Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[Kris Warkentin]

I asked this on the gcc mailing list but they suggested to ask binutils.

In the course of building some powerpc startup code we found ld complaining
about "allocated section `.rodata.str1.4' not in segment" and "allocated
section `.eh_frame' not in segment".  I observed that some of the objects in
the startup had .rodata.str1.4 sections in them.  I also observed that if I
took the ld from binutils 2.12.1 to 2.14, the problem went away.

I didn't find any talk of a hard binutils requirement for the 3.3 series of
gcc but perhaps I missed it.  Can anyone shed some light on this?  I hope
I'm not too astonishing in my ignorance but this .rodata.str1.x doesn't seem
to be generated by our older gcc and I don't really know what the
significance of it all is.

Essentially we're in the middle of a release and are stuck at our current
gcc and binutils versions for the moment so if anyone can give me an idea of
how to understand/work around the problem, I'd be very grateful.

cheers,

Kris

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[Alan Modra]

On Fri, Nov 28, 2003 at 01:07:21PM -0500, Kris Warkentin wrote:
> In the course of building some powerpc startup code we found ld complaining
> about "allocated section `.rodata.str1.4' not in segment" and "allocated
> section `.eh_frame' not in segment".  I observed that some of the objects in
> the startup had .rodata.str1.4 sections in them.  I also observed that if I
> took the ld from binutils 2.12.1 to 2.14, the problem went away.

This is very likely a bug in gcc.  I reckon that if you look at your
object file with readelf or objdump, you'll find that this .rodata
section is marked read/write.

The fix is easy.

--- gcc-ppc64-33.orig/gcc/config/rs6000/rs6000.c	2003-09-27 10:07:41.000000000 +0930
+++ gcc-ppc64-33/gcc/config/rs6000/rs6000.c	2003-11-29 12:04:22.000000000 +1030
@@ -13138,14 +13659,8 @@ rs6000_elf_section_type_flags (decl, nam
      const char *name;
      int reloc;
 {
-  unsigned int flags
-    = default_section_type_flags_1 (decl, name, reloc,
-				    flag_pic || DEFAULT_ABI == ABI_AIX);
-
-  if (TARGET_RELOCATABLE)
-    flags |= SECTION_WRITE;
-
-  return flags;
+  return default_section_type_flags_1 (decl, name, reloc,
+				       flag_pic || DEFAULT_ABI == ABI_AIX);
 }
 
 /* Record an element in the table of global constructors.  SYMBOL is


> I didn't find any talk of a hard binutils requirement for the 3.3 series of
> gcc but perhaps I missed it.  Can anyone shed some light on this?  I hope
> I'm not too astonishing in my ignorance but this .rodata.str1.x doesn't seem
> to be generated by our older gcc and I don't really know what the
> significance of it all is.

Sections like .rodata.str1.4 are generated by gcc to support merging of
duplicate constants and strings by ld.  Older gcc's won't do this, nor
will a newer gcc built for a system without the requisite ld support.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[Kris Warkentin]

Interesting.  So it might be an issue with gcc but it looks like the linker
can deal with it.  On our end, it actually turned out that the process that
was spawning ld (our bootable image generator) was using a special link
script that wasn't taking into account the presence of these sections.  When
we put them into the link script, the warnings went away and the image could
boot.  Thanks very much for the help.  I'll try out your patch and see if it
cleans up our objects for us.

cheers,

Kris

----- Original Message ----- 
From: "Alan Modra" <amodra@bigpond.net.au>
To: "Kris Warkentin" <kewarken@qnx.com>
Cc: "Gcc ML" <gcc@gcc.gnu.org>; <binutils@sources.redhat.com>
Sent: Friday, November 28, 2003 8:42 PM
Subject: Re: ppc problem with .rodata.str1.4. binutils requirement for gcc
3.3.1?


> On Fri, Nov 28, 2003 at 01:07:21PM -0500, Kris Warkentin wrote:
> > In the course of building some powerpc startup code we found ld
complaining
> > about "allocated section `.rodata.str1.4' not in segment" and "allocated
> > section `.eh_frame' not in segment".  I observed that some of the
objects in
> > the startup had .rodata.str1.4 sections in them.  I also observed that
if I
> > took the ld from binutils 2.12.1 to 2.14, the problem went away.
>
> This is very likely a bug in gcc.  I reckon that if you look at your
> object file with readelf or objdump, you'll find that this .rodata
> section is marked read/write.
>
> The fix is easy.
>
> --- gcc-ppc64-33.orig/gcc/config/rs6000/rs6000.c 2003-09-27
10:07:41.000000000 +0930
> +++ gcc-ppc64-33/gcc/config/rs6000/rs6000.c 2003-11-29 12:04:22.000000000
+1030
> @@ -13138,14 +13659,8 @@ rs6000_elf_section_type_flags (decl, nam
>       const char *name;
>       int reloc;
>  {
> -  unsigned int flags
> -    = default_section_type_flags_1 (decl, name, reloc,
> -     flag_pic || DEFAULT_ABI == ABI_AIX);
> -
> -  if (TARGET_RELOCATABLE)
> -    flags |= SECTION_WRITE;
> -
> -  return flags;
> +  return default_section_type_flags_1 (decl, name, reloc,
> +        flag_pic || DEFAULT_ABI == ABI_AIX);
>  }
>
>  /* Record an element in the table of global constructors.  SYMBOL is
>
>
> > I didn't find any talk of a hard binutils requirement for the 3.3 series
of
> > gcc but perhaps I missed it.  Can anyone shed some light on this?  I
hope
> > I'm not too astonishing in my ignorance but this .rodata.str1.x doesn't
seem
> > to be generated by our older gcc and I don't really know what the
> > significance of it all is.
>
> Sections like .rodata.str1.4 are generated by gcc to support merging of
> duplicate constants and strings by ld.  Older gcc's won't do this, nor
> will a newer gcc built for a system without the requisite ld support.
>
> -- 
> Alan Modra
> IBM OzLabs - Linux Technology Centre
>

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[David Edelsohn]

>>>>> Alan Modra writes:

Alan> This is very likely a bug in gcc.  I reckon that if you look at your
Alan> object file with readelf or objdump, you'll find that this .rodata
Alan> section is marked read/write.

Alan> The fix is easy.

Alan> --- gcc-ppc64-33.orig/gcc/config/rs6000/rs6000.c	2003-09-27 10:07:41.000000000 +0930
Alan> +++ gcc-ppc64-33/gcc/config/rs6000/rs6000.c	2003-11-29 12:04:22.000000000 +1030
Alan> @@ -13138,14 +13659,8 @@ rs6000_elf_section_type_flags (decl, nam
Alan> const char *name;
Alan> int reloc;
Alan> {
Alan> -  unsigned int flags
Alan> -    = default_section_type_flags_1 (decl, name, reloc,
Alan> -				    flag_pic || DEFAULT_ABI == ABI_AIX);
Alan> -
Alan> -  if (TARGET_RELOCATABLE)
Alan> -    flags |= SECTION_WRITE;
Alan> -
Alan> -  return flags;
Alan> +  return default_section_type_flags_1 (decl, name, reloc,
Alan> +				       flag_pic || DEFAULT_ABI == ABI_AIX);
Alan> }
 
Alan> /* Record an element in the table of global constructors.  SYMBOL is

	I am not sure that we can make this change unilaterally.  The
addition of the SECTION_WRITE flag may be necessary for embedded targets.
We need to understand the reason for the feature before removing it.

David

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[Alan Modra]

On Fri, Nov 28, 2003 at 09:32:25PM -0500, David Edelsohn wrote:
> 	I am not sure that we can make this change unilaterally.  The
> addition of the SECTION_WRITE flag may be necessary for embedded targets.
> We need to understand the reason for the feature before removing it.

See the thread starting with
http://gcc.gnu.org/ml/gcc-patches/2003-10/msg01516.html

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[David Edelsohn]

>>>>> Alan Modra writes:

Alan> See the thread starting with
Alan> http://gcc.gnu.org/ml/gcc-patches/2003-10/msg01516.html

	Okay, how about the following patch.

David

Index: eabi.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/rs6000/eabi.h,v
retrieving revision 1.11
diff -c -p -r1.11 eabi.h
*** eabi.h	10 Jun 2003 16:01:39 -0000	1.11
--- eabi.h	29 Nov 2003 04:03:02 -0000
***************
*** 55,57 ****
--- 55,60 ----
  #define TARGET_E500 (rs6000_cpu == PROCESSOR_PPC8540)
  #define TARGET_ISEL rs6000_isel
  #define TARGET_FPRS (!rs6000_float_gprs)
+ 
+ /* Generate entries in .fixup for relocatable addresses.  */
+ #define RELOCATABLE_NEEDS_FIXUP
Index: rs6000.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/rs6000/rs6000.c,v
retrieving revision 1.550
diff -c -p -r1.550 rs6000.c
*** rs6000.c	14 Nov 2003 20:05:34 -0000	1.550
--- rs6000.c	29 Nov 2003 04:03:02 -0000
*************** toc_section (void)
*** 14991,15002 ****
  static unsigned int
  rs6000_elf_section_type_flags (tree decl, const char *name, int reloc)
  {
!   unsigned int flags
!     = default_section_type_flags_1 (decl, name, reloc,
! 				    flag_pic || DEFAULT_ABI == ABI_AIX);
! 
    if (TARGET_RELOCATABLE)
!     flags |= SECTION_WRITE;
  
    return flags;
  }
--- 14991,15006 ----
  static unsigned int
  rs6000_elf_section_type_flags (tree decl, const char *name, int reloc)
  {
!   int rw_relocs = flag_pic || DEFAULT_ABI == ABI_AIX;
!   unsigned int flags;
!   
! #ifdef RELOCATABLE_NEEDS_FIXUP
    if (TARGET_RELOCATABLE)
!     rw_relocs = 1;
! #endif
! 
!   flags = default_section_type_flags_1 (decl, name, reloc,
! 					flag_pic || DEFAULT_ABI == ABI_AIX);
  
    return flags;
  }
Index: sysv4.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/rs6000/sysv4.h,v
retrieving revision 1.142
diff -c -p -r1.142 sysv4.h
*** sysv4.h	18 Nov 2003 12:56:44 -0000	1.142
--- sysv4.h	29 Nov 2003 04:03:02 -0000
*************** ncrtn.o%s"
*** 1346,1351 ****
  #define TARGET_ASM_EXCEPTION_SECTION readonly_data_section
  
  #define DOUBLE_INT_ASM_OP "\t.quad\t"
- 
- /* Generate entries in .fixup for relocatable addresses.  */
- #define RELOCATABLE_NEEDS_FIXUP
--- 1346,1348 ----

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[Andrew Pinski]

On Nov 28, 2003, at 20:06, David Edelsohn wrote:

> ! 					flag_pic || DEFAULT_ABI == ABI_AIX);


except this should be rw_relocs instead.

Thanks,
Andrew Pinski

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[David Edelsohn]

	Corrected patch.

Index: eabi.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/rs6000/eabi.h,v
retrieving revision 1.11
diff -c -p -r1.11 eabi.h
*** eabi.h	10 Jun 2003 16:01:39 -0000	1.11
--- eabi.h	29 Nov 2003 04:19:16 -0000
***************
*** 55,57 ****
--- 55,60 ----
  #define TARGET_E500 (rs6000_cpu == PROCESSOR_PPC8540)
  #define TARGET_ISEL rs6000_isel
  #define TARGET_FPRS (!rs6000_float_gprs)
+ 
+ /* Generate entries in .fixup for relocatable addresses.  */
+ #define RELOCATABLE_NEEDS_FIXUP
Index: rs6000.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/rs6000/rs6000.c,v
retrieving revision 1.550
diff -c -p -r1.550 rs6000.c
*** rs6000.c	14 Nov 2003 20:05:34 -0000	1.550
--- rs6000.c	29 Nov 2003 04:19:17 -0000
*************** toc_section (void)
*** 14991,15002 ****
  static unsigned int
  rs6000_elf_section_type_flags (tree decl, const char *name, int reloc)
  {
!   unsigned int flags
!     = default_section_type_flags_1 (decl, name, reloc,
! 				    flag_pic || DEFAULT_ABI == ABI_AIX);
! 
    if (TARGET_RELOCATABLE)
!     flags |= SECTION_WRITE;
  
    return flags;
  }
--- 14991,15005 ----
  static unsigned int
  rs6000_elf_section_type_flags (tree decl, const char *name, int reloc)
  {
!   int rw_relocs = flag_pic || DEFAULT_ABI == ABI_AIX;
!   unsigned int flags;
!   
! #ifdef RELOCATABLE_NEEDS_FIXUP
    if (TARGET_RELOCATABLE)
!     rw_relocs = 1;
! #endif
! 
!   flags = default_section_type_flags_1 (decl, name, reloc, rw_relocs);
  
    return flags;
  }
Index: sysv4.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/rs6000/sysv4.h,v
retrieving revision 1.142
diff -c -p -r1.142 sysv4.h
*** sysv4.h	18 Nov 2003 12:56:44 -0000	1.142
--- sysv4.h	29 Nov 2003 04:19:17 -0000
*************** ncrtn.o%s"
*** 1346,1351 ****
  #define TARGET_ASM_EXCEPTION_SECTION readonly_data_section
  
  #define DOUBLE_INT_ASM_OP "\t.quad\t"
- 
- /* Generate entries in .fixup for relocatable addresses.  */
- #define RELOCATABLE_NEEDS_FIXUP
--- 1346,1348 ----

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[Alan Modra]

On Fri, Nov 28, 2003 at 11:20:03PM -0500, David Edelsohn wrote:
> 	Corrected patch.

Geoff analyzed the problem enough to see that TARGET_RELOCATABLE
implies flag_pic == 2, so there's no need for any special hacks.

See http://gcc.gnu.org/ml/gcc-patches/2003-10/msg01575.html

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

Re: ppc problem with .rodata.str1.4. binutils requirement for gcc 3.3.1?

[David Edelsohn]

>>>>> Alan Modra writes:

Alan> Geoff analyzed the problem enough to see that TARGET_RELOCATABLE
Alan> implies flag_pic == 2, so there's no need for any special hacks.

Alan> See http://gcc.gnu.org/ml/gcc-patches/2003-10/msg01575.html

	Okay, go ahead and apply you patch to
rs6000_elf_section_type_flags().

	We may still want to move RELOCATABLE_NEEDS_FIXUP from sysv4.h to
eabi.h. 

David