Re: Map ".text.hot" and ".text.unlikely" input section prefixes to separate output sections.

[Ian Lance Taylor <iant at google dot com>]

On Fri, Dec 21, 2012 at 7:38 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Thu, Dec 20, 2012 at 10:26 PM, Ian Lance Taylor <iant@google.com> wrote:
>> On Thu, Dec 20, 2012 at 2:38 PM, Cary Coutant <ccoutant@google.com> wrote:
>>> This patch fixes the problems I've been seeing. It sets the "may sort
>>> attached input sections" flag for the .text section, which ensures
>>> that we track all input sections for that output section. The problem
>>> was that we can't start tracking input sections in mid-stream. We may
>>> not want to do this unconditionally for .text, so maybe you'll want a
>>> linker option to enable this reordering?
>>
>> Thanks for tracking this down.
>>
>> It would be nice to avoid doing it for .text, but I don't see how to
>> do that while remaining compatible with GNU ld.
>>
>> I reimplemented this code in a different way that I think fits in
>> better with the existing section ordering code.  Committed to
>> mainline.
>>
>> Ian
>>
>>
>> 2012-12-20  Ian Lance Taylor  <iant@google.com>
>>
>>         * layout.cc (Layout::special_ordering_of_input_section): New
>>         function.
>>         (Layout::layout): If input section requires special ordering, must
>>         sort input sections.
>>         (Layout::make_output_section): May sort .text input sections.
>>         (Layout::is_section_name_prefix_grouped): Remove.
>>         * layout.h (class Layout): Declare
>>         special_ordering_of_input_section.  Don't declare
>>         is_section_name_prefix_grouped.
>>         * output.cc (Output_section::add_input_section): Revert last
>>         change.
>>         (Output_section::Input_section_sort::match_file_name): Don't crash
>>         if called on output section data.
>>         (Output_section::Input_section_sort_compare): Sort based on
>>         special ordering.
>>         (Output_section::Input_section_sort_section_order_index_compare):
>>         Revert last patch.
>>         (Output_section::sort_attached_input_sections): Likewise.
>
> It doesn't fix regressions with x32 gold:
>
> `echo g++ -mx32  -W -Wall    -Werror -D_LARGEFILE_SOURCE
> -D_FILE_OFFSET_BITS=64 -fmerge-constants   -static-libstdc++
> -static-libgcc  -o incremental_test_2 | sed -e
> 's/-Wp,-D_FORTIFY_SOURCE=[0-9][0-9]*//'` -Wl,--incremental-update
> -Bgcctestdir/ two_file_test_tmp_2.o two_file_test_1b_ndebug.o
> two_file_test_2_ndebug.o two_file_test_main_ndebug.o
> gcctestdir/ld: internal error in convert_types, at
> /export/gnu/import/git/binutils/gold/gold.h:192
> collect2: error: ld returned 1 exit status
> make[3]: *** [incremental_test_2] Error 1

I don't really see how that could be related to this patch series.  I
don't have an x32 system so I also can't debug this.  Can you show a
backtrace at the point of failure?  You can set a breakpoint on
gold::do_gold_unreachable.

Ian