Re: Fix ld/418

[Alan Modra <amodra at bigpond dot net dot au>]

On Sun, Oct 10, 2004 at 05:24:38PM +0930, Alan Modra wrote:
> Those who peruse this patch may notice a FIXME.  I'll attack that later,
> probably by iterating over input sections using link_orders in
> bfd_elf_discard_info.  This code might fail if someone writes a linker
> script that deliberately reorders eh_frame sections, or uses HJ's new
> --sort-section linker options.

Hmm, link_orders aren't set up when bfd_elf_discard_info is called.  I
toyed with writing a linker function to traverse the script and call a
function on each input section mapped to a particular output section,
then decided it wasn't worth the bother.

	* elf-bfd.h (struct eh_frame_hdr_info): Add offsets_adjusted.
	* elf-eh-frame.c (_bfd_elf_write_section_eh_frame): Adjust
	offsets stored in struct eh_cie_fde entries before doing other
	work.

Index: bfd/elf-bfd.h
===================================================================
RCS file: /cvs/src/src/bfd/elf-bfd.h,v
retrieving revision 1.160
diff -u -p -r1.160 elf-bfd.h
--- bfd/elf-bfd.h	10 Oct 2004 07:58:07 -0000	1.160
+++ bfd/elf-bfd.h	10 Oct 2004 12:47:40 -0000
@@ -328,6 +328,7 @@ struct eh_frame_hdr_info
      We build it if we successfully read all .eh_frame input sections
      and recognize them.  */
   bfd_boolean table;
+  bfd_boolean offsets_adjusted;
 };
 
 /* ELF linker hash table.  */
Index: bfd/elf-eh-frame.c
===================================================================
RCS file: /cvs/src/src/bfd/elf-eh-frame.c,v
retrieving revision 1.33
diff -u -p -r1.33 elf-eh-frame.c
--- bfd/elf-eh-frame.c	10 Oct 2004 07:58:07 -0000	1.33
+++ bfd/elf-eh-frame.c	10 Oct 2004 12:47:40 -0000
@@ -798,6 +798,39 @@ _bfd_elf_write_section_eh_frame (bfd *ab
   sec_info = elf_section_data (sec)->sec_info;
   htab = elf_hash_table (info);
   hdr_info = &htab->eh_info;
+
+  /* First convert all offsets to output section offsets, so that a
+     CIE offset is valid if the CIE is used by a FDE from some other
+     section.  This can happen when duplicate CIEs are deleted in
+     _bfd_elf_discard_section_eh_frame.  We do all sections here because
+     this function might not be called on sections in the same order as
+     _bfd_elf_discard_section_eh_frame.  */
+  if (!hdr_info->offsets_adjusted)
+    {
+      bfd *ibfd;
+      asection *eh;
+      struct eh_frame_sec_info *eh_inf;
+
+      for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+	{
+	  if (bfd_get_flavour (ibfd) != bfd_target_elf_flavour
+	      || (ibfd->flags & DYNAMIC) != 0)
+	    continue;
+
+	  eh = bfd_get_section_by_name (ibfd, ".eh_frame");
+	  if (eh == NULL || eh->sec_info_type != ELF_INFO_TYPE_EH_FRAME)
+	    continue;
+
+	  eh_inf = elf_section_data (eh)->sec_info;
+	  for (ent = eh_inf->entry; ent < eh_inf->entry + eh_inf->count; ++ent)
+	    {
+	      ent->offset += eh->output_offset;
+	      ent->new_offset += eh->output_offset;
+	    }
+	}
+      hdr_info->offsets_adjusted = TRUE;
+    }
+
   if (hdr_info->table && hdr_info->array == NULL)
     hdr_info->array
       = bfd_malloc (hdr_info->fde_count * sizeof(*hdr_info->array));
@@ -810,16 +843,6 @@ _bfd_elf_write_section_eh_frame (bfd *ab
       if (ent->removed)
 	continue;
 
-      /* First convert all offsets to output section offsets, so that a
-	 CIE offset is valid if the CIE is used by a FDE from some other
-	 section.  This can happen when duplicate CIEs are deleted in
-	 _bfd_elf_discard_section_eh_frame.
-	 FIXME: This assumes that _bfd_elf_discard_section_eh_frame is
-	 called on sections in the same order as this function, which
-	 isn't necessarily so.  */
-      ent->offset += sec->output_offset;
-      ent->new_offset += sec->output_offset;
-
       if (ent->cie)
 	{
 	  /* CIE */

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre