[PATCH] Fix ppc32 dl-machine.h

[PATCH] Fix ppc32 dl-machine.h

[Jakub Jelinek]

Hi!

It seems the 2003-03-02 cleanups broke relocs against local symbols on ppc32
(e.g. a bunch of binutils tests now fail).
On ppc32 (and sparc32) for those relocs r_addend already includes st_value,
so ATM glibc adds them in twice.

  [ 6] .rodata           PROGBITS        00000918 000918 000010 00   A  0  0  4
  [ 7] .data             PROGBITS        00010928 000928 000008 00  WA  0  0  4

00000876  00000706 R_PPC_ADDR16_HA       00010928   .data + 10928
00000882  00000704 R_PPC_ADDR16_LO       00010928   .data + 10928
000008a2  00000706 R_PPC_ADDR16_HA       00010928   .data + 1092c
000008ae  00000704 R_PPC_ADDR16_LO       00010928   .data + 1092c
0000089e  00000606 R_PPC_ADDR16_HA       00000918   .rodata + 918
000008a6  00000604 R_PPC_ADDR16_LO       00000918   .rodata + 918
000008aa  00000604 R_PPC_ADDR16_LO       00000918   .rodata + 918

The patch is untested, but IMHO should work. Will test later today.

2003-03-25  Jakub Jelinek  <jakub@redhat.com>

	* sysdeps/powerpc/powerpc32/dl-machine.h (elf_machine_rela): Restore
	special handling of relocations against local symbols.

--- libc/sysdeps/powerpc/powerpc32/dl-machine.h.jj	2003-03-06 12:26:23.000000000 -0500
+++ libc/sysdeps/powerpc/powerpc32/dl-machine.h	2003-03-25 05:36:57.000000000 -0500
@@ -361,6 +361,9 @@ elf_machine_rela (struct link_map *map, 
   const Elf32_Sym *const refsym = sym;
   Elf32_Addr value;
   const int r_type = ELF32_R_TYPE (reloc->r_info);
+#if defined USE_TLS && !defined RTLD_BOOTSTRAP
+  struct link_map *sym_map;
+#endif
 
   if (r_type == R_PPC_RELATIVE)
     {
@@ -371,16 +374,24 @@ elf_machine_rela (struct link_map *map, 
   if (__builtin_expect (r_type == R_PPC_NONE, 0))
     return;
 
+  /* binutils on ppc32 includes st_value in r_addend for relocations
+     against local symbols.  */
+  if (__builtin_expect (ELF32_ST_BIND (sym->st_info) == STB_LOCAL, 0)
+      && sym->st_shndx != SHN_UNDEF)
+    value = map->l_addr;
+  else
+    {
 #if defined USE_TLS && !defined RTLD_BOOTSTRAP
-  struct link_map *sym_map = RESOLVE_MAP (&sym, version, r_type);
-  value = sym == NULL ? 0 : sym_map->l_addr + sym->st_value;
+      sym_map = RESOLVE_MAP (&sym, version, r_type);
+      value = sym == NULL ? 0 : sym_map->l_addr + sym->st_value;
 #else
-  value = RESOLVE (&sym, version, r_type);
+      value = RESOLVE (&sym, version, r_type);
 # ifndef RTLD_BOOTSTRAP
-  if (sym != NULL)
+      if (sym != NULL)
 # endif
-    value += sym->st_value;
+	value += sym->st_value;
 #endif
+    }
   value += reloc->r_addend;
 
   /* A small amount of code is duplicated here for speed.  In libc,

	Jakub

Re: [PATCH] Fix ppc32 dl-machine.h

[Jakub Jelinek]

On Tue, Mar 25, 2003 at 11:40:55AM +0100, Jakub Jelinek wrote:
> The patch is untested, but IMHO should work. Will test later today.

Tested, works just fine and fixes all binutils test failures.

> 2003-03-25  Jakub Jelinek  <jakub@redhat.com>
> 
> 	* sysdeps/powerpc/powerpc32/dl-machine.h (elf_machine_rela): Restore
> 	special handling of relocations against local symbols.

	Jakub

Re: [PATCH] Fix ppc32 dl-machine.h

[Roland McGrath]

> It seems the 2003-03-02 cleanups broke relocs against local symbols on ppc32
> (e.g. a bunch of binutils tests now fail).
> On ppc32 (and sparc32) for those relocs r_addend already includes st_value,
> so ATM glibc adds them in twice.

I can't find anything in the ABI spec to suggest this should be done.  Is
it considered part of the spec, or a binutils bug that we are just stuck with?
I had the impression that this was a sparc-specific bug, not the specified
behavior.  Do other Rela platforms use this meaning?  


Thanks,
Roland

Re: [PATCH] Fix ppc32 dl-machine.h

[Roland McGrath]

I put it in.  ld should definitely be fixed, but we must support the
existing binaries.