Re: cvs binutils+gcc4.0 build fail

Re: cvs binutils+gcc4.0 build fail

[Alan Modra]

On Wed, Mar 23, 2005 at 08:53:08AM +1030, Alan Modra wrote:
> On Wed, Mar 23, 2005 at 08:32:19AM +1100, Ben Elliston wrote:
> > > Does this look familiar?
> > 
> > Not immediately.
> > 
> > > /usr/local/ppc64-4.0/bin/powerpc-linux-ld:
> > > .libs/wstring-inst.o(.eh_frame+0x13c8): R_PPC64_REL64 reloc against
> > > `.gnu.linkonce.t._ZNKSbIwSt11char_traitsIwESaIwEE7compareEmmRKS2_mm':
> > > error 2
> > 
> > So _bfd_final_link_relocate is returning bfd_reloc_outofrange.  I
> > suspect the remaining errors just cascade from that.  As to why this
> > is happening, I don't know.  Alan?
> 
> My guess is that the .gnu.linkonce.* section in question has been
> removed because it's a duplicate, and something has gone wrong with
> .eh_frame editing.  We have had a lot of problems with linkonce
> sections in the past, but I thought it was all OK now..

Something was indeed going wrong with eh_frame editing.  It was being
done twice due to my 2005-03-21 ppc64elf.em change.  Oops.

	* emultempl/ppc64elf.em (need_laying_out): Delete.  Remove all refs.
	(ppc_finish): Don't call bfd_elf_discard_info.
	* emultempl/hppaelf.em: Similarly.

Index: ld/emultempl/ppc64elf.em
===================================================================
RCS file: /cvs/src/src/ld/emultempl/ppc64elf.em,v
retrieving revision 1.40
diff -u -p -r1.40 ppc64elf.em
--- ld/emultempl/ppc64elf.em	21 Mar 2005 09:39:53 -0000	1.40
+++ ld/emultempl/ppc64elf.em	29 Mar 2005 06:37:13 -0000
@@ -32,9 +32,6 @@ cat >>e${EMULATION_NAME}.c <<EOF
 static lang_input_statement_type *stub_file;
 static int stub_added = 0;
 
-/* Whether we need to call ppc_layout_sections_again.  */
-static int need_laying_out = 0;
-
 /* Maximum size of a group of input sections that can be handled by
    one stub section.  A value of +/-1 indicates the bfd back-end
    should use a suitable default size.  */
@@ -258,8 +255,6 @@ ppc_layout_sections_again (void)
   /* If we have changed sizes of the stub sections, then we need
      to recalculate all the section offsets.  This may mean we need to
      add even more stubs.  */
-  need_laying_out = 0;
-
   lang_reset_memory_regions ();
 
   /* Resize the sections.  */
@@ -321,13 +316,6 @@ ppc_finish (void)
      descriptor in the .opd section.  */
   entry_section = ".opd";
 
-  /* bfd_elf_discard_info just plays with debugging sections,
-     ie. doesn't affect any code, so we can delay resizing the
-     sections.  It's likely we'll resize everything in the process of
-     adding stubs.  */
-  if (bfd_elf_discard_info (output_bfd, &link_info))
-    need_laying_out = 1;
-
   /* If generating a relocatable output file, then we don't have any
      stubs.  */
   if (stub_file != NULL && !link_info.relocatable)
@@ -356,9 +344,6 @@ ppc_finish (void)
 	}
     }
 
-  if (need_laying_out)
-    ppc_layout_sections_again ();
-
   if (link_info.relocatable)
     {
       asection *toc = bfd_get_section_by_name (output_bfd, ".toc");
Index: ld/emultempl/hppaelf.em
===================================================================
RCS file: /cvs/src/src/ld/emultempl/hppaelf.em,v
retrieving revision 1.38
diff -u -p -r1.38 hppaelf.em
--- ld/emultempl/hppaelf.em	21 Mar 2005 09:39:53 -0000	1.38
+++ ld/emultempl/hppaelf.em	29 Mar 2005 06:37:13 -0000
@@ -36,9 +36,6 @@ static lang_input_statement_type *stub_f
    stubs.  */
 static int multi_subspace = 0;
 
-/* Whether we need to call hppa_layout_sections_again.  */
-static int need_laying_out = 0;
-
 /* Maximum size of a group of input sections that can be handled by
    one stub section.  A value of +/-1 indicates the bfd back-end
    should use a suitable default size.  */
@@ -220,8 +217,6 @@ hppaelf_layout_sections_again (void)
   /* If we have changed sizes of the stub sections, then we need
      to recalculate all the section offsets.  This may mean we need to
      add even more stubs.  */
-  need_laying_out = 0;
-
   lang_reset_memory_regions ();
 
   /* Resize the sections.  */
@@ -258,13 +253,6 @@ build_section_lists (lang_statement_unio
 static void
 hppaelf_finish (void)
 {
-  /* bfd_elf_discard_info just plays with debugging sections,
-     ie. doesn't affect any code, so we can delay resizing the
-     sections.  It's likely we'll resize everything in the process of
-     adding stubs.  */
-  if (bfd_elf_discard_info (output_bfd, &link_info))
-    need_laying_out = 1;
-
   /* If generating a relocatable output file, then we don't
      have to examine the relocs.  */
   if (stub_file != NULL && !link_info.relocatable)
@@ -296,9 +284,6 @@ hppaelf_finish (void)
 	}
     }
 
-  if (need_laying_out)
-    hppaelf_layout_sections_again ();
-
   if (! link_info.relocatable)
     {
       /* Set the global data pointer.  */

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

Re: cvs binutils+gcc4.0 build fail

[Alan Modra]

On Tue, Mar 29, 2005 at 04:21:39PM +0930, Alan Modra wrote:
> Something was indeed going wrong with eh_frame editing.  It was being
> done twice due to my 2005-03-21 ppc64elf.em change.  Oops.

Blah, it's worse than that.  .eh_frame needs to be edited before stubs
and suchlike are built.  If the section layout changes after we're done
adding stubs, then some of the stubs will have incorrect offsets.

	* emultempl/elf32.em (gld${EMULATION_NAME}_layout_sections_again):
	New function, extracted from static void gld${EMULATION_NAME}_finish.
	(gld${EMULATION_NAME}_strip_empty_sections): Likewise.
	(gld${EMULATION_NAME}_provide_init_fini_syms): Likewise.
	* emultempl/ppc64elf.em: Revert last change.
	(ppc_layout_sections_again): Use
	gld${EMULATION_NAME}_layout_sections_again.
	(ppc_finish): Don't call gld${EMULATION_NAME}_finish.  Instead call
	gld${EMULATION_NAME}_strip_empty_sections and
	gld${EMULATION_NAME}_provide_init_fini_syms.
	* emultempl/hppaelf.em: Similarly.

Index: ld/emultempl/elf32.em
===================================================================
RCS file: /cvs/src/src/ld/emultempl/elf32.em,v
retrieving revision 1.136
diff -c -p -r1.136 elf32.em
*** ld/emultempl/elf32.em	23 Mar 2005 15:35:50 -0000	1.136
--- ld/emultempl/elf32.em	29 Mar 2005 10:53:24 -0000
*************** static void gld${EMULATION_NAME}_after_o
*** 59,65 ****
  static void gld${EMULATION_NAME}_before_allocation (void);
  static bfd_boolean gld${EMULATION_NAME}_place_orphan
    (lang_input_statement_type *file, asection *s);
! static void gld${EMULATION_NAME}_finish (void);
  
  EOF
  
--- 59,68 ----
  static void gld${EMULATION_NAME}_before_allocation (void);
  static bfd_boolean gld${EMULATION_NAME}_place_orphan
    (lang_input_statement_type *file, asection *s);
! static void gld${EMULATION_NAME}_layout_sections_again (void);
! static void gld${EMULATION_NAME}_strip_empty_sections (void);
! static void gld${EMULATION_NAME}_provide_init_fini_syms (void);
! static void gld${EMULATION_NAME}_finish (void) ATTRIBUTE_UNUSED;
  
  EOF
  
*************** gld${EMULATION_NAME}_provide_bound_symbo
*** 1450,1474 ****
    _bfd_elf_provide_symbol (&link_info, end, end_val);
  }
  
  static void
! gld${EMULATION_NAME}_finish (void)
  {
!   if (bfd_elf_discard_info (output_bfd, &link_info))
      {
!       lang_reset_memory_regions ();
! 
!       /* Resize the sections.  */
!       lang_size_sections (stat_ptr->head, abs_output_section,
! 			  &stat_ptr->head, 0, (bfd_vma) 0, NULL, TRUE);
! 
!       /* Redo special stuff.  */
!       ldemul_after_allocation ();
! 
!       /* Do the assignments again.  */
!       lang_do_assignments (stat_ptr->head, abs_output_section,
! 			   (fill_type *) 0, (bfd_vma) 0);
      }
  
    if (!link_info.relocatable)
      {
        lang_output_section_statement_type *os;
--- 1453,1510 ----
    _bfd_elf_provide_symbol (&link_info, end, end_val);
  }
  
+ /* If not building a shared library, provide
+ 
+    __preinit_array_start
+    __preinit_array_end
+    __init_array_start
+    __init_array_end
+    __fini_array_start
+    __fini_array_end
+ 
+    They are set here rather than via PROVIDE in the linker
+    script, because using PROVIDE inside an output section
+    statement results in unnecessary output sections.  Using
+    PROVIDE outside an output section statement runs the risk of
+    section alignment affecting where the section starts.  */
+ 
  static void
! gld${EMULATION_NAME}_provide_init_fini_syms (void)
  {
!   if (!link_info.relocatable && !link_info.shared)
      {
!       gld${EMULATION_NAME}_provide_bound_symbols (".preinit_array",
! 						  "__preinit_array_start",
! 						  "__preinit_array_end");
!       gld${EMULATION_NAME}_provide_bound_symbols (".init_array",
! 						  "__init_array_start",
! 						  "__init_array_end");
!       gld${EMULATION_NAME}_provide_bound_symbols (".fini_array",
! 						  "__fini_array_start",
! 						  "__fini_array_end");
      }
+ }
  
+ static void
+ gld${EMULATION_NAME}_layout_sections_again (void)
+ {
+   lang_reset_memory_regions ();
+ 
+   /* Resize the sections.  */
+   lang_size_sections (stat_ptr->head, abs_output_section,
+ 		      &stat_ptr->head, 0, (bfd_vma) 0, NULL, TRUE);
+ 
+   /* Redo special stuff.  */
+   ldemul_after_allocation ();
+ 
+   /* Do the assignments again.  */
+   lang_do_assignments (stat_ptr->head, abs_output_section,
+ 		       (fill_type *) 0, (bfd_vma) 0);
+ }
+ 
+ static void
+ gld${EMULATION_NAME}_strip_empty_sections (void)
+ {
    if (!link_info.relocatable)
      {
        lang_output_section_statement_type *os;
*************** gld${EMULATION_NAME}_finish (void)
*** 1495,1529 ****
  		  }
  	    }
  	}
  
!       /* If not building shared library, provide
  
! 	 __preinit_array_start
! 	 __preinit_array_end
! 	 __init_array_start
! 	 __init_array_end
! 	 __fini_array_start
! 	 __fini_array_end
! 
! 	 They are set here rather than via PROVIDE in the linker
! 	 script, because using PROVIDE inside an output section
! 	 statement results in unnecessary output sections.  Using
! 	 PROVIDE outside an output section statement runs the risk of
! 	 section alignment affecting where the section starts.  */
! 
!       if (!link_info.shared)
! 	{
! 	  gld${EMULATION_NAME}_provide_bound_symbols
! 	    (".preinit_array", "__preinit_array_start",
! 	     "__preinit_array_end");
! 	  gld${EMULATION_NAME}_provide_bound_symbols
! 	    (".init_array", "__init_array_start",
! 	     "__init_array_end");
! 	  gld${EMULATION_NAME}_provide_bound_symbols
! 	    (".fini_array", "__fini_array_start",
! 	     "__fini_array_end");
! 	}
!     }
  }
  EOF
  fi
--- 1531,1547 ----
  		  }
  	    }
  	}
+     }
+ }
  
! static void
! gld${EMULATION_NAME}_finish (void)
! {
!   if (bfd_elf_discard_info (output_bfd, &link_info))
!     gld${EMULATION_NAME}_layout_sections_again ();
  
!   gld${EMULATION_NAME}_strip_empty_sections ();
!   gld${EMULATION_NAME}_provide_init_fini_syms ();
  }
  EOF
  fi
Index: ld/emultempl/ppc64elf.em
===================================================================
RCS file: /cvs/src/src/ld/emultempl/ppc64elf.em,v
retrieving revision 1.41
diff -u -p -r1.41 ppc64elf.em
--- ld/emultempl/ppc64elf.em	29 Mar 2005 06:52:22 -0000	1.41
+++ ld/emultempl/ppc64elf.em	29 Mar 2005 10:33:56 -0000
@@ -32,6 +32,9 @@ cat >>e${EMULATION_NAME}.c <<EOF
 static lang_input_statement_type *stub_file;
 static int stub_added = 0;
 
+/* Whether we need to call ppc_layout_sections_again.  */
+static int need_laying_out = 0;
+
 /* Maximum size of a group of input sections that can be handled by
    one stub section.  A value of +/-1 indicates the bfd back-end
    should use a suitable default size.  */
@@ -255,17 +258,9 @@ ppc_layout_sections_again (void)
   /* If we have changed sizes of the stub sections, then we need
      to recalculate all the section offsets.  This may mean we need to
      add even more stubs.  */
-  lang_reset_memory_regions ();
-
-  /* Resize the sections.  */
-  lang_size_sections (stat_ptr->head, abs_output_section,
-		      &stat_ptr->head, 0, 0, NULL, TRUE);
-
-  /* Recalculate TOC base.  */
-  ldemul_after_allocation ();
+  need_laying_out = 0;
 
-  /* Do the assignments again.  */
-  lang_do_assignments (stat_ptr->head, abs_output_section, NULL, 0);
+  gld${EMULATION_NAME}_layout_sections_again ();
 }
 
 
@@ -316,6 +311,13 @@ ppc_finish (void)
      descriptor in the .opd section.  */
   entry_section = ".opd";
 
+  /* bfd_elf_discard_info just plays with debugging sections,
+     ie. doesn't affect any code, so we can delay resizing the
+     sections.  It's likely we'll resize everything in the process of
+     adding stubs.  */
+  if (bfd_elf_discard_info (output_bfd, &link_info))
+    need_laying_out = 1;
+
   /* If generating a relocatable output file, then we don't have any
      stubs.  */
   if (stub_file != NULL && !link_info.relocatable)
@@ -344,6 +346,9 @@ ppc_finish (void)
 	}
     }
 
+  if (need_laying_out)
+    ppc_layout_sections_again ();
+
   if (link_info.relocatable)
     {
       asection *toc = bfd_get_section_by_name (output_bfd, ".toc");
@@ -374,7 +379,8 @@ ppc_finish (void)
     }
 
   ppc64_elf_restore_symbols (&link_info);
-  gld${EMULATION_NAME}_finish ();
+  gld${EMULATION_NAME}_strip_empty_sections ();
+  gld${EMULATION_NAME}_provide_init_fini_syms ();
 }
 
 
Index: ld/emultempl/hppaelf.em
===================================================================
RCS file: /cvs/src/src/ld/emultempl/hppaelf.em,v
retrieving revision 1.39
diff -u -p -r1.39 hppaelf.em
--- ld/emultempl/hppaelf.em	29 Mar 2005 06:52:22 -0000	1.39
+++ ld/emultempl/hppaelf.em	29 Mar 2005 10:33:56 -0000
@@ -36,6 +35,9 @@ static lang_input_statement_type *stub_f
    stubs.  */
 static int multi_subspace = 0;
 
+/* Whether we need to call hppa_layout_sections_again.  */
+static int need_laying_out = 0;
+
 /* Maximum size of a group of input sections that can be handled by
    one stub section.  A value of +/-1 indicates the bfd back-end
    should use a suitable default size.  */
@@ -217,18 +219,9 @@ hppaelf_layout_sections_again (void)
   /* If we have changed sizes of the stub sections, then we need
      to recalculate all the section offsets.  This may mean we need to
      add even more stubs.  */
-  lang_reset_memory_regions ();
+  need_laying_out = 0;
 
-  /* Resize the sections.  */
-  lang_size_sections (stat_ptr->head, abs_output_section,
-		      &stat_ptr->head, 0, (bfd_vma) 0, NULL, TRUE);
-
-  /* Redo special stuff.  */
-  ldemul_after_allocation ();
-
-  /* Do the assignments again.  */
-  lang_do_assignments (stat_ptr->head, abs_output_section,
-		       (fill_type *) 0, (bfd_vma) 0);
+  gld${EMULATION_NAME}_layout_sections_again ();
 }
 
 
@@ -253,6 +246,13 @@ build_section_lists (lang_statement_unio
 static void
 hppaelf_finish (void)
 {
+  /* bfd_elf_discard_info just plays with debugging sections,
+     ie. doesn't affect any code, so we can delay resizing the
+     sections.  It's likely we'll resize everything in the process of
+     adding stubs.  */
+  if (bfd_elf_discard_info (output_bfd, &link_info))
+    need_laying_out = 1;
+
   /* If generating a relocatable output file, then we don't
      have to examine the relocs.  */
   if (stub_file != NULL && !link_info.relocatable)
@@ -284,6 +284,9 @@ hppaelf_finish (void)
 	}
     }
 
+  if (need_laying_out)
+    hppaelf_layout_sections_again ();
+
   if (! link_info.relocatable)
     {
       /* Set the global data pointer.  */
@@ -297,14 +300,12 @@ hppaelf_finish (void)
       if (stub_file != NULL && stub_file->the_bfd->sections != NULL)
 	{
 	  if (! elf32_hppa_build_stubs (&link_info))
-	    {
-	      einfo ("%X%P: can not build stubs: %E\n");
-	      return;
-	    }
+	    einfo ("%X%P: can not build stubs: %E\n");
 	}
     }
 
-  gld${EMULATION_NAME}_finish ();
+  gld${EMULATION_NAME}_strip_empty_sections ();
+  gld${EMULATION_NAME}_provide_init_fini_syms ();
 }
 
 

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

Re: cvs binutils+gcc4.0 build fail

[Thorsten Glaser]

Alan Modra dixit:

>On Tue, Mar 29, 2005 at 04:21:39PM +0930, Alan Modra wrote:
>> Something was indeed going wrong with eh_frame editing.  It was being
>> done twice due to my 2005-03-21 ppc64elf.em change.  Oops.
>
>Blah, it's worse than that.  .eh_frame needs to be edited before stubs
>and suchlike are built.  If the section layout changes after we're done
>adding stubs, then some of the stubs will have incorrect offsets.

I've got a similar problem; just applying your emultempl/elf32.em diff
on top of a binutils-20050328 snapshot didn't help :-(

This is gcc-3.4.4 (20050325).


tg@odem:/home/tg $ g++ -static x.cc
/usr/bin/ld: `.gnu.linkonce.t._ZNKSt5ctypeIcE13_M_widen_initEv' referenced in section `.data' of /usr/lib/libstdc++.a(locale-inst.o): defined in discarded section `.gnu.linkonce.t._ZNKSt5ctypeIcE13_M_widen_initEv' of /usr/lib/libstdc++.a(locale-inst.o)

/usr/bin/ld: `.gnu.linkonce.t._ZStneIcSt11char_traitsIcESaIcEEbRKSbIT_T0_T1_EPKS3_' referenced in section `.data' of /usr/lib/libstdc++.a(locale_init.o): defined in discarded section `.gnu.linkonce.t._ZStneIcSt11char_traitsIcESaIcEEbRKSbIT_T0_T1_EPKS3_' of /usr/lib/libstdc++.a(locale_init.o)

tg@odem:/home/tg $ ./a.out
Segmentation fault (core dumped)
tg@odem:/home/tg $ cat x.cc
#include <iostream>
using namespace std;
main() { cout << "Hello, World!" << endl; }
tg@odem:/home/tg $ uname -a
MirBSD odem.66h.42h.de 8 Kv#8n24-20050315 GENERIC#660 i386


Do you have an idea if this is a binutils or a gcc problem?

Thanks in advance,
//mirabile, who doesn't know C++ but got the task to port it to MirOS

PS: Yes, this target isn't supported (yet) in stock binutils,
    but I've got my papers done, just got no time to split our
    diff up into smaller pieces and feed back.
-- 
> [...] Echtzeit hat weniger mit "Speed"[...] zu tun, sondern damit, daß der
> richtige Prozeß voraussagbar rechtzeitig sein Zeitscheibchen bekommt.
Wir haben uns[...] geeinigt, dass das verwendete Echtzeit-Betriebssystem[...]
weil selbst einfachste Operationen *echt* *Zeit* brauchen.	(aus d.a.s.r)

Re: cvs binutils+gcc4.0 build fail

[Alan Modra]

On Tue, Mar 29, 2005 at 07:09:31PM +0000, Thorsten Glaser wrote:
> tg@odem:/home/tg $ g++ -static x.cc
> /usr/bin/ld: `.gnu.linkonce.t._ZNKSt5ctypeIcE13_M_widen_initEv' referenced in section `.data' of /usr/lib/libstdc++.a(locale-inst.o): defined in discarded section `.gnu.linkonce.t._ZNKSt5ctypeIcE13_M_widen_initEv' of /usr/lib/libstdc++.a(locale-inst.o)

This is a gcc problem.  Some part of _ZNKSt5ctypeIcE13_M_widen_initEv is
not being emitted to linkonce sections, but instead finds its way into
.data.  This leaves a dangling reference when the linkonce sections are
removed due to some other copy of _ZNKSt5ctypeIcE13_M_widen_initEv
already being present.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre