* sysdeps/powerpc/powerpc32/dl-machine.h (elf_machine_runtime_setup):

	Handle prelinked libraries and binaries with new style PLT.

2005-06-07  Jakub Jelinek  <jakub@redhat.com>

	* elf/elf.h (R_PPC_REL16, R_PPC_REL16_LO, R_PPC_REL16_HI,
	R_PPC_REL16_HA): Define.

2005-06-14  Alan Modra  <amodra@bigpond.net.au>

	* config.h.in (HAVE_ASM_PPC_REL16): Add.
	* elf/elf.h (DT_PPC_GOT, DT_PPC_NUM): Define.
	* elf/tls-macros.h (PowerPC32): Include config.h.  Add variants of
	TLS_IE, TLS_LD and TLS_GD for new PLT/GOT layout.
	* sysdeps/powerpc/powerpc32/configure.in: New file,
	* sysdeps/powerpc/powerpc32/dl-dtprocnum.h: New file.
	* sysdeps/powerpc/powerpc32/dl-machine.h (DT_PPC): Define.
	(ppc_got): New inline function.
	(elf_machine_dynamic): Use ppc_got.  Add attribute const.
	(elf_machine_load_address): Add attribute const.  Don't use int vars.
	Use bcl rather than bl to save trashing branch target stack.  Use
	elf_machine_dynamic rather than duplicating code here.
	(elf_machine_runtime_setup): New inline function replacing define.
	Handle new PLT.
	(elf_machine_fixup_plt): Handle new PLT.
	(elf_machine_rela): Likewise.
	* sysdeps/powerpc/powerpc32/sysdep.h: Include config.h.
	(CALL_MCOUNT): Don't set up counter vars.
	* sysdeps/powerpc/powerpc32/ppc-mcount.S: Correct comment.
	* sysdeps/powerpc/powerpc32/elf/start.S (start_addressesp): Don't
	define when HAVE_ASM_PPC_REL16.
	(_start): Add HAVE_ASM_PPC_REL16 code.
	* sysdeps/powerpc/powerpc32/dl-start.S (_dl_start_user): Don't bl
	into the GOT when HAVE_ASM_PPC_REL16.
	* sysdeps/powerpc/powerpc32/memset.S (memset): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S (__longjmp): Ditto.
	* sysdeps/powerpc/powerpc32/fpu/s_ceil.S (__ceil): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_ceilf.S (__ceilf): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_floor.S (__floor): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_floorf.S (__floorf): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_lround.S (__lround): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_rint.S (__rint): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_rintf.S (__rintf): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_round.S (__round): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_roundf.S (__roundf): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_trunc.S (__trunc): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_truncf.S (__truncf): Likewise.
	* sysdeps/powerpc/powerpc32/fpu/setjmp-common.S (__sigsetjmp):
	Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/brk.S (__brk): Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/getcontext.S
	(__getcontext): Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/setcontext.S
	(__setcontext): Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/swapcontext.S
	(__swapcontext): Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/socket.S (stackblock):
	Comment.
	(__socket): Bomb if NARGS >= 7.  Invoke CGOTSETUP and CGOTRESTORE.

2005-06-17  Ulrich Drepper  <drepper@redhat.com>

	* sysdeps/posix/sigignore.c: Include <string.h> to tell the compiler
	to use __GI_memset.
	* sysdeps/posix/signal.c: Likewise.
	* sysdeps/posix/sigset.c: Likewise.
	* sysdeps/posix/sysv_signal.c: Likewise.
	* sysdeps/unix/sysv/linux/sleep.c: Likewise.
	* sysdeps/unix/sysv/linux/sysctl.c: Likewise.
	* sysdeps/unix/sysv/linux/system.c: Likewise.

22 files changed, 292 insertions, 62 deletions

diff --git a/sysdeps/powerpc/powerpc32/configure b/sysdeps/powerpc/powerpc32/configure
new file mode 100644
index 0000000000..9ebac38b57
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/configure
@@ -0,0 +1,33 @@
+# This file is generated from configure.in by Autoconf.  DO NOT EDIT!
+ # Local configure fragment for sysdeps/powerpc/powerpc32.
+
+# See whether gas has R_PPC_REL16 relocs.
+echo "$as_me:$LINENO: checking for R_PPC_REL16 gas support" >&5
+echo $ECHO_N "checking for R_PPC_REL16 gas support... $ECHO_C" >&6
+if test "${libc_cv_ppc_rel16+set}" = set; then
+  echo $ECHO_N "(cached) $ECHO_C" >&6
+else
+  cat > conftest.s <<\EOF
+	.text
+	addis 11,30,_GLOBAL_OFFSET_TABLE_-.@ha
+EOF
+if { ac_try='${CC-cc} -c $CFLAGS conftest.s 1>&5'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; }; then
+  libc_cv_ppc_rel16=yes
+else
+  libc_cv_ppc_rel16=no
+fi
+rm -f conftest*
+fi
+echo "$as_me:$LINENO: result: $libc_cv_ppc_rel16" >&5
+echo "${ECHO_T}$libc_cv_ppc_rel16" >&6
+if test $libc_cv_ppc_rel16 = yes; then
+  cat >>confdefs.h <<\_ACEOF
+#define HAVE_ASM_PPC_REL16 1
+_ACEOF
+
+fi
diff --git a/sysdeps/powerpc/powerpc32/configure.in b/sysdeps/powerpc/powerpc32/configure.in
new file mode 100644
index 0000000000..6d2c41cb9a
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/configure.in
@@ -0,0 +1,18 @@
+GLIBC_PROVIDES dnl See aclocal.m4 in the top level source directory.
+# Local configure fragment for sysdeps/powerpc/powerpc32.
+
+# See whether gas has R_PPC_REL16 relocs.
+AC_CACHE_CHECK(for R_PPC_REL16 gas support, libc_cv_ppc_rel16, [dnl
+cat > conftest.s <<\EOF
+	.text
+	addis 11,30,_GLOBAL_OFFSET_TABLE_-.@ha
+EOF
+if AC_TRY_COMMAND(${CC-cc} -c $CFLAGS conftest.s 1>&AS_MESSAGE_LOG_FD); then
+  libc_cv_ppc_rel16=yes
+else
+  libc_cv_ppc_rel16=no
+fi
+rm -f conftest*])
+if test $libc_cv_ppc_rel16 = yes; then
+  AC_DEFINE(HAVE_ASM_PPC_REL16)
+fi
diff --git a/sysdeps/powerpc/powerpc32/dl-dtprocnum.h b/sysdeps/powerpc/powerpc32/dl-dtprocnum.h
new file mode 100644
index 0000000000..7fe2be7939
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/dl-dtprocnum.h
@@ -0,0 +1,3 @@
+/* Number of extra dynamic section entries for this architecture.  By
+   default there are none.  */
+#define DT_THISPROCNUM	DT_PPC_NUM
diff --git a/sysdeps/powerpc/powerpc32/dl-machine.h b/sysdeps/powerpc/powerpc32/dl-machine.h
index 545c19b300..496fa71ecc 100644
--- a/sysdeps/powerpc/powerpc32/dl-machine.h
+++ b/sysdeps/powerpc/powerpc32/dl-machine.h
@@ -25,6 +25,10 @@
 #include <assert.h>
 #include <dl-tls.h>
 
+/* Translate a processor specific dynamic tag to the index
+   in l_info array.  */
+#define DT_PPC(x) (DT_PPC_##x - DT_LOPROC + DT_NUM)
+
 /* Return nonzero iff ELF header is compatible with the running host.  */
 static inline int
 elf_machine_matches_host (const Elf32_Ehdr *ehdr)
@@ -32,24 +36,38 @@ elf_machine_matches_host (const Elf32_Ehdr *ehdr)
   return ehdr->e_machine == EM_PPC;
 }
 
+/* Return the value of the GOT pointer.  */
+static inline Elf32_Addr * __attribute__ ((const))
+ppc_got (void)
+{
+  Elf32_Addr *got;
+#ifdef HAVE_ASM_PPC_REL16
+  asm ("bcl 20,31,1f\n"
+       "1:	mflr %0\n"
+       "	addis %0,%0,_GLOBAL_OFFSET_TABLE_-1b@ha\n"
+       "	addi %0,%0,_GLOBAL_OFFSET_TABLE_-1b@l\n"
+       : "=b" (got) : : "lr");
+#else
+  asm (" bl _GLOBAL_OFFSET_TABLE_-4@local"
+       : "=l" (got));
+#endif
+  return got;
+}
 
 /* Return the link-time address of _DYNAMIC, stored as
    the first value in the GOT. */
-static inline Elf32_Addr
+static inline Elf32_Addr __attribute__ ((const))
 elf_machine_dynamic (void)
 {
-  Elf32_Addr *got;
-  asm (" bl _GLOBAL_OFFSET_TABLE_-4@local"
-       : "=l"(got));
-  return *got;
+  return *ppc_got ();
 }
 
 /* Return the run-time load address of the shared object.  */
-static inline Elf32_Addr
+static inline Elf32_Addr __attribute__ ((const))
 elf_machine_load_address (void)
 {
-  unsigned int *got;
-  unsigned int *branchaddr;
+  Elf32_Addr *branchaddr;
+  Elf32_Addr runtime_dynamic;
 
   /* This is much harder than you'd expect.  Possibly I'm missing something.
      The 'obvious' way:
@@ -80,19 +98,17 @@ elf_machine_load_address (void)
      the address ourselves. That gives us the following code: */
 
   /* Get address of the 'b _DYNAMIC@local'...  */
-  asm ("bl 0f ;"
+  asm ("bcl 20,31,0f;"
        "b _DYNAMIC@local;"
        "0:"
-       : "=l"(branchaddr));
-
-  /* ... and the address of the GOT.  */
-  asm (" bl _GLOBAL_OFFSET_TABLE_-4@local"
-       : "=l"(got));
+       : "=l" (branchaddr));
 
   /* So now work out the difference between where the branch actually points,
      and the offset of that location in memory from the start of the file.  */
-  return ((Elf32_Addr)branchaddr - *got
-	  + ((int)(*branchaddr << 6 & 0xffffff00) >> 6));
+  runtime_dynamic = ((Elf32_Addr) branchaddr
+		     + ((Elf32_Sword) (*branchaddr << 6 & 0xffffff00) >> 6));
+
+  return runtime_dynamic - elf_machine_dynamic ();
 }
 
 #define ELF_MACHINE_BEFORE_RTLD_RELOC(dynamic_info) /* nothing */
@@ -144,13 +160,69 @@ __elf_preferred_address(struct link_map *loader, size_t maplength,
 /* The PowerPC never uses REL relocations.  */
 #define ELF_MACHINE_NO_REL 1
 
-/* Set up the loaded object described by L so its unrelocated PLT
+/* Set up the loaded object described by MAP so its unrelocated PLT
    entries will jump to the on-demand fixup code in dl-runtime.c.
    Also install a small trampoline to be used by entries that have
    been relocated to an address too far away for a single branch.  */
 extern int __elf_machine_runtime_setup (struct link_map *map,
 					int lazy, int profile);
-#define elf_machine_runtime_setup __elf_machine_runtime_setup
+
+static inline int
+elf_machine_runtime_setup (struct link_map *map,
+			   int lazy, int profile)
+{
+  if (map->l_info[DT_JMPREL] == 0)
+    return lazy;
+
+  if (map->l_info[DT_PPC(GOT)] == 0)
+    /* Handle old style PLT.  */
+    return __elf_machine_runtime_setup (map, lazy, profile);
+
+  /* New style non-exec PLT consisting of an array of addresses.  */
+  map->l_info[DT_PPC(GOT)]->d_un.d_ptr += map->l_addr;
+  if (lazy)
+    {
+      Elf32_Addr *plt, *got, glink;
+      Elf32_Word num_plt_entries;
+      void (*dlrr) (void);
+      extern void _dl_runtime_resolve (void);
+      extern void _dl_prof_resolve (void);
+
+      if (__builtin_expect (!profile, 1))
+	dlrr = _dl_runtime_resolve;
+      else
+	{
+	  if (GLRO(dl_profile) != NULL
+	      &&_dl_name_match_p (GLRO(dl_profile), map))
+	    GL(dl_profile_map) = map;
+	  dlrr = _dl_prof_resolve;
+	}
+      got = (Elf32_Addr *) map->l_info[DT_PPC(GOT)]->d_un.d_ptr;
+      glink = got[1];
+      got[1] = (Elf32_Addr) dlrr;
+      got[2] = (Elf32_Addr) map;
+
+      /* Relocate everything in .plt by the load address offset.  */
+      plt = (Elf32_Addr *) D_PTR (map, l_info[DT_PLTGOT]);
+      num_plt_entries = (map->l_info[DT_PLTRELSZ]->d_un.d_val
+			 / sizeof (Elf32_Rela));
+
+      /* If a library is prelinked but we have to relocate anyway,
+	 we have to be able to undo the prelinking of .plt section.
+	 The prelinker saved us at got[1] address of .glink
+	 section's start.  */
+      if (glink)
+	{
+	  glink += map->l_addr;
+	  while (num_plt_entries-- != 0)
+	    *plt++ = glink, glink += 4;
+	}
+      else
+	while (num_plt_entries-- != 0)
+	  *plt++ += map->l_addr;
+    }
+  return lazy;
+}
 
 /* Change the PLT entry whose reloc is 'reloc' to call the actual routine.  */
 extern Elf32_Addr __elf_machine_fixup_plt (struct link_map *map,
@@ -163,7 +235,12 @@ elf_machine_fixup_plt (struct link_map *map, lookup_t t,
 		       const Elf32_Rela *reloc,
 		       Elf32_Addr *reloc_addr, Elf64_Addr finaladdr)
 {
-  return __elf_machine_fixup_plt (map, reloc, reloc_addr, finaladdr);
+  if (map->l_info[DT_PPC(GOT)] == 0)
+    /* Handle old style PLT.  */
+    return __elf_machine_fixup_plt (map, reloc, reloc_addr, finaladdr);
+
+  *reloc_addr = finaladdr;
+  return finaladdr;
 }
 
 /* Return the final value of a plt relocation.  */
@@ -286,11 +363,16 @@ elf_machine_rela (struct link_map *map, const Elf32_Rela *reloc,
       break;
 #endif /* USE_TLS etc. */
 
-#ifdef RESOLVE_CONFLICT_FIND_MAP
     case R_PPC_JMP_SLOT:
+#ifdef RESOLVE_CONFLICT_FIND_MAP
       RESOLVE_CONFLICT_FIND_MAP (map, reloc_addr);
-      /* FALLTHROUGH */
 #endif
+      if (map->l_info[DT_PPC(GOT)] != 0)
+	{
+	  *reloc_addr = value;
+	  break;
+	}
+      /* FALLTHROUGH */
 
     default:
       __process_machine_rela (map, reloc, sym_map, sym, refsym,
diff --git a/sysdeps/powerpc/powerpc32/dl-start.S b/sysdeps/powerpc/powerpc32/dl-start.S
index d72202d4a4..e1f7f6e24a 100644
--- a/sysdeps/powerpc/powerpc32/dl-start.S
+++ b/sysdeps/powerpc/powerpc32/dl-start.S
@@ -47,8 +47,15 @@ ENTRY(_dl_start_user)
    passed by value!).  */
 
 /*  Put our GOT pointer in r31, */
+#ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r31
+	addis	r31,r31,_GLOBAL_OFFSET_TABLE_-1b@ha
+	addi	r31,r31,_GLOBAL_OFFSET_TABLE_-1b@l
+#else
 	bl	_GLOBAL_OFFSET_TABLE_-4@local
 	mflr	r31
+#endif
 /*  the address of _start in r30, */
 	mr	r30,r3
 /*  &_dl_argc in 29, &_dl_argv in 27, and _dl_loaded in 28.  */
diff --git a/sysdeps/powerpc/powerpc32/elf/start.S b/sysdeps/powerpc/powerpc32/elf/start.S
index 7827357a6c..bafd2ae001 100644
--- a/sysdeps/powerpc/powerpc32/elf/start.S
+++ b/sysdeps/powerpc/powerpc32/elf/start.S
@@ -52,7 +52,7 @@ L(start_addresses):
 	ASM_SIZE_DIRECTIVE(L(start_addresses))
 
 	.section ".text"
-#ifdef PIC
+#if defined PIC && !defined HAVE_ASM_PPC_REL16
 L(start_addressesp):
 	.long	L(start_addresses)-L(branch)
 #endif
@@ -73,11 +73,19 @@ L(branch):
 	mtlr	r0
 	stw	r0,0(r1)
  /* Set r13 to point at the 'small data area', and put the address of
-    start_addresses in r8...  */
+    start_addresses in r8.  Also load the GOT pointer so that new PLT
+    calls work, like the one to __libc_start_main.  */
 #ifdef PIC
+# ifdef HAVE_ASM_PPC_REL16
+	addis	r30,r13,_GLOBAL_OFFSET_TABLE_-L(branch)@ha
+	addis	r8,r13,L(start_addresses)-L(branch)@ha
+	addi	r30,r30,_GLOBAL_OFFSET_TABLE_-L(branch)@l
+	lwzu	r13,L(start_addresses)-L(branch)@l(r8)
+# else
 	lwz	r8,L(start_addressesp)-L(branch)(r13)
 	add	r8,r13,r8
 	lwz	r13,0(r8)
+# endif
 #else
 	lis	r8,L(start_addresses)@ha
 	lwzu	r13,L(start_addresses)@l(r8)
diff --git a/sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S b/sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S
index 6dfe6a67d5..73cc8181f9 100644
--- a/sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S
+++ b/sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S
@@ -34,8 +34,15 @@ ENTRY (BP_SYM (__longjmp))
 #ifndef __NO_VMX__
 # ifdef PIC
 	mflr    r6
+#  ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r5
+	addis	r5,r5,_GLOBAL_OFFSET_TABLE_-1b@ha
+	addi	r5,r5,_GLOBAL_OFFSET_TABLE_-1b@l
+#  else
 	bl      _GLOBAL_OFFSET_TABLE_@local-4
 	mflr    r5
+#  endif
 #  ifdef SHARED
 	lwz     r5,_rtld_global_ro@got(r5)
 	mtlr    r6
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_ceil.S b/sysdeps/powerpc/powerpc32/fpu/s_ceil.S
index 7924e34648..13afba88f0 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_ceil.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_ceil.S
@@ -29,11 +29,19 @@ ENTRY (__ceil)
 	mffs	fp11		/* Save current FPU rounding mode.  */
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	mtlr	r11
+	lfs	fp13,.LC0-1b@l(r9)
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
 	lfs	fp13,0(r9)
+# endif
 #else
 	lis	r9,.LC0@ha
 	lfs	fp13,.LC0@l(r9)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S b/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
index 9315d8d2df..f8ca1de08c 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
@@ -20,7 +20,7 @@
 #include <sysdep.h>
 
 	.section	.rodata.cst4,"aM",@progbits,4
-	.align 2
+	.align	2
 .LC0:	/* 2**23 */
 	.long 0x4b000000
 
@@ -29,11 +29,19 @@ ENTRY (__ceilf)
 	mffs	fp11		/* Save current FPU rounding mode.  */
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	mtlr	r11
+	lfs	fp13,.LC0-1b@l(r9)
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
 	lfs	fp13,0(r9)
+# endif
 #else
 	lis	r9,.LC0@ha
 	lfs	fp13,.LC0@l(r9)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_floor.S b/sysdeps/powerpc/powerpc32/fpu/s_floor.S
index c8f59c24a6..5dfe8f2d9a 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_floor.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_floor.S
@@ -29,11 +29,19 @@ ENTRY (__floor)
 	mffs	fp11		/* Save current FPU rounding mode.  */
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	mtlr	r11
+	lfs	fp13,.LC0-1b@l(r9)
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
 	lfs	fp13,0(r9)
+# endif
 #else
 	lis	r9,.LC0@ha
 	lfs	fp13,.LC0@l(r9)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_floorf.S b/sysdeps/powerpc/powerpc32/fpu/s_floorf.S
index 8ee0644ac9..31b71ad229 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_floorf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_floorf.S
@@ -20,7 +20,7 @@
 #include <sysdep.h>
 
 	.section	.rodata.cst4,"aM",@progbits,4
-	.align 2
+	.align	2
 .LC0:	/* 2**23 */
 	.long 0x4b000000
 
@@ -29,11 +29,19 @@ ENTRY (__floorf)
 	mffs	fp11		/* Save current FPU rounding mode.  */
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	mtlr	r11
+	lfs	fp13,.LC0-1b@l(r9)
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
 	lfs	fp13,0(r9)
+# endif
 #else
 	lis	r9,.LC0@ha
 	lfs	fp13,.LC0@l(r9)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_lround.S b/sysdeps/powerpc/powerpc32/fpu/s_lround.S
index 72fd49ba46..a85743164c 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_lround.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_lround.S
@@ -41,9 +41,16 @@
 ENTRY (__lround)
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	addi	r9,r9,.LC0-1b@l
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
+# endif
 	mtlr	r11
 	lfs	fp12,0(r9)
 #else
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_rint.S b/sysdeps/powerpc/powerpc32/fpu/s_rint.S
index 4abdcedfe8..1cfcd78b5c 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_rint.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_rint.S
@@ -31,11 +31,19 @@
 ENTRY (__rint)
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	mtlr	r11
+	lfs	fp13,.LC0-1b@l(r9)
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
 	lfs	fp13,0(r9)
+# endif
 #else
 	lis	r9,.LC0@ha
 	lfs	fp13,.LC0@l(r9)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_rintf.S b/sysdeps/powerpc/powerpc32/fpu/s_rintf.S
index d02bd066b8..93c02667fe 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_rintf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_rintf.S
@@ -20,7 +20,7 @@
 #include <sysdep.h>
 
 	.section	.rodata.cst4,"aM",@progbits,4
-	.align 2
+	.align	2
 .LC0:	/* 2**23 */
 	.long 0x4b000000
 
@@ -28,11 +28,19 @@
 ENTRY (__rintf)
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	mtlr	r11
+	lfs	fp13,.LC0-1b@l(r9)
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
 	lfs	fp13,0(r9)
+# endif
 #else
 	lis	r9,.LC0@ha
 	lfs	fp13,.LC0@l(r9)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_round.S b/sysdeps/powerpc/powerpc32/fpu/s_round.S
index 96fc2984fd..53b45916d1 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_round.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_round.S
@@ -41,9 +41,16 @@ ENTRY (__round)
 	mffs	fp11		/* Save current FPU rounding mode.  */
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	addi	r9,r9,.LC0-1b@l
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
+# endif
 	mtlr	r11
 	lfs	fp13,0(r9)
 #else
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_roundf.S b/sysdeps/powerpc/powerpc32/fpu/s_roundf.S
index 87965dea80..39ba08655a 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_roundf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_roundf.S
@@ -41,9 +41,16 @@ ENTRY (__roundf )
 	mffs	fp11		/* Save current FPU rounding mode.  */
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	addi	r9,r9,.LC0-1b@l
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
+# endif
 	mtlr	r11
 	lfs	fp13,0(r9)
 #else
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_trunc.S b/sysdeps/powerpc/powerpc32/fpu/s_trunc.S
index 7a3e705a81..827e8cb940 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_trunc.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_trunc.S
@@ -36,11 +36,19 @@ ENTRY (__trunc)
 	mffs	fp11		/* Save current FPU rounding mode.  */
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	mtlr	r11
+	lfs	fp13,.LC0-1b@l(r9)
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
 	lfs	fp13,0(r9)
+# endif
 #else
 	lis	r9,.LC0@ha
 	lfs	fp13,.LC0@l(r9)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_truncf.S b/sysdeps/powerpc/powerpc32/fpu/s_truncf.S
index 5275c69d29..55e7a74b41 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_truncf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_truncf.S
@@ -20,7 +20,7 @@
 #include <sysdep.h>
 
 	.section	.rodata.cst4,"aM",@progbits,4
-	.align 2
+	.align	2
 .LC0:	/* 2**23 */
 	.long 0x4b000000
 
@@ -36,11 +36,19 @@ ENTRY (__truncf)
 	mffs	fp11		/* Save current FPU rounding mode.  */
 #ifdef SHARED
 	mflr	r11
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r9
+	addis	r9,r9,.LC0-1b@ha
+	mtlr	r11
+	lfs	fp13,.LC0-1b@l(r9)
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
 	lfs	fp13,0(r9)
+# endif
 #else
 	lis	r9,.LC0@ha
 	lfs	fp13,.LC0@l(r9)
diff --git a/sysdeps/powerpc/powerpc32/fpu/setjmp-common.S b/sysdeps/powerpc/powerpc32/fpu/setjmp-common.S
index be2cf4d69d..cf3f215f2d 100644
--- a/sysdeps/powerpc/powerpc32/fpu/setjmp-common.S
+++ b/sysdeps/powerpc/powerpc32/fpu/setjmp-common.S
@@ -76,8 +76,15 @@ ENTRY (BP_SYM (__sigsetjmp))
 #ifndef __NO_VMX__
 # ifdef PIC
 	mflr    r6
+#  ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	r5
+	addis	r5,r5,_GLOBAL_OFFSET_TABLE_-1b@ha
+	addi	r5,r5,_GLOBAL_OFFSET_TABLE_-1b@l
+#  else
 	bl      _GLOBAL_OFFSET_TABLE_@local-4
 	mflr    r5
+#  endif
 #  ifdef SHARED
 	lwz     r5,_rtld_global_ro@got(r5)
 	mtlr    r6
diff --git a/sysdeps/powerpc/powerpc32/memset.S b/sysdeps/powerpc/powerpc32/memset.S
index 4c0edc8e45..f09c294674 100644
--- a/sysdeps/powerpc/powerpc32/memset.S
+++ b/sysdeps/powerpc/powerpc32/memset.S
@@ -264,10 +264,17 @@ L(checklinesize):
 	beq	L(medium)
 /* Establishes GOT addressability so we can load __cache_line_size
    from static. This value was set from the aux vector during startup.  */
+# ifdef HAVE_ASM_PPC_REL16
+	bcl	20,31,1f
+1:	mflr	rGOT
+	addis	rGOT,rGOT,__cache_line_size-1b@ha
+	lwz	rCLS,__cache_line_size-1b@l(rGOT)
+# else
 	bl	_GLOBAL_OFFSET_TABLE_@local-4
 	mflr	rGOT
 	lwz	rGOT,__cache_line_size@got(rGOT)
 	lwz	rCLS,0(rGOT)
+# endif
 	mtlr	rTMP
 #else
 /* Load __cache_line_size from static. This value was set from the
diff --git a/sysdeps/powerpc/powerpc32/ppc-mcount.S b/sysdeps/powerpc/powerpc32/ppc-mcount.S
index 314c8ee703..c1a08d379c 100644
--- a/sysdeps/powerpc/powerpc32/ppc-mcount.S
+++ b/sysdeps/powerpc/powerpc32/ppc-mcount.S
@@ -1,5 +1,5 @@
 /* PowerPC-specific implementation of profiling support.
-   Copyright (C) 1997, 1999 Free Software Foundation, Inc.
+   Copyright (C) 1997, 1999, 2005 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -24,25 +24,19 @@
 
 #include <sysdep.h>
 
-/* We do profiling as described in the SYSV ELF ABI, _mcount is called
-   with the address of a data word in r0 (that is different for every
-   routine, initialised to 0, and otherwise unused).  The caller has put
-   the address the caller will return to in the usual place on the stack,
-   4(r1).  _mcount is responsible for ensuring that when it returns no
-   argument-passing registers are disturbed, and that the LR is set back
-   to (what the caller sees as) 4(r1).
+/* We do profiling as described in the SYSV ELF ABI, except that glibc
+   _mcount manages its own counters.  The caller has put the address the
+   caller will return to in the usual place on the stack, 4(r1).  _mcount
+   is responsible for ensuring that when it returns no argument-passing
+   registers are disturbed, and that the LR is set back to (what the
+   caller sees as) 4(r1).
 
    This is intended so that the following code can be inserted at the
    front of any routine without changing the routine:
 
 	.data
-	.align	2
-   0:	.long	0
-	.previous
 	mflr	r0
-	lis	r11,0b@ha
 	stw	r0,4(r1)
-	addi	r0,r11,0b@l
 	bl	_mcount
 */
 
diff --git a/sysdeps/powerpc/powerpc32/sysdep.h b/sysdeps/powerpc/powerpc32/sysdep.h
index 775073f325..552f595a10 100644
--- a/sysdeps/powerpc/powerpc32/sysdep.h
+++ b/sysdeps/powerpc/powerpc32/sysdep.h
@@ -29,31 +29,10 @@
 /* The mcount code relies on a the return address being on the stack
    to locate our caller and so it can restore it; so store one just
    for its benefit.  */
-# ifdef PIC
-#  define CALL_MCOUNT							      \
-  .pushsection;								      \
-  .section ".data";    							      \
-  .align ALIGNARG(2);							      \
-0:.long 0;								      \
-  .previous;								      \
-  mflr  r0;								      \
-  stw   r0,4(r1);	       						      \
-  bl    _GLOBAL_OFFSET_TABLE_@local-4;					      \
-  mflr  r11;								      \
-  lwz   r0,0b@got(r11);							      \
-  bl    JUMPTARGET(_mcount);
-# else  /* PIC */
-#  define CALL_MCOUNT							      \
-  .section ".data";							      \
-  .align ALIGNARG(2);							      \
-0:.long 0;								      \
-  .previous;								      \
+# define CALL_MCOUNT							      \
   mflr  r0;								      \
-  lis   r11,0b@ha;		       					      \
   stw   r0,4(r1);	       						      \
-  addi  r0,r11,0b@l;							      \
   bl    JUMPTARGET(_mcount);
-# endif /* PIC */
 #else  /* PROF */
 # define CALL_MCOUNT		/* Do nothing.  */
 #endif /* PROF */