[PATCH][GOLD] Handle symbols with section index SHN_XINDEX correctly.

[Ian Lance Taylor]

"Doug Kwan (關振德)" <dougkwan@google.com> writes:

> 2010-04-07  Doug Kwan  <dougkwan@google.com>
>
>         * arm.cc (Arm_relobj::scan_section_for_cortex_a8_erratum): Warn
>         if section is marked as containing instructions but has no mapping
>         symbols.
>         (Arm_relobj::do_count_local_symbols): Call adjust_sym_shndx to get
>         correct section index.
>         (Arm_relobj::find_linked_text_section): Ditto.


> @@ -6101,7 +6108,9 @@ Arm_relobj<big_endian>::do_count_local_s
>        const char* sym_name = pnames + sym.get_st_name();
>        if (Target_arm<big_endian>::is_mapping_symbol_name(sym_name))
>  	{
> -	  unsigned int input_shndx = sym.get_st_shndx();  
> +	  bool is_ordinary;
> +	  unsigned int input_shndx =
> +	    this->adjust_sym_shndx(i, sym.get_st_shndx(), &is_ordinary);
>  
>  	  // Strip of LSB in case this is a THUMB symbol.
>  	  Mapping_symbol_position msp(input_shndx, input_value & ~1U);
> @@ -6284,7 +6293,9 @@ Arm_relobj<big_endian>::find_linked_text
>        elfcpp::Sym<32, big_endian> sym(psyms + r_sym * sym_size);
>        if (sym.get_st_type() == elfcpp::STT_SECTION)
>  	{
> -	  *pshndx = this->adjust_shndx(sym.get_st_shndx());
> +	  bool is_ordinary;
> +	  *pshndx =
> +	    this->adjust_sym_shndx(r_sym, sym.get_st_shndx(), &is_ordinary);
>  	  return true;
>  	}
>        else

I'm not sure it's safe to ignore is_ordinary.  I think you should at
least add
    gold_assert(is_ordinary);
Otherwise you could silently do something really weird.  Presumably
is_ordinary is expected to be always true for these symbols.

This is OK with that change.

Ian