Re: Some ppc fixes breaks on sparc

[Martin Schlemmer <azarah at gentoo dot org>]

On Mon, 2003-11-10 at 02:02, Alan Modra wrote:
> On Sun, Nov 09, 2003 at 05:42:26PM +0200, Martin Schlemmer wrote:
> > http://sources.redhat.com/ml/binutils/2003-11/msg00069.html
> > http://sources.redhat.com/ml/binutils/2003-11/msg00077.html
> > http://sources.redhat.com/ml/binutils/2003-11/msg00082.html
> > 
> > from Alan Modra breaks sparc - basically ld quits all the time
> > (even with simple helloworld.c) with SIGBUS signal.
> > 
> > If you want the specific one, we can try to track it further.
> 
> It can't be the last two, they only touch ppc.  sparc or sparc64?

sparc64 (64bit kernel, 32bit userland)

> I'm a bit surprised my patch broke anything, as I didn't make it in a
> hurry, and ran the binutils testsuite for most of the affected
> architectures.  sparc-linux and sparc64-linux were among the targets
> tested.  Do you have local patches to any of the files listed in the
> first ChangeLog?
> 

The one for Jakub (can drop it if not needed anymore ?) attached, and
your tls-section-alignment.patch.

> When reporting problems about gas and ld, it's useful to provide a
> testcase that triggers the problem.  In the case of a gas problem, we
> want input files to gas and command line switches used.  The inputs to
> gas are _not_ .c or .i files, but rather .s files.  If your original
> source was a C program, you can generate the .s file and see the command
> line options by passing -v -save-temps to gcc in addition to all the
> usual options you use.  The reason we don't want C files is that we
> might not have a C compiler around for the target you use.  While it
> might be possible to build a compiler, that takes considerable time and
> disk space, and we might not end up with exactly the same compiler you
> use.
> 
> In the case of a ld problem, the input files are .o, .a and .so files,
> and possibly a linker script specified with -T.  Again, when using gcc
> to link, you can see these files by adding options to the gcc command
> line.  Use -v -save-temps -Wl,-t, except that on targets that use gcc's
> collect2, you would add -v -save-temps -Wl,-t,-debug.  The -t option
> tells ld to print all files and libraries used, so that, for example,
> you can associate -lc on the ld command line with the actual libc used.
> Note that your simple two line C program to trigger a problem typically
> expands into several megabytes of objects by the time you include
> libraries.  It is antisocial to post megabyte sized attachments to
> mailing lists, so please put large ld testcases somewhere on an ftp or
> web site so that only interested developers need to download them, or
> offer to email them on request.  Better still, try to reduce the
> testcase to not use system libraries.

Jason do you think you can merge binutils-2.14.90.0.7-r2 again, and
include above ?


Thanks,

-- 

Martin Schlemmer



2002-04-20  Jakub Jelinek  <jakub@redhat.com>

	* elf32-sparc.c (elf32_sparc_relocate_section): Find real output
	section with SEC_MERGE.
	* elf64-sparc.c (sparc64_elf_relocate_section): Likewise.

--- bfd/elf32-sparc.c.jj	Thu Feb  6 08:07:10 2003
+++ bfd/elf32-sparc.c	Thu Feb  6 08:31:54 2003
@@ -2168,7 +2168,7 @@ elf32_sparc_relocate_section (output_bfd
       struct elf_link_hash_entry *h;
       Elf_Internal_Sym *sym;
       asection *sec;
-      bfd_vma relocation, off;
+      bfd_vma relocation, off, orig_addend = 0;
       bfd_reloc_status_type r;
       bfd_boolean is_plt = FALSE;
       bfd_boolean unresolved_reloc;
@@ -2196,6 +2196,7 @@ elf32_sparc_relocate_section (output_bfd
 	{
 	  sym = local_syms + r_symndx;
 	  sec = local_sections[r_symndx];
+	  orig_addend = rel->r_addend;
 	  relocation = _bfd_elf_rela_local_sym (output_bfd, sym, sec, rel);
 	}
       else
@@ -2525,6 +2526,24 @@ elf32_sparc_relocate_section (output_bfd
 			  asection *osec;
 
 			  osec = sec->output_section;
+			  /* FIXME: As soon as making sections zero size
+			     is possible, this if can go away.  */
+			  if (bfd_is_abs_section (osec)
+			      && h == NULL
+			      && (sec->flags & SEC_MERGE)
+			      && ELF_ST_TYPE (sym->st_info) == STT_SECTION
+			      && sec->sec_info_type == ELF_INFO_TYPE_MERGE)
+			    {
+			      asection *msec;
+
+			      msec = sec;
+			      _bfd_merged_section_offset (output_bfd, &msec,
+					elf_section_data (sec)->sec_info,
+					sym->st_value + orig_addend,
+					(bfd_vma) 0);
+			      osec = msec->output_section;
+			     }
+
 			  indx = elf_section_data (osec)->dynindx;
 
 			  /* FIXME: we really should be able to link non-pic
--- bfd/elf64-sparc.c.jj	Thu Feb  6 08:07:10 2003
+++ bfd/elf64-sparc.c	Thu Feb  6 08:31:54 2003
@@ -2016,7 +2016,7 @@ sparc64_elf_relocate_section (output_bfd
       struct elf_link_hash_entry *h;
       Elf_Internal_Sym *sym;
       asection *sec;
-      bfd_vma relocation, off;
+      bfd_vma relocation, off, orig_addend = 0;
       bfd_reloc_status_type r;
       bfd_boolean is_plt = FALSE;
       bfd_boolean unresolved_reloc;
@@ -2039,6 +2039,7 @@ sparc64_elf_relocate_section (output_bfd
 	{
 	  sym = local_syms + r_symndx;
 	  sec = local_sections[r_symndx];
+	  orig_addend = rel->r_addend;
 	  relocation = _bfd_elf_rela_local_sym (output_bfd, sym, sec, rel);
 	}
       else
@@ -2267,6 +2268,24 @@ sparc64_elf_relocate_section (output_bfd
 			    asection *osec;
 
 			    osec = sec->output_section;
+			    /* FIXME: As soon as making sections zero size
+			       is possible, this if can go away.  */
+			    if (bfd_is_abs_section (osec)
+				&& h == NULL
+				&& (sec->flags & SEC_MERGE)
+				&& ELF_ST_TYPE (sym->st_info) == STT_SECTION
+				&& sec->sec_info_type == ELF_INFO_TYPE_MERGE)
+			      {
+				asection *msec;
+
+				msec = sec;
+				_bfd_merged_section_offset (output_bfd, &msec,
+					elf_section_data (sec)->sec_info,
+					sym->st_value + orig_addend,
+					(bfd_vma) 0);
+				osec = msec->output_section;
+			      }
+
 			    indx = elf_section_data (osec)->dynindx;
 
 			    /* We are turning this relocation into one

Attachment: signature.asc
Description: This is a digitally signed message part