Index PowerPC64 linker generated .eh_frame in .eh_frame_hdr

I noticed recently that .eh_frame FDEs generated by the linker for
call stubs and .glink weren't being indexed in .eh_frame_hdr, due to
bfd_elf_discard_info being run before the linker generated .eh_frame
sections were available for parsing.  This patch moves code around in
elf64-ppc.c and ppc64elf.em to avoid that problem.

Another problem fixed here is that --gc-sections parses .eh_frame
early, and the existing machinery allows only one go at parsing the
.eh_frame sections.  That resulted in the linker generated .eh_frame
CIEs not being merged and no .eh_frame_hdr index entries for those
FDEs.  It turns out that all the info from parsing .eh_frame is
attached to the section, so order of parsing isn't important, and
after parsing sec_info_type being set will prevent a section being
parsed again.  At least, when parsing doesn't hit an error.  So there
isn't really any need for "parsed_eh_frame".  "merge_cies" is also
redundant, which means _bfd_elf_{begin,end}_eh_frame_parsing can also
disappear.

bfd/
	* elf-bfd.h (struct eh_frame_hdr_info): Delete merge_cies and
	parsed_eh_frames.
	(_bfd_elf_begin_eh_frame_parsing): Delete.
	(_bfd_elf_end_eh_frame_parsing): Delete.
	* elf-eh-frame.c (_bfd_elf_begin_eh_frame_parsing): Delete.
	(_bfd_elf_end_eh_frame_parsing): Delete.
	(_bfd_elf_parse_eh_frame): Don't test parsed_eh_frame.  Test
	!info->relocatable in place of merge_cies.
	* elflink.c (bfd_elf_gc_sections, bfd_elf_discard_info): Adjust.
	* elf64-ppc.c (glink_eh_frame_cie): Pad to multiple of 8.
	(ppc64_elf_size_stubs): Likewise pad stub FDE.
	(ppc64_elf_build_stubs): Move code setting glink .eh_frame to..
	(ppc64_elf_size_stubs): ..here and..
	(ppc64_elf_finish_dynamic_sections): ..here.
ld/
	* emultempl/ppc64elf.em (gld${EMULATION_NAME}_after_allocation): Call
	bfd_elf_discard_info after generating glink .eh_frame.  Delete
	redundant test on ppc64_elf_setup_section_lists status.

1 files changed, 14 insertions, 14 deletions

diff --git a/ld/emultempl/ppc64elf.em b/ld/emultempl/ppc64elf.em
index 3e316af9751..914fc52183c 100644
--- a/ld/emultempl/ppc64elf.em
+++ b/ld/emultempl/ppc64elf.em
@@ -460,19 +460,6 @@ gld${EMULATION_NAME}_after_allocation (void)
 {
   int ret;
 
-  /* bfd_elf_discard_info just plays with data and debugging sections,
-     ie. doesn't affect code size, so we can delay resizing the
-     sections.  It's likely we'll resize everything in the process of
-     adding stubs.  */
-  ret = bfd_elf_discard_info (link_info.output_bfd, &link_info);
-  if (ret < 0)
-    {
-      einfo ("%X%P: .eh_frame/.stab edit: %E\n");
-      return;
-    }
-  else if (ret > 0)
-    need_laying_out = 1;
-
   /* If generating a relocatable output file, then we don't have any
      stubs.  */
   if (stub_file != NULL && !link_info.relocatable)
@@ -480,7 +467,7 @@ gld${EMULATION_NAME}_after_allocation (void)
       ret = ppc64_elf_setup_section_lists (&link_info);
       if (ret < 0)
 	einfo ("%X%P: can not size stub section: %E\n");
-      else if (ret > 0)
+      else
 	{
 	  ppc64_elf_start_multitoc_partition (&link_info);
 
@@ -510,6 +497,19 @@ gld${EMULATION_NAME}_after_allocation (void)
 	}
     }
 
+  /* We can't parse and merge .eh_frame until the glink .eh_frame has
+     been generated.  Otherwise the glink .eh_frame CIE won't be
+     merged with other CIEs, and worse, the glink .eh_frame FDEs won't
+     be listed in .eh_frame_hdr.  */
+  ret = bfd_elf_discard_info (link_info.output_bfd, &link_info);
+  if (ret < 0)
+    {
+      einfo ("%X%P: .eh_frame/.stab edit: %E\n");
+      return;
+    }
+  else if (ret > 0)
+    need_laying_out = 1;
+
   if (need_laying_out != -1)
     {
       gld${EMULATION_NAME}_map_segments (need_laying_out);