +    targ_selvecs="bfd_elf32_bigmips_vec"
> +    targ_cflags=-D__QNXTARGET__
> +    ;;

LE seems to be default, so you can remove the explicit mips*el-*-nto*
spec.

[snip]
> --- bfd/elf32-mips.c	29 Jun 2004 13:46:30 -0000	1.178
> +++ bfd/elf32-mips.c	17 Aug 2004 17:16:21 -0000
[snip]
> @@ -1315,21 +1337,23 @@ static const struct ecoff_debug_swap mip
>    ecoff_swap_ext_out,
>    _bfd_ecoff_swap_tir_out,
>    _bfd_ecoff_swap_rndx_out,
>    /* Function to read in symbolic data.  */
>    _bfd_mips_elf_read_ecoff_info
>  };
>  
>  #define ELF_ARCH			bfd_arch_mips
>  #define ELF_MACHINE_CODE		EM_MIPS
>  
> +#if !defined(__QNXNTO__) && !defined(__QNXTARGET__)
>  #define elf_backend_collect		TRUE
> +#endif

It's probably better to do a runtime switch instead of hardcoding.

[snip]
> --- bfd/elfxx-mips.c	13 Aug 2004 03:15:59 -0000	1.109
> +++ bfd/elfxx-mips.c	17 Aug 2004 17:16:22 -0000
[snip]
> @@ -1173,20 +1181,32 @@ _bfd_mips_elf_got16_reloc (bfd *abfd, ar
>  			   bfd *output_bfd, char **error_message)
>  {
>    if ((symbol->flags & (BSF_GLOBAL | BSF_WEAK)) != 0
>        || bfd_is_und_section (bfd_get_section (symbol))
>        || bfd_is_com_section (bfd_get_section (symbol)))
>      /* The relocation is against a global symbol.  */
>      return _bfd_mips_elf_generic_reloc (abfd, reloc_entry, symbol, data,
>  					input_section, output_bfd,
>  					error_message);
>  
> +#ifdef __QNXTARGET__
> +/* Else we might be doing a reloc from a local symbol, but we want it
> +   to be a got16 reloc, instead of a reloc against a section */
> +if (output_bfd != (bfd *) NULL
> +    && (symbol->flags & BSF_GLOBAL))
> +  {
> +    reloc_entry->address += input_section->output_offset;
> +  //  reloc_entry->addend = 0;
> +    return bfd_reloc_ok;
> +  }
> +#endif

Same here, and the other __QNXTARGET__ instances. Can you also
split this monster up a bit, starting with the new reloc type?
It would make review quite a bit easier.


Thiemo