Re: [PATCH] ld: Allow section groups to be resolved as part of a relocatable link

[Andrew Burgess <andrew dot burgess at embecosm dot com>]

* Alan Modra <amodra@gmail.com> [2017-05-10 12:15:31 +0930]:

> On Tue, May 09, 2017 at 06:29:42PM +0100, Andrew Burgess wrote:
> > Any followup to the below?
> 
> Sorry, I put this aside as being a little difficult, and it
> disappeared among too many other emails.

Thanks for the follow up.  It's taken me a while to find time for this
again, but hopefully this should be it now...

> 
> > * Andrew Burgess <andrew.burgess@embecosm.com> [2017-04-13 23:07:13 +0100]:
> > 
> > > * Alan Modra <amodra@gmail.com> [2017-04-04 19:31:21 +0930]:
> > > 
> > > > On Fri, Mar 24, 2017 at 03:11:29PM +0000, Andrew Burgess wrote:
> > > > > This commit adds a new linker feature: the ability to resolve section
> > > > > groups as part of a relocatable link.
> > > > 
> > > > What happens with dwarf debug or .eh_frame info for discarded
> > > > sections?
> > > 
> > > Alan,
> > > 
> > > Thanks for the feedback, sorry for the slow reply.
> > > 
> > > I wasn't totally sure what you're asking for, as far as I'm aware
> > > debug and eh_frame sections will work largely as before.  However,
> > > given your question I suspect you've spotted something that I'm still
> > > not getting.
> 
> One of the "little difficulties" is admitting that when I wrote the
> email, I was thinking that ld -r kept all the groups. ;-)  It doesn't,
> so the change isn't as major as I was thinking.
> 
> > +@item --inhibit-group-allocation
> > +This option prevents the linker from resolving sections groups,
> > +sections that are part of a group remain group members in the output
> > +file and the section groups are retained.  This is the default
> > +behaviour for a relocatable link (@samp{-r}) but this option can be
> > +used to change the behaviour of a final link.  The script command
> > +@code{INHIBIT_GROUP_ALLOCATION} has the same
> > +effect. @xref{Miscellaneous Commands}.
> 
> This part is a problem.  The ELF gABI says "Sections of type SHT_GROUP
> may appear only in relocatable objects".  Did you add the option for
> symmetry?  If so, let's drop this part.

Yes, as I took the treatment of common sections as my model I just
added a FORCE/INHIBIT mechanism, but it's really only the FORCE that I
care about.

I've updated the patch to remove all trace (I believe) of the INHIBIT
side of things.  The only _new_ code is a couple of asserts inside bfd
that if we're _not_ resolving, then we _must_ be doing a relocatable
link.

Would you be happy with this version?

Thanks,
Andrew

---

ld: Allow section groups to be resolved as part of a relocatable link

This commit adds a new linker feature: the ability to resolve section
groups as part of a relocatable link.

Currently section groups are automatically resolved when performing a
final link, and are carried through when performing a relocatable link.
By carried through this means that one copy of each section group (from
all the copies that might be found in all the input files) is placed
into the output file.  Sections that are part of a section group will
not match input section specifiers within a linker script and are
forcibly kept as separate sections.

There is a slight resemblance between section groups and common
section.  Like section groups, common sections are carried through when
performing a relocatable link, and resolved (allocated actual space)
only at final link time.

However, with common sections there is an ability to force the linker to
allocate space for the common sections when performing a relocatable
link, there's currently no such ability for section groups.

This commit adds such a mechanism.  This new facility can be accessed in
two ways, first there's a command line switch --force-group-allocation,
second, there's a new linker script command FORCE_GROUP_ALLOCATION.  If
one of these is used when performing a relocatable link then the linker
will resolve the section groups as though it were performing a final
link, the section group will be deleted, and the members of the group
will be placed like normal input sections.  If there are multiple copies
of the group (from multiple input files) then only one copy of the group
members will be placed, the duplicate copies will be discarded.

Unlike common sections that have the --no-define-common command line
flag, and INHIBIT_COMMON_ALLOCATION linker script command there is no
way to prevent group resolution during a final link, this is because the
ELF gABI specifically prohibits the presence of SHT_GROUP sections in a
fully linked executable.  However, the code as written should make
adding such a feature trivial, setting the new resolve_section_groups
flag to false during a final link should work as you'd expect.

bfd/ChangeLog:

	* elf.c (_bfd_elf_make_section_from_shdr): Don't initially mark
	SEC_GROUP sections as SEC_EXCLUDE.
	(bfd_elf_set_group_contents): Replace use of abort with an assert.
	(assign_section_numbers): Use resolve_section_groups flag instead
	of relocatable link type.
	(_bfd_elf_init_private_section_data): Use resolve_section_groups
	flag instead of checking the final_link flag for part of the
	checks in here.  Fix white space as a result.
	* elflink.c (elf_link_input_bfd): Use resolve_section_groups flag
	instead of relocatable link type.
	(bfd_elf_final_link): Likewise.

include/ChangeLog:

	* bfdlink.h (struct bfd_link_info): Add new resolve_section_groups
	flag.

ld/ChangeLog:

	* ld.h (struct args_type): Add force_group_allocation field.
	* ldgram.y: Add support for FORCE_GROUP_ALLOCATION.
	* ldlex.h: Likewise.
	* ldlex.l: Likewise.
	* lexsup.c: Likewise.
	* ldlang.c (unique_section_p): Check resolve_section_groups flag
	not the relaxable link flag.
	(lang_add_section): Discard section groups when we're resolving
	groups.  Clear the SEC_LINK_ONCE flag if we're resolving section
	groups.
	* ldmain.c (main): Initialise resolve_section_groups flag in
	link_info based on command line flags.
	* testsuite/ld-elf/group11.d: New file.
	* testsuite/ld-elf/group12.d: New file.
	* testsuite/ld-elf/group12.ld: New file.
	* NEWS: Mention new features.
	* ld.texinfo (Options): Document --force-group-allocation.
	(Miscellaneous Commands): Document FORCE_GROUP_ALLOCATION.
---
 bfd/ChangeLog                  | 14 ++++++++++++++
 bfd/elf.c                      | 35 +++++++++++++++++------------------
 bfd/elflink.c                  |  6 ++++--
 include/ChangeLog              |  5 +++++
 include/bfdlink.h              |  3 +++
 ld/ChangeLog                   | 21 +++++++++++++++++++++
 ld/NEWS                        |  5 +++++
 ld/ld.h                        |  5 +++++
 ld/ld.texinfo                  | 23 +++++++++++++++++++++++
 ld/ldgram.y                    |  4 +++-
 ld/ldlang.c                    | 18 +++++++++++++++---
 ld/ldlex.h                     |  1 +
 ld/ldlex.l                     |  1 +
 ld/ldmain.c                    |  6 ++++++
 ld/lexsup.c                    |  6 ++++++
 ld/testsuite/ld-elf/group11.d  |  6 ++++++
 ld/testsuite/ld-elf/group12.d  |  6 ++++++
 ld/testsuite/ld-elf/group12.ld | 14 ++++++++++++++
 18 files changed, 155 insertions(+), 24 deletions(-)
 create mode 100644 ld/testsuite/ld-elf/group11.d
 create mode 100644 ld/testsuite/ld-elf/group12.d
 create mode 100644 ld/testsuite/ld-elf/group12.ld

diff --git a/bfd/elf.c b/bfd/elf.c
index 863bd61..22e3cfe 100644
--- a/bfd/elf.c
+++ b/bfd/elf.c
@@ -983,7 +983,7 @@ _bfd_elf_make_section_from_shdr (bfd *abfd,
   if (hdr->sh_type != SHT_NOBITS)
     flags |= SEC_HAS_CONTENTS;
   if (hdr->sh_type == SHT_GROUP)
-    flags |= SEC_GROUP | SEC_EXCLUDE;
+    flags |= SEC_GROUP;
   if ((hdr->sh_flags & SHF_ALLOC) != 0)
     {
       flags |= SEC_ALLOC;
@@ -3532,8 +3532,8 @@ bfd_elf_set_group_contents (bfd *abfd, asection *sec, void *failedptrarg)
 	break;
     }
 
-  if ((loc -= 4) != sec->contents)
-    abort ();
+  loc -= 4;
+  BFD_ASSERT (loc == sec->contents);
 
   H_PUT_32 (abfd, sec->flags & SEC_LINK_ONCE ? GRP_COMDAT : 0, loc);
 }
@@ -3608,7 +3608,7 @@ assign_section_numbers (bfd *abfd, struct bfd_link_info *link_info)
   _bfd_elf_strtab_clear_all_refs (elf_shstrtab (abfd));
 
   /* SHT_GROUP sections are in relocatable files only.  */
-  if (link_info == NULL || bfd_link_relocatable (link_info))
+  if (link_info == NULL || !link_info->resolve_section_groups)
     {
       size_t reloc_count = 0;
 
@@ -7444,23 +7444,22 @@ _bfd_elf_init_private_section_data (bfd *ibfd,
      SHT_GROUP section will have its elf_next_in_group pointing back
      to the input group members.  Ignore linker created group section.
      See elfNN_ia64_object_p in elfxx-ia64.c.  */
-  if (!final_link)
+  if ((link_info == NULL
+       || !link_info->resolve_section_groups)
+      && (elf_sec_group (isec) == NULL
+	  || (elf_sec_group (isec)->flags & SEC_LINKER_CREATED) == 0))
     {
-      if (elf_sec_group (isec) == NULL
-	  || (elf_sec_group (isec)->flags & SEC_LINKER_CREATED) == 0)
-	{
-	  if (elf_section_flags (isec) & SHF_GROUP)
-	    elf_section_flags (osec) |= SHF_GROUP;
-	  elf_next_in_group (osec) = elf_next_in_group (isec);
-	  elf_section_data (osec)->group = elf_section_data (isec)->group;
-	}
-
-      /* If not decompress, preserve SHF_COMPRESSED.  */
-      if ((ibfd->flags & BFD_DECOMPRESS) == 0)
-	elf_section_flags (osec) |= (elf_section_flags (isec)
-				     & SHF_COMPRESSED);
+      if (elf_section_flags (isec) & SHF_GROUP)
+	elf_section_flags (osec) |= SHF_GROUP;
+      elf_next_in_group (osec) = elf_next_in_group (isec);
+      elf_section_data (osec)->group = elf_section_data (isec)->group;
     }
 
+  /* If not decompress, preserve SHF_COMPRESSED.  */
+  if (!final_link && (ibfd->flags & BFD_DECOMPRESS) == 0)
+    elf_section_flags (osec) |= (elf_section_flags (isec)
+				 & SHF_COMPRESSED);
+
   ihdr = &elf_section_data (isec)->this_hdr;
 
   /* We need to handle elf_linked_to_section for SHF_LINK_ORDER. We
diff --git a/bfd/elflink.c b/bfd/elflink.c
index 71da4c9..5e8b807 100644
--- a/bfd/elflink.c
+++ b/bfd/elflink.c
@@ -10278,7 +10278,7 @@ elf_link_input_bfd (struct elf_final_link_info *flinfo, bfd *input_bfd)
 	  continue;
 	}
 
-      if (bfd_link_relocatable (flinfo->info)
+      if (!flinfo->info->resolve_section_groups
 	  && (o->flags & (SEC_LINKER_CREATED | SEC_GROUP)) == SEC_GROUP)
 	{
 	  /* Deal with the group signature symbol.  */
@@ -10286,6 +10286,7 @@ elf_link_input_bfd (struct elf_final_link_info *flinfo, bfd *input_bfd)
 	  unsigned long symndx = sec_data->this_hdr.sh_info;
 	  asection *osec = o->output_section;
 
+	  BFD_ASSERT (bfd_link_relocatable (flinfo->info));
 	  if (symndx >= locsymcount
 	      || (elf_bad_symtab (input_bfd)
 		  && flinfo->sections[symndx] == NULL))
@@ -12464,10 +12465,11 @@ bfd_elf_final_link (bfd *abfd, struct bfd_link_info *info)
 	}
     }
 
-  if (bfd_link_relocatable (info))
+  if (!info->resolve_section_groups)
     {
       bfd_boolean failed = FALSE;
 
+      BFD_ASSERT (bfd_link_relocatable (info));
       bfd_map_over_sections (abfd, bfd_elf_set_group_contents, &failed);
       if (failed)
 	goto error_return;
diff --git a/include/bfdlink.h b/include/bfdlink.h
index 371822c..2e3f0b1 100644
--- a/include/bfdlink.h
+++ b/include/bfdlink.h
@@ -348,6 +348,9 @@ struct bfd_link_info
   /* TRUE if all data symbols should be dynamic.  */
   unsigned int dynamic_data: 1;
 
+  /* TRUE if section groups should be resolved.  */
+  unsigned int resolve_section_groups: 1;
+
   /* Which symbols to strip.  */
   ENUM_BITFIELD (bfd_link_strip) strip : 2;
 
diff --git a/ld/NEWS b/ld/NEWS
index 52daa6b..98055b5 100644
--- a/ld/NEWS
+++ b/ld/NEWS
@@ -15,6 +15,11 @@
 * Orphan sections placed after an empty section that has an AT(LMA) will now
   take an load memory address starting from LMA.
 
+* Section groups can now be resolved (the group deleted and the group members
+  placed like normal sections) at partial link time either using the new linker
+  option --force-group-allocation or by placing FORCE_GROUP_ALLOCATION into the
+  linker script.
+
 Changes in 2.28:
 
 * The EXCLUDE_FILE linker script construct can now be applied outside of the
diff --git a/ld/ld.h b/ld/ld.h
index 104bb8e..d9bb653 100644
--- a/ld/ld.h
+++ b/ld/ld.h
@@ -172,6 +172,11 @@ typedef struct
   /* If set, display the target memory usage (per memory region).  */
   bfd_boolean print_memory_usage;
 
+  /* Shold we force section groups to be resolved?  Controlled with
+     --force-group-allocation on the command line or FORCE_GROUP_ALLOCATION
+     in the linker script.  */
+  bfd_boolean force_group_allocation;
+
   /* Big or little endian as set on command line.  */
   enum endian_enum endian;
 
diff --git a/ld/ld.texinfo b/ld/ld.texinfo
index 9a72cb9..9da35a1 100644
--- a/ld/ld.texinfo
+++ b/ld/ld.texinfo
@@ -1486,6 +1486,18 @@
 duplicate when there are many dynamic modules with specialized search
 paths for runtime symbol resolution.
 
+@cindex group allocation in linker script
+@cindex section groups
+@cindex COMDAT
+@kindex --force-group-allocation
+@item --force-group-allocation
+This option causes the linker to place section group members like
+normal input sections, and to delete the section groups.  This is the
+default behaviour for a final link but this option can be used to
+change the behaviour of a relocatable link (@samp{-r}).  The script
+command @code{FORCE_GROUP_ALLOCATION} has the same
+effect. @xref{Miscellaneous Commands}.
+
 @cindex symbols, from command line
 @kindex --defsym=@var{symbol}=@var{exp}
 @item --defsym=@var{symbol}=@var{expression}
@@ -3728,6 +3740,17 @@
 command-line option: to make @code{ld} omit the assignment of addresses
 to common symbols even for a non-relocatable output file.
 
+@item FORCE_GROUP_ALLOCATION
+@kindex FORCE_GROUP_ALLOCATION
+@cindex group allocation in linker script
+@cindex section groups
+@cindex COMDAT
+This command has the same effect as the
+@samp{--force-group-allocation} command-line option: to make
+@command{ld} place section group members like normal input sections,
+and to delete the section groups even if a relocatable output file is
+specified (@samp{-r}).
+
 @item INSERT [ AFTER | BEFORE ] @var{output_section}
 @kindex INSERT
 @cindex insert user script into default script
diff --git a/ld/ldgram.y b/ld/ldgram.y
index 849f272..4c1efdc 100644
--- a/ld/ldgram.y
+++ b/ld/ldgram.y
@@ -131,7 +131,7 @@ static int error_index;
 %token SORT_BY_INIT_PRIORITY
 %token '{' '}'
 %token SIZEOF_HEADERS OUTPUT_FORMAT FORCE_COMMON_ALLOCATION OUTPUT_ARCH
-%token INHIBIT_COMMON_ALLOCATION
+%token INHIBIT_COMMON_ALLOCATION FORCE_GROUP_ALLOCATION
 %token SEGMENT_START
 %token INCLUDE
 %token MEMORY
@@ -336,6 +336,8 @@ ifile_p1:
 		  { ldfile_set_output_arch ($3, bfd_arch_unknown); }
 	|	FORCE_COMMON_ALLOCATION
 		{ command_line.force_common_definition = TRUE ; }
+	|	FORCE_GROUP_ALLOCATION
+		{ command_line.force_group_allocation = TRUE ; }
 	|	INHIBIT_COMMON_ALLOCATION
 		{ command_line.inhibit_common_definition = TRUE ; }
 	|	INPUT '(' input_list ')'
diff --git a/ld/ldlang.c b/ld/ldlang.c
index ed7e552..252400b 100644
--- a/ld/ldlang.c
+++ b/ld/ldlang.c
@@ -207,7 +207,7 @@ unique_section_p (const asection *sec,
   struct unique_sections *unam;
   const char *secnam;
 
-  if (bfd_link_relocatable (&link_info)
+  if (!link_info.resolve_section_groups
       && sec->owner != NULL
       && bfd_is_group_section (sec->owner, sec))
     return !(os != NULL
@@ -2349,6 +2349,12 @@ lang_add_section (lang_statement_list_type *ptr,
   if (strcmp (output->name, DISCARD_SECTION_NAME) == 0)
     discard = TRUE;
 
+  /* Discard the group descriptor sections when we're finally placing the
+     sections from within the group.  */
+  if ((section->flags & SEC_GROUP) == SEC_GROUP
+      && link_info.resolve_section_groups)
+    discard = TRUE;
+
   /* Discard debugging sections if we are stripping debugging
      information.  */
   if ((link_info.strip == strip_debugger || link_info.strip == strip_all)
@@ -2389,8 +2395,14 @@ lang_add_section (lang_statement_list_type *ptr,
      already been processed.  One reason to do this is that on pe
      format targets, .text$foo sections go into .text and it's odd
      to see .text with SEC_LINK_ONCE set.  */
-
-  if (!bfd_link_relocatable (&link_info))
+  if ((flags & (SEC_LINK_ONCE | SEC_GROUP)) == (SEC_LINK_ONCE | SEC_GROUP))
+    {
+      if (link_info.resolve_section_groups)
+        flags &= ~(SEC_LINK_ONCE | SEC_LINK_DUPLICATES | SEC_RELOC);
+      else
+        flags &= ~(SEC_LINK_DUPLICATES | SEC_RELOC);
+    }
+  else if (!bfd_link_relocatable (&link_info))
     flags &= ~(SEC_LINK_ONCE | SEC_LINK_DUPLICATES | SEC_RELOC);
 
   switch (output->sectype)
diff --git a/ld/ldlex.h b/ld/ldlex.h
index dac152b..5aa7f6b 100644
--- a/ld/ldlex.h
+++ b/ld/ldlex.h
@@ -146,6 +146,7 @@ enum option_values
   OPTION_PRINT_MEMORY_USAGE,
   OPTION_REQUIRE_DEFINED_SYMBOL,
   OPTION_ORPHAN_HANDLING,
+  OPTION_FORCE_GROUP_ALLOCATION,
 };
 
 /* The initial parser states.  */
diff --git a/ld/ldlex.l b/ld/ldlex.l
index acba1a2..ba618ec 100644
--- a/ld/ldlex.l
+++ b/ld/ldlex.l
@@ -274,6 +274,7 @@ V_IDENTIFIER [*?.$_a-zA-Z\[\]\-\!\^\\]([*?.$_a-zA-Z0-9\[\]\-\!\^\\]|::)*
 <BOTH,SCRIPT>"CREATE_OBJECT_SYMBOLS"	{ RTOKEN(CREATE_OBJECT_SYMBOLS);}
 <BOTH,SCRIPT>"CONSTRUCTORS"		{ RTOKEN( CONSTRUCTORS);}
 <BOTH,SCRIPT>"FORCE_COMMON_ALLOCATION"	{ RTOKEN(FORCE_COMMON_ALLOCATION);}
+<BOTH,SCRIPT>"FORCE_GROUP_ALLOCATION"	{ RTOKEN(FORCE_GROUP_ALLOCATION);}
 <BOTH,SCRIPT>"INHIBIT_COMMON_ALLOCATION" { RTOKEN(INHIBIT_COMMON_ALLOCATION);}
 <BOTH,SCRIPT>"SECTIONS"			{ RTOKEN(SECTIONS);}
 <BOTH,SCRIPT>"INSERT"			{ RTOKEN(INSERT_K);}
diff --git a/ld/ldmain.c b/ld/ldmain.c
index e049de3..8aa308d 100644
--- a/ld/ldmain.c
+++ b/ld/ldmain.c
@@ -383,6 +383,12 @@ main (int argc, char **argv)
       info_msg ("\n==================================================\n");
     }
 
+  if (command_line.force_group_allocation
+      || !bfd_link_relocatable (&link_info))
+    link_info.resolve_section_groups = TRUE;
+  else
+    link_info.resolve_section_groups = FALSE;
+
   if (command_line.print_output_format)
     info_msg ("%s\n", lang_get_output_target ());
 
diff --git a/ld/lexsup.c b/ld/lexsup.c
index 0b7d497..95c7e59 100644
--- a/ld/lexsup.c
+++ b/ld/lexsup.c
@@ -112,6 +112,9 @@ static const struct ld_option ld_options[] =
     'd', NULL, N_("Force common symbols to be defined"), ONE_DASH },
   { {"dp", no_argument, NULL, 'd'},
     '\0', NULL, NULL, ONE_DASH },
+  { {"force-group-allocation", no_argument, NULL,
+     OPTION_FORCE_GROUP_ALLOCATION},
+    '\0', NULL, N_("Force group members out of groups"), TWO_DASHES },
   { {"entry", required_argument, NULL, 'e'},
     'e', N_("ADDRESS"), N_("Set start address"), TWO_DASHES },
   { {"export-dynamic", no_argument, NULL, OPTION_EXPORT_DYNAMIC},
@@ -767,6 +770,9 @@ parse_args (unsigned argc, char **argv)
 	case 'd':
 	  command_line.force_common_definition = TRUE;
 	  break;
+        case OPTION_FORCE_GROUP_ALLOCATION:
+          command_line.force_group_allocation = TRUE;
+          break;
 	case OPTION_DEFSYM:
 	  lex_string = optarg;
 	  lex_redirect (optarg, "--defsym", ++defsym_count);
diff --git a/ld/testsuite/ld-elf/group11.d b/ld/testsuite/ld-elf/group11.d
new file mode 100644
index 0000000..245cf440
--- /dev/null
+++ b/ld/testsuite/ld-elf/group11.d
@@ -0,0 +1,6 @@
+#source: group1a.s
+#source: group1b.s
+#ld: -r --force-group-allocation
+#readelf: -g
+
+There are no section groups in this file.
diff --git a/ld/testsuite/ld-elf/group12.d b/ld/testsuite/ld-elf/group12.d
new file mode 100644
index 0000000..3fa7be8
--- /dev/null
+++ b/ld/testsuite/ld-elf/group12.d
@@ -0,0 +1,6 @@
+#source: group1a.s
+#source: group1b.s
+#ld: -r -T group12.ld
+#readelf: -g
+
+There are no section groups in this file.
diff --git a/ld/testsuite/ld-elf/group12.ld b/ld/testsuite/ld-elf/group12.ld
new file mode 100644
index 0000000..4a36ee5
--- /dev/null
+++ b/ld/testsuite/ld-elf/group12.ld
@@ -0,0 +1,14 @@
+FORCE_GROUP_ALLOCATION
+
+PHDRS
+{
+  header PT_PHDR PHDRS ;
+  image  PT_LOAD PHDRS;
+}
+
+SECTIONS
+{
+  . = 0x1000;
+  .text : { *(.text) *(.rodata.brlt) } :image :header
+  /DISCARD/ : { *(.dropme) *(.reginfo) *(.MIPS.abiflags) }
+}
-- 
2.5.1