Re: deleting relocs, objcopy and BFD

["Maciej W. Rozycki" <macro at imgtec dot com>]

On Thu, 27 Apr 2017, Jose E. Marchesi wrote:

> The recent commit
> 
> commit 1d15e434f43bc41a07bc7b0648fcb7e6ccbe8dcc
> Author: Nick Clifton <nickc@redhat.com>
> Date:   Thu Apr 13 14:50:56 2017 +0100
> 
>     Add note merging to strip and add code to merge stack size notes.
> 
>     * objcopy.c: Add --no-merge-notes option to disable note merging.
>       Add --[no-]merge-notes option to strip, and enable it by default.
>       (num_bytes): New function.
>       (merge_gnu_build_notes): Add code to merge stack size notes.
>     * binutils.texi: Update strip and objcopy documentation.
>     * readelf.c (print_gnu_build_attribute_name): Use defined
>       constants for note types.
> 
> Introduced a regression in two ELF targets: elf64-sparc and elf64-mips:
> 
> FAIL: merge notes section (64-bits)
[...]
> For example:
> 
> 1. Build binutils with --target=mips64-unknown-openbsd
> 
> 2. ./gas/as-new binutils/testsuite/binutils-all/note-2-64.s -o foo.o
> 
> 3. ./binutils/objcopy --merge-notes foo.o bar.o will segfault.
> 
> 4. Run the above command under GDB and check that:
> 
>    - At the beginning of `merge_gnu_build_notes' the
>     .gnu_build_attributes section has 3 external relocations.  This
>     makes `bfd_canonicalize_reloc' to return a count of 3 * 3 = 9
>     internal relocs.
> 
>   - The deletion logic in `merge_gnu_build_notes' then deletes two
>     internal relocs, leaving 7.  This sounds like a sequence of three
>     internal relocs gets broken.  I am not sure if this really a problem
>     (don't know much of mips64) but it is worth a check.

 Thanks for raising this issue.  I actually noticed the regression, but 
incorrectly attributed it to the usual test framework breakage (that I 
mean to address in a systematic way sometime) for MIPS targets that 
default to n64.  I'll see what I can do about it unless Alan or Nick beat 
me to it.

  Maciej