include/elf/

	* ppc.h (DT_PPC_TLSOPT): Define.
	* ppc64.h (DT_PPC64_TLSOPT): Define.
bfd/
	* elf32-ppc.c (TLS_GET_ADDR_GLINK_SIZE): Define.
	(ADD_3_12_2, BEQLR, CMPWI_11_0, LWZ_11_3, LWZ_12_3): Define.
	(MR_0_3, MR_3_0): Define.
	(struct ppc_elf_link_hash_table): Add no_tls_get_addr_opt.
	(ppc_elf_select_plt_layout): Save emit_stub_syms param earlier.
	(ppc_elf_tls_setup): Add no_tls_get_addr_opt param and save to hash
	table.  Check for presense of __tls_get_addr_opt
	(allocate_dynrelocs): Increase glink entry size for __tls_get_addr.
	(ppc_elf_size_dynamic_sections): Add DT_PPC_TLS_OPT tag.
	(write_glink_stub): Add param p.
	(ppc_elf_relocate_section): Adjust write_glink_stub call.
	(ppc_elf_finish_dynamic_symbol): Emit special glink call stub for
	__tls_get_addr.
	* elf32-ppc.h (ppc_elf_tls_setup): Update prototype.
	* elf64-ppc.c (struct ppc_link_hash_table): Add no_tls_get_addr_opt.
	(ppc64_elf_tls_setup): Add no_tls_get_addr_opt param and save to hash
	table.  Check for presense of __tls_get_addr_opt.
	(ppc64_elf_size_dynamic_sections): Add DT_PPC64_TLS_OPT tag.
	(LD_R11_0R3, LD_R12_0R3, MR_R0_R3, CMPDI_R11_0, ADD_R3_R12_R13,
	BEQLR, MR_R3_R0, MFLR_R11, STD_R11_0R1, BCTRL, LD_R11_0R1,
	LD_R2_0R1, MTLR_R11): Define.
	(build_tls_get_addr_stub): New function.
	(ppc_build_one_stub): Call it.
	(ppc_size_one_stub): Add extra size for __tls_get_addr stub.
	(ppc64_elf_relocate_section): Don't change nop to ld 2,40(1) for
	__tls_get_addr plt call.
	* elf64-ppc.h (ppc64_elf_tls_setup): Update prototype.
binutils/
	* readelf.c (get_ppc_dynamic_type): Add TLSOPT.
	(get_ppc64_dynamic_type): Likewise.
ld/
	* emultempl/ppc32elf.em (no_tls_get_addr_opt): New var.
	(ppc_before_allocation): Pass to ppc_elf_tls_setup.
	(OPTION_NO_TLS_GET_ADDR_OPT): Define.  Redefine other options in
	terms of previous option.
	(PARSE_AND_LIST_LONGOPTS, PARSE_AND_LIST_OPTIONS): Add
	--no-tls-get-addr-optimize.
	(PARSE_AND_LIST_ARGS_CASES): Handle it.
	* emultempl/ppc64elf.em (no_tls_get_addr_opt): New var.
	(ppc_before_allocation): Pass to ppc64_elf_tls_setup.
	(OPTION_NO_TLS_GET_ADDR_OPT): Define.
	(PARSE_AND_LIST_LONGOPTS, PARSE_AND_LIST_OPTIONS): Add
	--no-tls-get-addr-optimize.
	(PARSE_AND_LIST_ARGS_CASES): Handle it.
ld/testsuite/
	* ld-powerpc/tlslib.s: Delete dot-symbol entry syms.  Add
	__tls_get_addr_opt.
	* ld-powerpc/tlslib32.s: Add __tls_get_addr_opt.
	* ld-powerpc/oldtlslib.s: New file, old-abi version of tlslib.s.
	* ld-powerpc/powerpc.exp: Build old-abi library and use it in
	two new link tests.
	* ld-powerpc/tlsexe.d: Update for new __tls_get_addr stub.
	* ld-powerpc/tlsexe.g, * ld-powerpc/tlsexe.r, *ld-powerpc/tlsexe32.d,
	* ld-powerpc/tlsexe32.g, * ld-powerpc/tlsexe32.r,
	* ld-powerpc/tlsexetoc.d, * ld-powerpc/tlsexetoc.g,
	* ld-powerpc/tlsexetoc.r: Likewise.

5 files changed, 284 insertions, 16 deletions

diff --git a/bfd/ChangeLog b/bfd/ChangeLog
index a4f99e7..9900119 100644
--- a/bfd/ChangeLog
+++ b/bfd/ChangeLog
@@ -1,3 +1,33 @@
+2009-09-21  Alan Modra  <amodra@bigpond.net.au>
+
+	* elf32-ppc.c (TLS_GET_ADDR_GLINK_SIZE): Define.
+	(ADD_3_12_2, BEQLR, CMPWI_11_0, LWZ_11_3, LWZ_12_3): Define.
+	(MR_0_3, MR_3_0): Define.
+	(struct ppc_elf_link_hash_table): Add no_tls_get_addr_opt.
+	(ppc_elf_select_plt_layout): Save emit_stub_syms param earlier.
+	(ppc_elf_tls_setup): Add no_tls_get_addr_opt param and save to hash
+	table.  Check for presense of __tls_get_addr_opt
+	(allocate_dynrelocs): Increase glink entry size for __tls_get_addr.
+	(ppc_elf_size_dynamic_sections): Add DT_PPC_TLS_OPT tag.
+	(write_glink_stub): Add param p.
+	(ppc_elf_relocate_section): Adjust write_glink_stub call.
+	(ppc_elf_finish_dynamic_symbol): Emit special glink call stub for
+	__tls_get_addr.
+	* elf32-ppc.h (ppc_elf_tls_setup): Update prototype.
+	* elf64-ppc.c (struct ppc_link_hash_table): Add no_tls_get_addr_opt.
+	(ppc64_elf_tls_setup): Add no_tls_get_addr_opt param and save to hash
+	table.  Check for presense of __tls_get_addr_opt.
+	(ppc64_elf_size_dynamic_sections): Add DT_PPC64_TLS_OPT tag.
+	(LD_R11_0R3, LD_R12_0R3, MR_R0_R3, CMPDI_R11_0, ADD_R3_R12_R13,
+	BEQLR, MR_R3_R0, MFLR_R11, STD_R11_0R1, BCTRL, LD_R11_0R1,
+	LD_R2_0R1, MTLR_R11): Define.
+	(build_tls_get_addr_stub): New function.
+	(ppc_build_one_stub): Call it.
+	(ppc_size_one_stub): Add extra size for __tls_get_addr stub.
+	(ppc64_elf_relocate_section): Don't change nop to ld 2,40(1) for
+	__tls_get_addr plt call.
+	* elf64-ppc.h (ppc64_elf_tls_setup): Update prototype.
+
 2009-09-19  Richard Sandiford  <rdsandiford@googlemail.com>
 
 	* elf-bfd.h (eh_cie_fde): Add personality_offset and
diff --git a/bfd/elf32-ppc.c b/bfd/elf32-ppc.c
index bbe94db..95058a2 100644
--- a/bfd/elf32-ppc.c
+++ b/bfd/elf32-ppc.c
@@ -61,6 +61,7 @@ static bfd_reloc_status_type ppc_elf_unhandled_reloc
 /* For new-style .glink and .plt.  */
 #define GLINK_PLTRESOLVE 16*4
 #define GLINK_ENTRY_SIZE 4*4
+#define TLS_GET_ADDR_GLINK_SIZE 12*4
 
 /* VxWorks uses its own plt layout, filled in by the static linker.  */
 
@@ -135,17 +136,24 @@ static const bfd_vma ppc_elf_vxworks_pic_plt0_entry
 #define ADDIS_12_12	0x3d8c0000
 #define ADDI_11_11	0x396b0000
 #define ADD_0_11_11	0x7c0b5a14
+#define ADD_3_12_2	0x7c6c1214
 #define ADD_11_0_11	0x7d605a14
 #define B		0x48000000
 #define BCL_20_31	0x429f0005
 #define BCTR		0x4e800420
+#define BEQLR		0x4d820020
+#define CMPWI_11_0	0x2c0b0000
 #define LIS_11		0x3d600000
 #define LIS_12		0x3d800000
 #define LWZU_0_12	0x840c0000
 #define LWZ_0_12	0x800c0000
+#define LWZ_11_3	0x81630000
 #define LWZ_11_11	0x816b0000
 #define LWZ_11_30	0x817e0000
+#define LWZ_12_3	0x81830000
 #define LWZ_12_12	0x818c0000
+#define MR_0_3		0x7c601b78
+#define MR_3_0		0x7c030378
 #define MFLR_0		0x7c0802a6
 #define MFLR_12		0x7d8802a6
 #define MTCTR_0		0x7c0903a6
@@ -2754,6 +2762,9 @@ struct ppc_elf_link_hash_table
   /* Set if we should emit symbols for stubs.  */
   unsigned int emit_stub_syms:1;
 
+  /* Set if __tls_get_addr optimization should not be done.  */
+  unsigned int no_tls_get_addr_opt:1;
+
   /* True if the target system is VxWorks.  */
   unsigned int is_vxworks:1;
 
@@ -4277,6 +4288,8 @@ ppc_elf_select_plt_layout (bfd *output_bfd ATTRIBUTE_UNUSED,
 
   htab = ppc_elf_hash_table (info);
 
+  htab->emit_stub_syms = emit_stub_syms;
+
   if (htab->plt_type == PLT_UNSET)
     {
       if (plt_style == PLT_OLD)
@@ -4310,8 +4323,6 @@ ppc_elf_select_plt_layout (bfd *output_bfd ATTRIBUTE_UNUSED,
   if (htab->plt_type == PLT_OLD && plt_style == PLT_NEW)
     info->callbacks->info (_("Using bss-plt due to %B"), htab->old_bfd);
 
-  htab->emit_stub_syms = emit_stub_syms;
-
   BFD_ASSERT (htab->plt_type != PLT_VXWORKS);
 
   if (htab->plt_type == PLT_NEW)
@@ -4539,11 +4550,62 @@ ppc_elf_gc_sweep_hook (bfd *abfd,
    generic ELF tls_setup function.  */
 
 asection *
-ppc_elf_tls_setup (bfd *obfd, struct bfd_link_info *info)
+ppc_elf_tls_setup (bfd *obfd,
+		   struct bfd_link_info *info,
+		   int no_tls_get_addr_opt)
 {
   struct ppc_elf_link_hash_table *htab;
 
   htab = ppc_elf_hash_table (info);
+  htab->tls_get_addr = elf_link_hash_lookup (&htab->elf, "__tls_get_addr",
+					     FALSE, FALSE, TRUE);
+  if (!no_tls_get_addr_opt)
+    {
+      struct elf_link_hash_entry *opt, *tga;
+      opt = elf_link_hash_lookup (&htab->elf, "__tls_get_addr_opt",
+				  FALSE, FALSE, TRUE);
+      if (opt != NULL
+	  && (opt->root.type == bfd_link_hash_defined
+	      || opt->root.type == bfd_link_hash_defweak))
+	{
+	  /* If glibc supports an optimized __tls_get_addr call stub,
+	     signalled by the presence of __tls_get_addr_opt, and we'll
+	     be calling __tls_get_addr via a plt call stub, then
+	     make __tls_get_addr point to __tls_get_addr_opt.  */
+	  tga = htab->tls_get_addr;
+	  if (htab->elf.dynamic_sections_created
+	      && tga != NULL
+	      && (tga->type == STT_FUNC
+		  || tga->needs_plt)
+	      && !(SYMBOL_CALLS_LOCAL (info, tga)
+		   || (ELF_ST_VISIBILITY (tga->other) != STV_DEFAULT
+		       && tga->root.type == bfd_link_hash_undefweak)))
+	    {
+	      struct plt_entry *ent;
+	      ent = find_plt_ent (&tga->plt.plist, NULL, 0);
+	      if (ent != NULL
+		  && ent->plt.refcount > 0)
+		{
+		  tga->root.type = bfd_link_hash_indirect;
+		  tga->root.u.i.link = &opt->root;
+		  ppc_elf_copy_indirect_symbol (info, opt, tga);
+		  if (opt->dynindx != -1)
+		    {
+		      /* Use __tls_get_addr_opt in dynamic relocations.  */
+		      opt->dynindx = -1;
+		      _bfd_elf_strtab_delref (elf_hash_table (info)->dynstr,
+					      opt->dynstr_index);
+		      if (!bfd_elf_link_record_dynamic_symbol (info, opt))
+			return FALSE;
+		    }
+		  htab->tls_get_addr = opt;
+		}
+	    }
+	}
+      else
+	no_tls_get_addr_opt = TRUE;
+    }
+  htab->no_tls_get_addr_opt = no_tls_get_addr_opt;
   if (htab->plt_type == PLT_NEW
       && htab->plt != NULL
       && htab->plt->output_section != NULL)
@@ -4552,8 +4614,6 @@ ppc_elf_tls_setup (bfd *obfd, struct bfd_link_info *info)
       elf_section_flags (htab->plt->output_section) = SHF_ALLOC + SHF_WRITE;
     }
 
-  htab->tls_get_addr = elf_link_hash_lookup (&htab->elf, "__tls_get_addr",
-					     FALSE, FALSE, TRUE);
   return _bfd_elf_tls_setup (obfd, info);
 }
 
@@ -5144,6 +5204,9 @@ allocate_dynrelocs (struct elf_link_hash_entry *h, void *inf)
 		      {
 			glink_offset = s->size;
 			s->size += GLINK_ENTRY_SIZE;
+			if (h == htab->tls_get_addr
+			    && !htab->no_tls_get_addr_opt)
+			  s->size += TLS_GET_ADDR_GLINK_SIZE - GLINK_ENTRY_SIZE;
 		      }
 		    if (!doneone
 			&& !info->shared
@@ -5820,6 +5883,11 @@ ppc_elf_size_dynamic_sections (bfd *output_bfd ATTRIBUTE_UNUSED,
 	{
 	  if (!add_dynamic_entry (DT_PPC_GOT, 0))
 	    return FALSE;
+	  if (!htab->no_tls_get_addr_opt
+	      && htab->tls_get_addr != NULL
+	      && htab->tls_get_addr->plt.plist != NULL
+	      && !add_dynamic_entry (DT_PPC_TLSOPT, 0))
+	    return FALSE;
 	}
 
       if (relocs)
@@ -6474,18 +6542,16 @@ elf_finish_pointer_linker_section (bfd *input_bfd,
 #define PPC_HA(v) PPC_HI ((v) + 0x8000)
 
 static void
-write_glink_stub (struct plt_entry *ent, asection *plt_sec,
+write_glink_stub (struct plt_entry *ent, asection *plt_sec, unsigned char *p,
 		  struct bfd_link_info *info)
 {
   struct ppc_elf_link_hash_table *htab = ppc_elf_hash_table (info);
   bfd *output_bfd = info->output_bfd;
   bfd_vma plt;
-  unsigned char *p;
 
   plt = ((ent->plt.offset & ~1)
 	 + plt_sec->output_section->vma
 	 + plt_sec->output_offset);
-  p = (unsigned char *) htab->glink->contents + ent->glink_offset;
 
   if (info->shared)
     {
@@ -7045,7 +7111,9 @@ ppc_elf_relocate_section (bfd *output_bfd,
 		}
 	      if (h == NULL && (ent->glink_offset & 1) == 0)
 		{
-		  write_glink_stub (ent, htab->iplt, info);
+		  unsigned char *p = ((unsigned char *) htab->glink->contents
+				      + ent->glink_offset);
+		  write_glink_stub (ent, htab->iplt, p, info);
 		  ent->glink_offset |= 1;
 		}
 
@@ -8274,12 +8342,35 @@ ppc_elf_finish_dynamic_symbol (bfd *output_bfd,
 	    || !htab->elf.dynamic_sections_created
 	    || h->dynindx == -1)
 	  {
+	    unsigned char *p;
 	    asection *splt = htab->plt;
 	    if (!htab->elf.dynamic_sections_created
 		|| h->dynindx == -1)
 	      splt = htab->iplt;
 
-	    write_glink_stub (ent, splt, info);
+	    p = (unsigned char *) htab->glink->contents + ent->glink_offset;
+
+	    if (h == htab->tls_get_addr && !htab->no_tls_get_addr_opt)
+	      {
+		bfd_put_32 (output_bfd, LWZ_11_3, p);
+		p += 4;
+		bfd_put_32 (output_bfd, LWZ_12_3 + 4, p);
+		p += 4;
+		bfd_put_32 (output_bfd, MR_0_3, p);
+		p += 4;
+		bfd_put_32 (output_bfd, CMPWI_11_0, p);
+		p += 4;
+		bfd_put_32 (output_bfd, ADD_3_12_2, p);
+		p += 4;
+		bfd_put_32 (output_bfd, BEQLR, p);
+		p += 4;
+		bfd_put_32 (output_bfd, MR_3_0, p);
+		p += 4;
+		bfd_put_32 (output_bfd, NOP, p);
+		p += 4;
+	      }
+
+	    write_glink_stub (ent, splt, p, info);
 
 	    if (!info->shared)
 	      /* We only need one non-PIC glink stub.  */
diff --git a/bfd/elf32-ppc.h b/bfd/elf32-ppc.h
index ef1f0ac..4becb30 100644
--- a/bfd/elf32-ppc.h
+++ b/bfd/elf32-ppc.h
@@ -28,6 +28,6 @@ enum ppc_elf_plt_type
 
 int ppc_elf_select_plt_layout (bfd *, struct bfd_link_info *,
 			       enum ppc_elf_plt_type, int);
-asection *ppc_elf_tls_setup (bfd *, struct bfd_link_info *);
+asection *ppc_elf_tls_setup (bfd *, struct bfd_link_info *, int);
 bfd_boolean ppc_elf_tls_optimize (bfd *, struct bfd_link_info *);
 void ppc_elf_set_sdata_syms (bfd *, struct bfd_link_info *);
diff --git a/bfd/elf64-ppc.c b/bfd/elf64-ppc.c
index 99bfec2..35757b8 100644
--- a/bfd/elf64-ppc.c
+++ b/bfd/elf64-ppc.c
@@ -3771,6 +3771,9 @@ struct ppc_link_hash_table
   /* Set if we should emit symbols for stubs.  */
   unsigned int emit_stub_syms:1;
 
+  /* Set if __tls_get_addr optimization should not be done.  */
+  unsigned int no_tls_get_addr_opt:1;
+
   /* Support for multiple toc sections.  */
   unsigned int no_multi_toc:1;
   unsigned int multi_toc_needed:1;
@@ -7218,7 +7221,9 @@ ppc64_elf_edit_opd (bfd *obfd, struct bfd_link_info *info,
 /* Set htab->tls_get_addr and call the generic ELF tls_setup function.  */
 
 asection *
-ppc64_elf_tls_setup (bfd *obfd, struct bfd_link_info *info)
+ppc64_elf_tls_setup (bfd *obfd,
+		     struct bfd_link_info *info,
+		     int no_tls_get_addr_opt)
 {
   struct ppc_link_hash_table *htab;
 
@@ -7226,9 +7231,83 @@ ppc64_elf_tls_setup (bfd *obfd, struct bfd_link_info *info)
   htab->tls_get_addr = ((struct ppc_link_hash_entry *)
 			elf_link_hash_lookup (&htab->elf, ".__tls_get_addr",
 					      FALSE, FALSE, TRUE));
+  /* Move dynamic linking info to the function descriptor sym.  */
+  if (htab->tls_get_addr != NULL)
+    func_desc_adjust (&htab->tls_get_addr->elf, info);
   htab->tls_get_addr_fd = ((struct ppc_link_hash_entry *)
 			   elf_link_hash_lookup (&htab->elf, "__tls_get_addr",
 						 FALSE, FALSE, TRUE));
+  if (!no_tls_get_addr_opt)
+    {
+      struct elf_link_hash_entry *opt, *opt_fd, *tga, *tga_fd;
+
+      opt = elf_link_hash_lookup (&htab->elf, ".__tls_get_addr_opt",
+				  FALSE, FALSE, TRUE);
+      if (opt != NULL)
+	func_desc_adjust (opt, info);
+      opt_fd = elf_link_hash_lookup (&htab->elf, "__tls_get_addr_opt",
+				     FALSE, FALSE, TRUE);
+      if (opt_fd != NULL
+	  && (opt_fd->root.type == bfd_link_hash_defined
+	      || opt_fd->root.type == bfd_link_hash_defweak))
+	{
+	  /* If glibc supports an optimized __tls_get_addr call stub,
+	     signalled by the presence of __tls_get_addr_opt, and we'll
+	     be calling __tls_get_addr via a plt call stub, then
+	     make __tls_get_addr point to __tls_get_addr_opt.  */
+	  tga_fd = &htab->tls_get_addr_fd->elf;
+	  if (htab->elf.dynamic_sections_created
+	      && tga_fd != NULL
+	      && (tga_fd->type == STT_FUNC
+		  || tga_fd->needs_plt)
+	      && !(SYMBOL_CALLS_LOCAL (info, tga_fd)
+		   || (ELF_ST_VISIBILITY (tga_fd->other) != STV_DEFAULT
+		       && tga_fd->root.type == bfd_link_hash_undefweak)))
+	    {
+	      struct plt_entry *ent;
+
+	      for (ent = tga_fd->plt.plist; ent != NULL; ent = ent->next)
+		if (ent->plt.refcount > 0)
+		  break;
+	      if (ent != NULL)
+		{
+		  tga_fd->root.type = bfd_link_hash_indirect;
+		  tga_fd->root.u.i.link = &opt_fd->root;
+		  ppc64_elf_copy_indirect_symbol (info, opt_fd, tga_fd);
+		  if (opt_fd->dynindx != -1)
+		    {
+		      /* Use __tls_get_addr_opt in dynamic relocations.  */
+		      opt_fd->dynindx = -1;
+		      _bfd_elf_strtab_delref (elf_hash_table (info)->dynstr,
+					      opt_fd->dynstr_index);
+		      if (!bfd_elf_link_record_dynamic_symbol (info, opt_fd))
+			return FALSE;
+		    }
+		  htab->tls_get_addr_fd = (struct ppc_link_hash_entry *) opt_fd;
+		  tga = &htab->tls_get_addr->elf;
+		  if (opt != NULL && tga != NULL)
+		    {
+		      tga->root.type = bfd_link_hash_indirect;
+		      tga->root.u.i.link = &opt->root;
+		      ppc64_elf_copy_indirect_symbol (info, opt, tga);
+		      _bfd_elf_link_hash_hide_symbol (info, opt,
+						      tga->forced_local);
+		      htab->tls_get_addr = (struct ppc_link_hash_entry *) opt;
+		    }
+		  htab->tls_get_addr_fd->oh = htab->tls_get_addr;
+		  htab->tls_get_addr_fd->is_func_descriptor = 1;
+		  if (htab->tls_get_addr != NULL)
+		    {
+		      htab->tls_get_addr->oh = htab->tls_get_addr_fd;
+		      htab->tls_get_addr->is_func = 1;
+		    }
+		}
+	    }
+	}
+      else
+	no_tls_get_addr_opt = TRUE;
+    }
+  htab->no_tls_get_addr_opt = no_tls_get_addr_opt;
   return _bfd_elf_tls_setup (obfd, info);
 }
 
@@ -8666,6 +8745,12 @@ ppc64_elf_size_dynamic_sections (bfd *output_bfd ATTRIBUTE_UNUSED,
 	    return FALSE;
 	}
 
+      if (!htab->no_tls_get_addr_opt
+	  && htab->tls_get_addr_fd != NULL
+	  && htab->tls_get_addr_fd->elf.plt.plist != NULL
+	  && !add_dynamic_entry (DT_PPC64_TLSOPT, 0))
+	return FALSE;
+
       if (relocs)
 	{
 	  if (!add_dynamic_entry (DT_RELA, 0)
@@ -8859,6 +8944,49 @@ build_plt_stub (bfd *obfd, bfd_byte *p, int offset, Elf_Internal_Rela *r)
   return p;
 }
 
+/* Build a special .plt call stub for __tls_get_addr.  */
+
+#define LD_R11_0R3	0xe9630000
+#define LD_R12_0R3	0xe9830000
+#define MR_R0_R3	0x7c601b78
+#define CMPDI_R11_0	0x2c2b0000
+#define ADD_R3_R12_R13	0x7c6c6a14
+#define BEQLR		0x4d820020
+#define MR_R3_R0	0x7c030378
+#define MFLR_R11	0x7d6802a6
+#define STD_R11_0R1	0xf9610000
+#define BCTRL		0x4e800421
+#define LD_R11_0R1	0xe9610000
+#define LD_R2_0R1	0xe8410000
+#define MTLR_R11	0x7d6803a6
+
+static inline bfd_byte *
+build_tls_get_addr_stub (bfd *obfd, bfd_byte *p, int offset,
+			 Elf_Internal_Rela *r)
+{
+  bfd_put_32 (obfd, LD_R11_0R3 + 0, p),		p += 4;
+  bfd_put_32 (obfd, LD_R12_0R3 + 8, p),		p += 4;
+  bfd_put_32 (obfd, MR_R0_R3, p),		p += 4;
+  bfd_put_32 (obfd, CMPDI_R11_0, p),		p += 4;
+  bfd_put_32 (obfd, ADD_R3_R12_R13, p),		p += 4;
+  bfd_put_32 (obfd, BEQLR, p),			p += 4;
+  bfd_put_32 (obfd, MR_R3_R0, p),		p += 4;
+  bfd_put_32 (obfd, MFLR_R11, p),		p += 4;
+  bfd_put_32 (obfd, STD_R11_0R1 + 32, p),	p += 4;
+
+  if (r != NULL)
+    r[0].r_offset += 9 * 4;
+  p = build_plt_stub (obfd, p, offset, r);
+  bfd_put_32 (obfd, BCTRL, p - 4);
+
+  bfd_put_32 (obfd, LD_R11_0R1 + 32, p),	p += 4;
+  bfd_put_32 (obfd, LD_R2_0R1 + 40, p),		p += 4;
+  bfd_put_32 (obfd, MTLR_R11, p),		p += 4;
+  bfd_put_32 (obfd, BLR, p),			p += 4;
+
+  return p;
+}
+
 static Elf_Internal_Rela *
 get_relocs (asection *sec, int count)
 {
@@ -9227,7 +9355,13 @@ ppc_build_one_stub (struct bfd_hash_entry *gen_entry, void *in_arg)
 	    r[0].r_offset += 2;
 	  r[0].r_addend = dest;
 	}
-      p = build_plt_stub (htab->stub_bfd, loc, off, r);
+      if (stub_entry->h != NULL
+	  && (stub_entry->h == htab->tls_get_addr_fd
+	      || stub_entry->h == htab->tls_get_addr)
+	  && !htab->no_tls_get_addr_opt)
+	p = build_tls_get_addr_stub (htab->stub_bfd, loc, off, r);
+      else
+	p = build_plt_stub (htab->stub_bfd, loc, off, r);
       size = p - loc;
       break;
 
@@ -9316,6 +9450,11 @@ ppc_size_one_stub (struct bfd_hash_entry *gen_entry, void *in_arg)
 	size -= 4;
       if (PPC_HA (off + 16) != PPC_HA (off))
 	size += 4;
+      if (stub_entry->h != NULL
+	  && (stub_entry->h == htab->tls_get_addr_fd
+	      || stub_entry->h == htab->tls_get_addr)
+	  && !htab->no_tls_get_addr_opt)
+	size += 13 * 4;
       if (info->emitrelocations)
 	{
 	  stub_entry->stub_sec->reloc_count
@@ -11255,8 +11394,16 @@ ppc64_elf_relocate_section (bfd *output_bfd,
 		  if (nop == NOP
 		      || nop == CROR_151515 || nop == CROR_313131)
 		    {
-		      bfd_put_32 (input_bfd, LD_R2_40R1,
-				  contents + rel->r_offset + 4);
+		      if (h != NULL
+			  && (h == htab->tls_get_addr_fd
+			      || h == htab->tls_get_addr)
+			  && !htab->no_tls_get_addr_opt)
+			{
+			  /* Special stub used, leave nop alone.  */
+			}
+		      else
+			bfd_put_32 (input_bfd, LD_R2_40R1,
+				    contents + rel->r_offset + 4);
 		      can_plt_call = TRUE;
 		    }
 		}
diff --git a/bfd/elf64-ppc.h b/bfd/elf64-ppc.h
index 687b4a1..e5f7140 100644
--- a/bfd/elf64-ppc.h
+++ b/bfd/elf64-ppc.h
@@ -24,7 +24,7 @@ void ppc64_elf_init_stub_bfd
 bfd_boolean ppc64_elf_edit_opd
   (bfd *, struct bfd_link_info *, bfd_boolean);
 asection *ppc64_elf_tls_setup
-  (bfd *, struct bfd_link_info *);
+(bfd *, struct bfd_link_info *, int);
 bfd_boolean ppc64_elf_tls_optimize
   (bfd *, struct bfd_link_info *);
 bfd_boolean ppc64_elf_edit_toc