Re: PATCH: Linker is broken

[Jakub Jelinek <jakub at redhat dot com>]

On Thu, May 13, 2004 at 07:48:06PM -0700, H. J. Lu wrote:
> The problem is .eh_frame is readonly. The new linker script moves it
> at the beginning of data segment. When it is empty, the new alignment
> causes map_sections_to_segments to allocate a segment for an empty
> .eh_frame section. This patch fixes the testcase. Should
> map_sections_to_segments be able to deal with empty readonly
> section at the alignment boundary at the beginning of data segment?

This patch will break linking if .eh_frame is read-write (e.g. with
older GCCs, e.g. all GCC 2.x .eh_frames need relocations and are read-write,
with newer GCCs they are read-write unless configure found it is desirable
to make them read-only).  Also, .gcc_except_table is another section
which is sometimes read-only and sometimes read-write.

The reason why I have never seen such problems is that because I have
in my rpms also the
http://sources.redhat.com/ml/binutils/2002-11/msg00592.html
patch (well, adjusted for current sources), so .eh_frame and
.gcc_except_table are put into the end of the read-only segment if
they are read-only and into the read-write segment if they are read-write.
...
  .rodata         : { *(.rodata .rodata.* .gnu.linkonce.r.*) }
  .rodata1        : { *(.rodata1) }
  .eh_frame_hdr : { *(.eh_frame_hdr) }
  .eh_frame       : ONLY_IF_RO { KEEP (*(.eh_frame)) }
  .gcc_except_table   : ONLY_IF_RO { *(.gcc_except_table) }
  /* Adjust the address for the data segment.  We want to adjust up to
     the same address within the page on the next page up.  */
  . = ALIGN (0x100000) - ((0x100000 - .) & (0x100000 - 1)); . = DATA_SEGMENT_ALIGN (0x100000, 0x1000);
  /* Exception handling  */
  .eh_frame       : ONLY_IF_RW { KEEP (*(.eh_frame)) }
  .gcc_except_table   : ONLY_IF_RW { *(.gcc_except_table) }
  /* Thread Local Storage sections  */
  .tdata          : { *(.tdata .tdata.* .gnu.linkonce.td.*) }
  .tbss           : { *(.tbss .tbss.* .gnu.linkonce.tb.*) *(.tcommon) }
...

Now, there are several options what to do with CVS:
1) put that patch into CVS (if somebody can come up with a nicer
   syntax, let me know)
2) implement what Alan proposed, i.e. AUTOPLACE keyword and hardcode
   that .eh_frame should come before .gcc_except_table (if any),
   at the very end of RX segment if read-only and at the very beginning
   of RW segment if read-write (and similarly for .gcc_except_table)
3) do a hack and mark .eh_frame/.gcc_except_table output sections
   read-write
4) somehow deal with read-only sections at the beginning of read-write
   segment in section to segment mapping code (not sure how though)

This problem did not pop up in the make check testing I have done
on CVS binutils with the relro patch applied, I'm sorry for missing
this problem.

	Jakub