* configure.ac (ia64*-*-*vms*): Add support for ld.
authorNick Clifton <nickc@redhat.com>
Thu, 12 Apr 2012 07:35:07 +0000 (07:35 +0000)
committerNick Clifton <nickc@redhat.com>
Thu, 12 Apr 2012 07:35:07 +0000 (07:35 +0000)
* configure: Regenerate.

* scripttempl/ia64vms.sc: New file.
* emultempl/vms.em (_before_parse): Support for ia64.
(elf64-ia64-vms): New fragment for ia64.
* emulparams/elf64_ia64_vms.sh: New file.
* configure.tgt (ia64-*-*vms*): Add.
* Makefile.am (ALL_64_EMULATION_SOURCES): Add eelf64_ia64_vms.c
(eelf64_ia64_vms.c): New rule.
* Makefile.in: Regenerate.

* elflink.c (elf_link_output_extsym): Add a guard.
(bfd_elf_final_link): Remove assertion.
(bfd_elf_final_link): Add a guard.
* elfnn-ia64.c (INCLUDE_IA64_VMS): Removed.
(elfNN_vms_section_from_shdr, elfNN_vms_object_p)
(elfNN_vms_post_process_headers, elfNN_vms_section_processing)
(elfNN_vms_final_write_processing, elfNN_vms_close_and_cleanup):
Remove.
(elfNN-ia64-vms target): Move to ...
* elf64-ia64-vms.c: New file.
* configure.in (bfd_elf64_ia64_vms_vec): Add elf64-ia64-vms.lo
* Makefile.am (BFD64_BACKENDS): Add elf64-ia64-vms.lo.
(BFD64_BACKENDS_CFILES): Ad elf64-ia64-vms.c.
* configure: Regenerate.
* Makefile.in: Regenerate.

18 files changed:
ChangeLog
bfd/ChangeLog
bfd/Makefile.am
bfd/Makefile.in
bfd/configure
bfd/configure.in
bfd/elf64-ia64-vms.c [new file with mode: 0644]
bfd/elflink.c
bfd/elfnn-ia64.c
configure
configure.ac
ld/ChangeLog
ld/Makefile.am
ld/Makefile.in
ld/configure.tgt
ld/emulparams/elf64_ia64_vms.sh [new file with mode: 0644]
ld/emultempl/vms.em
ld/scripttempl/ia64vms.sc [new file with mode: 0644]

index cff1dfb1e61cbf75e10c2a1f0de2218c471cabf7..c4b159fd5cb213fb0f50355c6184ec5409759e00 100644 (file)
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,8 @@
+2012-03-19  Tristan Gingold  <gingold@adacore.com>
+
+       * configure.ac (ia64*-*-*vms*): Add support for ld.
+       * configure: Regenerate.
+
 2012-03-14  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
 
        * configure.ac (enable_libgomp): Remove *-*-irix6*.
index 74f8c3a355922113abaa3e3773742c3f0f73b0be..f84e7584b018e2ee5201c36119b0d08735be6cad 100644 (file)
@@ -1,3 +1,21 @@
+2012-04-12  Tristan Gingold  <gingold@adacore.com>
+
+       * elflink.c (elf_link_output_extsym): Add a guard.
+       (bfd_elf_final_link): Remove assertion.
+       (bfd_elf_final_link): Add a guard.
+       * elfnn-ia64.c (INCLUDE_IA64_VMS): Removed.
+       (elfNN_vms_section_from_shdr, elfNN_vms_object_p)
+       (elfNN_vms_post_process_headers, elfNN_vms_section_processing)
+       (elfNN_vms_final_write_processing, elfNN_vms_close_and_cleanup):
+       Remove.
+       (elfNN-ia64-vms target): Move to ...
+       * elf64-ia64-vms.c: New file.
+       * configure.in (bfd_elf64_ia64_vms_vec): Add elf64-ia64-vms.lo
+       * Makefile.am (BFD64_BACKENDS): Add elf64-ia64-vms.lo.
+       (BFD64_BACKENDS_CFILES): Ad elf64-ia64-vms.c.
+       * configure: Regenerate.
+       * Makefile.in: Regenerate.
+
 2012-04-11  Nick Clifton  <nickc@redhat.com>
 
        PR binutils/13897
index 50c84ee0ea5f724692f54e9c34532d06af5c18db..627f1bb8dcc118496efb337e2bba9ceb1827f8bd 100644 (file)
@@ -619,6 +619,7 @@ BFD64_BACKENDS = \
        elf64-gen.lo \
        elf64-hppa.lo \
        elf64-ia64.lo \
+       elf64-ia64-vms.lo \
        elf64-mips.lo \
        elf64-mmix.lo \
        elf64-ppc.lo \
@@ -654,6 +655,7 @@ BFD64_BACKENDS_CFILES = \
        elf64-alpha.c \
        elf64-gen.c \
        elf64-hppa.c \
+       elf64-ia64-vms.c \
        elf64-mips.c \
        elf64-mmix.c \
        elf64-ppc.c \
index cf7b3807917d0de8657d3950a5f98d8771fec1da..6d9c760cee8e7f58b4735be71968538aac808387 100644 (file)
@@ -920,6 +920,7 @@ BFD64_BACKENDS = \
        elf64-gen.lo \
        elf64-hppa.lo \
        elf64-ia64.lo \
+       elf64-ia64-vms.lo \
        elf64-mips.lo \
        elf64-mmix.lo \
        elf64-ppc.lo \
@@ -955,6 +956,7 @@ BFD64_BACKENDS_CFILES = \
        elf64-alpha.c \
        elf64-gen.c \
        elf64-hppa.c \
+       elf64-ia64-vms.c \
        elf64-mips.c \
        elf64-mmix.c \
        elf64-ppc.c \
@@ -1416,6 +1418,7 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/elf64-alpha.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/elf64-gen.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/elf64-hppa.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/elf64-ia64-vms.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/elf64-ia64.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/elf64-mips.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/elf64-mmix.Plo@am__quote@
index dd274bd8868ffdfcd3fd6ce1a87f1311321596ab..d90e52f2306bd1b978433d9d3fd8585d192dfa1e 100755 (executable)
@@ -15341,7 +15341,7 @@ do
     bfd_elf64_ia64_big_vec)    tb="$tb elf64-ia64.lo elfxx-ia64.lo elf64.lo $elf"; target_size=64 ;;
     bfd_elf64_ia64_hpux_big_vec) tb="$tb elf64-ia64.lo elfxx-ia64.lo elf64.lo $elf"; target_size=64 ;;
     bfd_elf64_ia64_little_vec) tb="$tb elf64-ia64.lo elfxx-ia64.lo elf64.lo $elf"; target_size=64 ;;
-    bfd_elf64_ia64_vms_vec)    tb="$tb elf64-ia64.lo elfxx-ia64.lo  elf64.lo vms-lib.lo vms-misc.lo $elf"; target_size=64 ;;
+    bfd_elf64_ia64_vms_vec)    tb="$tb elf64-ia64-vms.lo elf64-ia64.lo elfxx-ia64.lo elf64.lo vms-lib.lo vms-misc.lo $elf"; target_size=64 ;;
     bfd_elf64_little_generic_vec) tb="$tb elf64-gen.lo elf64.lo $elf"; target_size=64 ;;
     bfd_elf64_littlemips_vec)  tb="$tb elf64-mips.lo elf64.lo elfxx-mips.lo elf-vxworks.lo elf32.lo $elf ecofflink.lo"; target_size=64 ;;
     bfd_elf64_mmix_vec)        tb="$tb elf64-mmix.lo elf64.lo $elf" target_size=64 ;;
index af3622b426f051c3960828b6b7e89c4bda357c41..37cc769290ade38117ebb1896a3c4b3f685c2937 100644 (file)
@@ -827,7 +827,7 @@ do
     bfd_elf64_ia64_big_vec)    tb="$tb elf64-ia64.lo elfxx-ia64.lo elf64.lo $elf"; target_size=64 ;;
     bfd_elf64_ia64_hpux_big_vec) tb="$tb elf64-ia64.lo elfxx-ia64.lo elf64.lo $elf"; target_size=64 ;;
     bfd_elf64_ia64_little_vec) tb="$tb elf64-ia64.lo elfxx-ia64.lo elf64.lo $elf"; target_size=64 ;;
-    bfd_elf64_ia64_vms_vec)    tb="$tb elf64-ia64.lo elfxx-ia64.lo  elf64.lo vms-lib.lo vms-misc.lo $elf"; target_size=64 ;;
+    bfd_elf64_ia64_vms_vec)    tb="$tb elf64-ia64-vms.lo elf64-ia64.lo elfxx-ia64.lo elf64.lo vms-lib.lo vms-misc.lo $elf"; target_size=64 ;;    
     bfd_elf64_little_generic_vec) tb="$tb elf64-gen.lo elf64.lo $elf"; target_size=64 ;;
     bfd_elf64_littlemips_vec)  tb="$tb elf64-mips.lo elf64.lo elfxx-mips.lo elf-vxworks.lo elf32.lo $elf ecofflink.lo"; target_size=64 ;;
     bfd_elf64_mmix_vec)        tb="$tb elf64-mmix.lo elf64.lo $elf" target_size=64 ;;
diff --git a/bfd/elf64-ia64-vms.c b/bfd/elf64-ia64-vms.c
new file mode 100644 (file)
index 0000000..1ab8e38
--- /dev/null
@@ -0,0 +1,5601 @@
+/* IA-64 support for OpenVMS
+   Copyright 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007,
+   2008, 2009, 2010, 2012  Free Software Foundation, Inc.
+
+   This file is part of BFD, the Binary File Descriptor library.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
+   MA 02110-1301, USA.  */
+
+#include "sysdep.h"
+#include "bfd.h"
+#include "libbfd.h"
+#include "elf-bfd.h"
+#include "opcode/ia64.h"
+#include "elf/ia64.h"
+#include "objalloc.h"
+#include "hashtab.h"
+#include "elfxx-ia64.h"
+#include "vms.h"
+#include "bfdver.h"
+
+/* THE RULES for all the stuff the linker creates --
+
+  GOT          Entries created in response to LTOFF or LTOFF_FPTR
+               relocations.  Dynamic relocs created for dynamic
+               symbols in an application; REL relocs for locals
+               in a shared library.
+
+  FPTR         The canonical function descriptor.  Created for local
+               symbols in applications.  Descriptors for dynamic symbols
+               and local symbols in shared libraries are created by
+               ld.so.  Thus there are no dynamic relocs against these
+               objects.  The FPTR relocs for such _are_ passed through
+               to the dynamic relocation tables.
+
+  FULL_PLT     Created for a PCREL21B relocation against a dynamic symbol.
+               Requires the creation of a PLTOFF entry.  This does not
+               require any dynamic relocations.
+
+  PLTOFF       Created by PLTOFF relocations.  For local symbols, this
+               is an alternate function descriptor, and in shared libraries
+               requires two REL relocations.  Note that this cannot be
+               transformed into an FPTR relocation, since it must be in
+               range of the GP.  For dynamic symbols, this is a function
+               descriptor.  */
+
+typedef struct bfd_hash_entry *(*new_hash_entry_func)
+  (struct bfd_hash_entry *, struct bfd_hash_table *, const char *);
+
+/* In dynamically (linker-) created sections, we generally need to keep track
+   of the place a symbol or expression got allocated to. This is done via hash
+   tables that store entries of the following type.  */
+
+struct elf64_ia64_dyn_sym_info
+{
+  /* The addend for which this entry is relevant.  */
+  bfd_vma addend;
+
+  bfd_vma got_offset;
+  bfd_vma fptr_offset;
+  bfd_vma pltoff_offset;
+  bfd_vma plt_offset;
+  bfd_vma plt2_offset;
+
+  /* The symbol table entry, if any, that this was derived from.  */
+  struct elf_link_hash_entry *h;
+
+  /* Used to count non-got, non-plt relocations for delayed sizing
+     of relocation sections.  */
+  struct elf64_ia64_dyn_reloc_entry
+  {
+    struct elf64_ia64_dyn_reloc_entry *next;
+    asection *srel;
+    int type;
+    int count;
+  } *reloc_entries;
+
+  /* TRUE when the section contents have been updated.  */
+  unsigned got_done : 1;
+  unsigned fptr_done : 1;
+  unsigned pltoff_done : 1;
+
+  /* TRUE for the different kinds of linker data we want created.  */
+  unsigned want_got : 1;
+  unsigned want_gotx : 1;
+  unsigned want_fptr : 1;
+  unsigned want_ltoff_fptr : 1;
+  unsigned want_plt : 1;       /* A MIN_PLT entry.  */
+  unsigned want_plt2 : 1;      /* A FULL_PLT.  */
+  unsigned want_pltoff : 1;
+};
+
+struct elf64_ia64_local_hash_entry
+{
+  int id;
+  unsigned int r_sym;
+  /* The number of elements in elf64_ia64_dyn_sym_info array.  */
+  unsigned int count;
+  /* The number of sorted elements in elf64_ia64_dyn_sym_info array.  */
+  unsigned int sorted_count;
+  /* The size of elf64_ia64_dyn_sym_info array.  */
+  unsigned int size;
+  /* The array of elf64_ia64_dyn_sym_info.  */
+  struct elf64_ia64_dyn_sym_info *info;
+
+  /* TRUE if this hash entry's addends was translated for
+     SHF_MERGE optimization.  */
+  unsigned sec_merge_done : 1;
+};
+
+struct elf64_ia64_link_hash_entry
+{
+  struct elf_link_hash_entry root;
+
+  /* Set if this symbol is defined in a shared library.
+     We can't use root.u.def.section->owner as the symbol is an absolute
+     symbol.  */
+  bfd *shl;
+
+  /* The number of elements in elf64_ia64_dyn_sym_info array.  */
+  unsigned int count;
+  /* The number of sorted elements in elf64_ia64_dyn_sym_info array.  */
+  unsigned int sorted_count;
+  /* The size of elf64_ia64_dyn_sym_info array.  */
+  unsigned int size;
+  /* The array of elf64_ia64_dyn_sym_info.  */
+  struct elf64_ia64_dyn_sym_info *info;
+};
+
+struct elf64_ia64_link_hash_table
+{
+  /* The main hash table.  */
+  struct elf_link_hash_table root;
+
+  asection *fptr_sec;          /* Function descriptor table (or NULL).  */
+  asection *rel_fptr_sec;      /* Dynamic relocation section for same.  */
+  asection *pltoff_sec;                /* Private descriptors for plt (or NULL).  */
+  asection *fixups_sec;                /* Fixups section.  */
+  asection *transfer_sec;      /* Transfer vector section.  */
+  asection *note_sec;          /* .note section.  */
+
+  /* There are maybe R_IA64_GPREL22 relocations, including those
+     optimized from R_IA64_LTOFF22X, against non-SHF_IA_64_SHORT
+     sections.  We need to record those sections so that we can choose
+     a proper GP to cover all R_IA64_GPREL22 relocations.  */
+  asection *max_short_sec;     /* Maximum short output section.  */
+  bfd_vma max_short_offset;    /* Maximum short offset.  */
+  asection *min_short_sec;     /* Minimum short output section.  */
+  bfd_vma min_short_offset;    /* Minimum short offset.  */
+
+  htab_t loc_hash_table;
+  void *loc_hash_memory;
+};
+
+struct elf64_ia64_allocate_data
+{
+  struct bfd_link_info *info;
+  bfd_size_type ofs;
+};
+
+#define elf64_ia64_hash_table(p) \
+  (elf_hash_table_id ((struct elf_link_hash_table *) ((p)->hash)) \
+  == IA64_ELF_DATA ? ((struct elf64_ia64_link_hash_table *) ((p)->hash)) : NULL)
+
+struct elf64_ia64_vms_obj_tdata
+{
+  struct elf_obj_tdata root;
+
+  /* Ident for shared library.  */
+  bfd_uint64_t ident;
+
+  /* Used only during link: offset in the .fixups section for this bfd.  */
+  bfd_vma fixups_off;
+
+  /* Max number of shared libraries.  */
+  unsigned int needed_count;
+};
+
+#define elf_ia64_vms_tdata(abfd) \
+  ((struct elf64_ia64_vms_obj_tdata *)((abfd)->tdata.any))
+#define elf_ia64_vms_ident(abfd) (elf_ia64_vms_tdata(abfd)->ident)
+
+struct elf64_vms_transfer
+{
+  unsigned char size[4];
+  unsigned char spare[4];
+  unsigned char tfradr1[8];
+  unsigned char tfradr2[8];
+  unsigned char tfradr3[8];
+  unsigned char tfradr4[8];
+  unsigned char tfradr5[8];
+
+  /* Local function descriptor for tfr3.  */
+  unsigned char tfr3_func[8];
+  unsigned char tfr3_gp[8];
+};
+
+typedef struct
+{
+  Elf64_External_Ehdr ehdr;
+  unsigned char vms_needed_count[8];
+} Elf64_External_VMS_Ehdr;
+
+static struct elf64_ia64_dyn_sym_info * get_dyn_sym_info
+  (struct elf64_ia64_link_hash_table *,
+   struct elf_link_hash_entry *,
+   bfd *, const Elf_Internal_Rela *, bfd_boolean);
+static bfd_boolean elf64_ia64_dynamic_symbol_p
+  (struct elf_link_hash_entry *);
+static bfd_boolean elf64_ia64_choose_gp
+  (bfd *, struct bfd_link_info *, bfd_boolean);
+static void elf64_ia64_dyn_sym_traverse
+  (struct elf64_ia64_link_hash_table *,
+   bfd_boolean (*) (struct elf64_ia64_dyn_sym_info *, void *),
+   void *);
+static bfd_boolean allocate_global_data_got
+  (struct elf64_ia64_dyn_sym_info *, void *);
+static bfd_boolean allocate_global_fptr_got
+  (struct elf64_ia64_dyn_sym_info *, void *);
+static bfd_boolean allocate_local_got
+  (struct elf64_ia64_dyn_sym_info *, void *);
+static bfd_boolean allocate_dynrel_entries
+  (struct elf64_ia64_dyn_sym_info *, void *);
+static asection *get_pltoff
+  (bfd *, struct elf64_ia64_link_hash_table *);
+static asection *get_got
+  (bfd *, struct elf64_ia64_link_hash_table *);
+
+
+/* Given a ELF reloc, return the matching HOWTO structure.  */
+
+static void
+elf64_ia64_info_to_howto (bfd *abfd ATTRIBUTE_UNUSED,
+                         arelent *bfd_reloc,
+                         Elf_Internal_Rela *elf_reloc)
+{
+  bfd_reloc->howto
+    = ia64_elf_lookup_howto ((unsigned int) ELF64_R_TYPE (elf_reloc->r_info));
+}
+
+
+#define PLT_FULL_ENTRY_SIZE    (2 * 16)
+
+static const bfd_byte plt_full_entry[PLT_FULL_ENTRY_SIZE] =
+{
+  0x0b, 0x78, 0x00, 0x02, 0x00, 0x24,  /*   [MMI]       addl r15=0,r1;;   */
+  0x00, 0x41, 0x3c, 0x70, 0x29, 0xc0,  /*               ld8.acq r16=[r15],8*/
+  0x01, 0x08, 0x00, 0x84,              /*               mov r14=r1;;      */
+  0x11, 0x08, 0x00, 0x1e, 0x18, 0x10,  /*   [MIB]       ld8 r1=[r15]      */
+  0x60, 0x80, 0x04, 0x80, 0x03, 0x00,  /*               mov b6=r16        */
+  0x60, 0x00, 0x80, 0x00               /*               br.few b6;;       */
+};
+
+static const bfd_byte oor_brl[16] =
+{
+  0x05, 0x00, 0x00, 0x00, 0x01, 0x00,  /*  [MLX]        nop.m 0           */
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00,  /*               brl.sptk.few tgt;;*/
+  0x00, 0x00, 0x00, 0xc0
+};
+
+
+/* These functions do relaxation for IA-64 ELF.  */
+
+/* Rename some of the generic section flags to better document how they
+   are used here.  */
+#define skip_relax_pass_0 sec_flg0
+#define skip_relax_pass_1 sec_flg1
+
+static void
+elf64_ia64_update_short_info (asection *sec, bfd_vma offset,
+                             struct elf64_ia64_link_hash_table *ia64_info)
+{
+  /* Skip ABS and SHF_IA_64_SHORT sections.  */
+  if (sec == bfd_abs_section_ptr
+      || (sec->flags & SEC_SMALL_DATA) != 0)
+    return;
+
+  if (!ia64_info->min_short_sec)
+    {
+      ia64_info->max_short_sec = sec;
+      ia64_info->max_short_offset = offset;
+      ia64_info->min_short_sec = sec;
+      ia64_info->min_short_offset = offset;
+    }
+  else if (sec == ia64_info->max_short_sec
+          && offset > ia64_info->max_short_offset)
+    ia64_info->max_short_offset = offset;
+  else if (sec == ia64_info->min_short_sec
+          && offset < ia64_info->min_short_offset)
+    ia64_info->min_short_offset = offset;
+  else if (sec->output_section->vma
+          > ia64_info->max_short_sec->vma)
+    {
+      ia64_info->max_short_sec = sec;
+      ia64_info->max_short_offset = offset;
+    }
+  else if (sec->output_section->vma
+          < ia64_info->min_short_sec->vma)
+    {
+      ia64_info->min_short_sec = sec;
+      ia64_info->min_short_offset = offset;
+    }
+}
+
+/* Use a two passes algorithm.  In the first pass, branches are relaxed
+   (which may increase the size of the section).  In the second pass,
+   the other relaxations are done.
+*/
+
+static bfd_boolean
+elf64_ia64_relax_section (bfd *abfd, asection *sec,
+                         struct bfd_link_info *link_info,
+                         bfd_boolean *again)
+{
+  struct one_fixup
+    {
+      struct one_fixup *next;
+      asection *tsec;
+      bfd_vma toff;
+      bfd_vma trampoff;
+    };
+
+  Elf_Internal_Shdr *symtab_hdr;
+  Elf_Internal_Rela *internal_relocs;
+  Elf_Internal_Rela *irel, *irelend;
+  bfd_byte *contents;
+  Elf_Internal_Sym *isymbuf = NULL;
+  struct elf64_ia64_link_hash_table *ia64_info;
+  struct one_fixup *fixups = NULL;
+  bfd_boolean changed_contents = FALSE;
+  bfd_boolean changed_relocs = FALSE;
+  bfd_boolean skip_relax_pass_0 = TRUE;
+  bfd_boolean skip_relax_pass_1 = TRUE;
+  bfd_vma gp = 0;
+
+  /* Assume we're not going to change any sizes, and we'll only need
+     one pass.  */
+  *again = FALSE;
+
+  if (link_info->relocatable)
+    (*link_info->callbacks->einfo)
+      (_("%P%F: --relax and -r may not be used together\n"));
+
+  /* Don't even try to relax for non-ELF outputs.  */
+  if (!is_elf_hash_table (link_info->hash))
+    return FALSE;
+
+  /* Nothing to do if there are no relocations or there is no need for
+     the current pass.  */
+  if ((sec->flags & SEC_RELOC) == 0
+      || sec->reloc_count == 0
+      || (link_info->relax_pass == 0 && sec->skip_relax_pass_0)
+      || (link_info->relax_pass == 1 && sec->skip_relax_pass_1))
+    return TRUE;
+
+  ia64_info = elf64_ia64_hash_table (link_info);
+  if (ia64_info == NULL)
+    return FALSE;
+
+  symtab_hdr = &elf_tdata (abfd)->symtab_hdr;
+
+  /* Load the relocations for this section.  */
+  internal_relocs = (_bfd_elf_link_read_relocs
+                    (abfd, sec, NULL, (Elf_Internal_Rela *) NULL,
+                     link_info->keep_memory));
+  if (internal_relocs == NULL)
+    return FALSE;
+
+  irelend = internal_relocs + sec->reloc_count;
+
+  /* Get the section contents.  */
+  if (elf_section_data (sec)->this_hdr.contents != NULL)
+    contents = elf_section_data (sec)->this_hdr.contents;
+  else
+    {
+      if (!bfd_malloc_and_get_section (abfd, sec, &contents))
+       goto error_return;
+    }
+
+  for (irel = internal_relocs; irel < irelend; irel++)
+    {
+      unsigned long r_type = ELF64_R_TYPE (irel->r_info);
+      bfd_vma symaddr, reladdr, trampoff, toff, roff;
+      asection *tsec;
+      struct one_fixup *f;
+      bfd_size_type amt;
+      bfd_boolean is_branch;
+      struct elf64_ia64_dyn_sym_info *dyn_i;
+
+      switch (r_type)
+       {
+       case R_IA64_PCREL21B:
+       case R_IA64_PCREL21BI:
+       case R_IA64_PCREL21M:
+       case R_IA64_PCREL21F:
+         /* In pass 1, all br relaxations are done. We can skip it. */
+         if (link_info->relax_pass == 1)
+           continue;
+         skip_relax_pass_0 = FALSE;
+         is_branch = TRUE;
+         break;
+
+       case R_IA64_PCREL60B:
+         /* We can't optimize brl to br in pass 0 since br relaxations
+            will increase the code size. Defer it to pass 1.  */
+         if (link_info->relax_pass == 0)
+           {
+             skip_relax_pass_1 = FALSE;
+             continue;
+           }
+         is_branch = TRUE;
+         break;
+
+       case R_IA64_GPREL22:
+         /* Update max_short_sec/min_short_sec.  */
+
+       case R_IA64_LTOFF22X:
+       case R_IA64_LDXMOV:
+         /* We can't relax ldx/mov in pass 0 since br relaxations will
+            increase the code size. Defer it to pass 1.  */
+         if (link_info->relax_pass == 0)
+           {
+             skip_relax_pass_1 = FALSE;
+             continue;
+           }
+         is_branch = FALSE;
+         break;
+
+       default:
+         continue;
+       }
+
+      /* Get the value of the symbol referred to by the reloc.  */
+      if (ELF64_R_SYM (irel->r_info) < symtab_hdr->sh_info)
+       {
+         /* A local symbol.  */
+         Elf_Internal_Sym *isym;
+
+         /* Read this BFD's local symbols.  */
+         if (isymbuf == NULL)
+           {
+             isymbuf = (Elf_Internal_Sym *) symtab_hdr->contents;
+             if (isymbuf == NULL)
+               isymbuf = bfd_elf_get_elf_syms (abfd, symtab_hdr,
+                                               symtab_hdr->sh_info, 0,
+                                               NULL, NULL, NULL);
+             if (isymbuf == 0)
+               goto error_return;
+           }
+
+         isym = isymbuf + ELF64_R_SYM (irel->r_info);
+         if (isym->st_shndx == SHN_UNDEF)
+           continue;   /* We can't do anything with undefined symbols.  */
+         else if (isym->st_shndx == SHN_ABS)
+           tsec = bfd_abs_section_ptr;
+         else if (isym->st_shndx == SHN_COMMON)
+           tsec = bfd_com_section_ptr;
+         else if (isym->st_shndx == SHN_IA_64_ANSI_COMMON)
+           tsec = bfd_com_section_ptr;
+         else
+           tsec = bfd_section_from_elf_index (abfd, isym->st_shndx);
+
+         toff = isym->st_value;
+         dyn_i = get_dyn_sym_info (ia64_info, NULL, abfd, irel, FALSE);
+       }
+      else
+       {
+         unsigned long indx;
+         struct elf_link_hash_entry *h;
+
+         indx = ELF64_R_SYM (irel->r_info) - symtab_hdr->sh_info;
+         h = elf_sym_hashes (abfd)[indx];
+         BFD_ASSERT (h != NULL);
+
+         while (h->root.type == bfd_link_hash_indirect
+                || h->root.type == bfd_link_hash_warning)
+           h = (struct elf_link_hash_entry *) h->root.u.i.link;
+
+         dyn_i = get_dyn_sym_info (ia64_info, h, abfd, irel, FALSE);
+
+         /* For branches to dynamic symbols, we're interested instead
+            in a branch to the PLT entry.  */
+         if (is_branch && dyn_i && dyn_i->want_plt2)
+           {
+             /* Internal branches shouldn't be sent to the PLT.
+                Leave this for now and we'll give an error later.  */
+             if (r_type != R_IA64_PCREL21B)
+               continue;
+
+             tsec = ia64_info->root.splt;
+             toff = dyn_i->plt2_offset;
+             BFD_ASSERT (irel->r_addend == 0);
+           }
+
+         /* Can't do anything else with dynamic symbols.  */
+         else if (elf64_ia64_dynamic_symbol_p (h))
+           continue;
+
+         else
+           {
+             /* We can't do anything with undefined symbols.  */
+             if (h->root.type == bfd_link_hash_undefined
+                 || h->root.type == bfd_link_hash_undefweak)
+               continue;
+
+             tsec = h->root.u.def.section;
+             toff = h->root.u.def.value;
+           }
+       }
+
+      toff += irel->r_addend;
+
+      symaddr = tsec->output_section->vma + tsec->output_offset + toff;
+
+      roff = irel->r_offset;
+
+      if (is_branch)
+       {
+         bfd_signed_vma offset;
+
+         reladdr = (sec->output_section->vma
+                    + sec->output_offset
+                    + roff) & (bfd_vma) -4;
+
+         /* The .plt section is aligned at 32byte and the .text section
+            is aligned at 64byte. The .text section is right after the
+            .plt section.  After the first relaxation pass, linker may
+            increase the gap between the .plt and .text sections up
+            to 32byte.  We assume linker will always insert 32byte
+            between the .plt and .text sections after the the first
+            relaxation pass.  */
+         if (tsec == ia64_info->root.splt)
+           offset = -0x1000000 + 32;
+         else
+           offset = -0x1000000;
+
+         /* If the branch is in range, no need to do anything.  */
+         if ((bfd_signed_vma) (symaddr - reladdr) >= offset
+             && (bfd_signed_vma) (symaddr - reladdr) <= 0x0FFFFF0)
+           {
+             /* If the 60-bit branch is in 21-bit range, optimize it. */
+             if (r_type == R_IA64_PCREL60B)
+               {
+                 ia64_elf_relax_brl (contents, roff);
+
+                 irel->r_info = ELF64_R_INFO (ELF64_R_SYM (irel->r_info),
+                                               R_IA64_PCREL21B);
+
+                 /* If the original relocation offset points to slot
+                    1, change it to slot 2.  */
+                 if ((irel->r_offset & 3) == 1)
+                   irel->r_offset += 1;
+               }
+
+             continue;
+           }
+         else if (r_type == R_IA64_PCREL60B)
+           continue;
+         else if (ia64_elf_relax_br (contents, roff))
+           {
+             irel->r_info = ELF64_R_INFO (ELF64_R_SYM (irel->r_info),
+                                           R_IA64_PCREL60B);
+
+             /* Make the relocation offset point to slot 1.  */
+             irel->r_offset = (irel->r_offset & ~((bfd_vma) 0x3)) + 1;
+             continue;
+           }
+
+         /* We can't put a trampoline in a .init/.fini section. Issue
+            an error.  */
+         if (strcmp (sec->output_section->name, ".init") == 0
+             || strcmp (sec->output_section->name, ".fini") == 0)
+           {
+             (*_bfd_error_handler)
+               (_("%B: Can't relax br at 0x%lx in section `%A'. Please use brl or indirect branch."),
+                sec->owner, sec, (unsigned long) roff);
+             bfd_set_error (bfd_error_bad_value);
+             goto error_return;
+           }
+
+         /* If the branch and target are in the same section, you've
+            got one honking big section and we can't help you unless
+            you are branching backwards.  You'll get an error message
+            later.  */
+         if (tsec == sec && toff > roff)
+           continue;
+
+         /* Look for an existing fixup to this address.  */
+         for (f = fixups; f ; f = f->next)
+           if (f->tsec == tsec && f->toff == toff)
+             break;
+
+         if (f == NULL)
+           {
+             /* Two alternatives: If it's a branch to a PLT entry, we can
+                make a copy of the FULL_PLT entry.  Otherwise, we'll have
+                to use a `brl' insn to get where we're going.  */
+
+             size_t size;
+
+             if (tsec == ia64_info->root.splt)
+               size = sizeof (plt_full_entry);
+             else
+               size = sizeof (oor_brl);
+
+             /* Resize the current section to make room for the new branch. */
+             trampoff = (sec->size + 15) & (bfd_vma) -16;
+
+             /* If trampoline is out of range, there is nothing we
+                can do.  */
+             offset = trampoff - (roff & (bfd_vma) -4);
+             if (offset < -0x1000000 || offset > 0x0FFFFF0)
+               continue;
+
+             amt = trampoff + size;
+             contents = (bfd_byte *) bfd_realloc (contents, amt);
+             if (contents == NULL)
+               goto error_return;
+             sec->size = amt;
+
+             if (tsec == ia64_info->root.splt)
+               {
+                 memcpy (contents + trampoff, plt_full_entry, size);
+
+                 /* Hijack the old relocation for use as the PLTOFF reloc.  */
+                 irel->r_info = ELF64_R_INFO (ELF64_R_SYM (irel->r_info),
+                                              R_IA64_PLTOFF22);
+                 irel->r_offset = trampoff;
+               }
+             else
+               {
+                  memcpy (contents + trampoff, oor_brl, size);
+                  irel->r_info = ELF64_R_INFO (ELF64_R_SYM (irel->r_info),
+                                               R_IA64_PCREL60B);
+                  irel->r_offset = trampoff + 2;
+               }
+
+             /* Record the fixup so we don't do it again this section.  */
+             f = (struct one_fixup *)
+               bfd_malloc ((bfd_size_type) sizeof (*f));
+             f->next = fixups;
+             f->tsec = tsec;
+             f->toff = toff;
+             f->trampoff = trampoff;
+             fixups = f;
+           }
+         else
+           {
+             /* If trampoline is out of range, there is nothing we
+                can do.  */
+             offset = f->trampoff - (roff & (bfd_vma) -4);
+             if (offset < -0x1000000 || offset > 0x0FFFFF0)
+               continue;
+
+             /* Nop out the reloc, since we're finalizing things here.  */
+             irel->r_info = ELF64_R_INFO (0, R_IA64_NONE);
+           }
+
+         /* Fix up the existing branch to hit the trampoline.  */
+         if (ia64_elf_install_value (contents + roff, offset, r_type)
+             != bfd_reloc_ok)
+           goto error_return;
+
+         changed_contents = TRUE;
+         changed_relocs = TRUE;
+       }
+      else
+       {
+         /* Fetch the gp.  */
+         if (gp == 0)
+           {
+             bfd *obfd = sec->output_section->owner;
+             gp = _bfd_get_gp_value (obfd);
+             if (gp == 0)
+               {
+                 if (!elf64_ia64_choose_gp (obfd, link_info, FALSE))
+                   goto error_return;
+                 gp = _bfd_get_gp_value (obfd);
+               }
+           }
+
+         /* If the data is out of range, do nothing.  */
+         if ((bfd_signed_vma) (symaddr - gp) >= 0x200000
+             ||(bfd_signed_vma) (symaddr - gp) < -0x200000)
+           continue;
+
+         if (r_type == R_IA64_GPREL22)
+           elf64_ia64_update_short_info (tsec->output_section,
+                                         tsec->output_offset + toff,
+                                         ia64_info);
+         else if (r_type == R_IA64_LTOFF22X)
+           {
+              /* Can't deal yet correctly with ABS symbols.  */
+              if (bfd_is_abs_section (tsec))
+                continue;
+
+             irel->r_info = ELF64_R_INFO (ELF64_R_SYM (irel->r_info),
+                                          R_IA64_GPREL22);
+             changed_relocs = TRUE;
+
+             elf64_ia64_update_short_info (tsec->output_section,
+                                           tsec->output_offset + toff,
+                                           ia64_info);
+           }
+         else
+           {
+             ia64_elf_relax_ldxmov (contents, roff);
+             irel->r_info = ELF64_R_INFO (0, R_IA64_NONE);
+             changed_contents = TRUE;
+             changed_relocs = TRUE;
+           }
+       }
+    }
+
+  /* ??? If we created fixups, this may push the code segment large
+     enough that the data segment moves, which will change the GP.
+     Reset the GP so that we re-calculate next round.  We need to
+     do this at the _beginning_ of the next round; now will not do.  */
+
+  /* Clean up and go home.  */
+  while (fixups)
+    {
+      struct one_fixup *f = fixups;
+      fixups = fixups->next;
+      free (f);
+    }
+
+  if (isymbuf != NULL
+      && symtab_hdr->contents != (unsigned char *) isymbuf)
+    {
+      if (! link_info->keep_memory)
+       free (isymbuf);
+      else
+       {
+         /* Cache the symbols for elf_link_input_bfd.  */
+         symtab_hdr->contents = (unsigned char *) isymbuf;
+       }
+    }
+
+  if (contents != NULL
+      && elf_section_data (sec)->this_hdr.contents != contents)
+    {
+      if (!changed_contents && !link_info->keep_memory)
+       free (contents);
+      else
+       {
+         /* Cache the section contents for elf_link_input_bfd.  */
+         elf_section_data (sec)->this_hdr.contents = contents;
+       }
+    }
+
+  if (elf_section_data (sec)->relocs != internal_relocs)
+    {
+      if (!changed_relocs)
+       free (internal_relocs);
+      else
+       elf_section_data (sec)->relocs = internal_relocs;
+    }
+
+  if (link_info->relax_pass == 0)
+    {
+      /* Pass 0 is only needed to relax br.  */
+      sec->skip_relax_pass_0 = skip_relax_pass_0;
+      sec->skip_relax_pass_1 = skip_relax_pass_1;
+    }
+
+  *again = changed_contents || changed_relocs;
+  return TRUE;
+
+ error_return:
+  if (isymbuf != NULL && (unsigned char *) isymbuf != symtab_hdr->contents)
+    free (isymbuf);
+  if (contents != NULL
+      && elf_section_data (sec)->this_hdr.contents != contents)
+    free (contents);
+  if (internal_relocs != NULL
+      && elf_section_data (sec)->relocs != internal_relocs)
+    free (internal_relocs);
+  return FALSE;
+}
+#undef skip_relax_pass_0
+#undef skip_relax_pass_1
+
+/* Return TRUE if NAME is an unwind table section name.  */
+
+static inline bfd_boolean
+is_unwind_section_name (bfd *abfd ATTRIBUTE_UNUSED, const char *name)
+{
+  return ((CONST_STRNEQ (name, ELF_STRING_ia64_unwind)
+          && ! CONST_STRNEQ (name, ELF_STRING_ia64_unwind_info))
+         || CONST_STRNEQ (name, ELF_STRING_ia64_unwind_once));
+}
+
+
+/* Convert IA-64 specific section flags to bfd internal section flags.  */
+
+/* ??? There is no bfd internal flag equivalent to the SHF_IA_64_NORECOV
+   flag.  */
+
+static bfd_boolean
+elf64_ia64_section_flags (flagword *flags,
+                         const Elf_Internal_Shdr *hdr)
+{
+  if (hdr->sh_flags & SHF_IA_64_SHORT)
+    *flags |= SEC_SMALL_DATA;
+
+  return TRUE;
+}
+
+/* Set the correct type for an IA-64 ELF section.  We do this by the
+   section name, which is a hack, but ought to work.  */
+
+static bfd_boolean
+elf64_ia64_fake_sections (bfd *abfd, Elf_Internal_Shdr *hdr,
+                         asection *sec)
+{
+  const char *name;
+
+  name = bfd_get_section_name (abfd, sec);
+
+  if (is_unwind_section_name (abfd, name))
+    {
+      /* We don't have the sections numbered at this point, so sh_info
+        is set later, in elf64_ia64_final_write_processing.  */
+      hdr->sh_type = SHT_IA_64_UNWIND;
+      hdr->sh_flags |= SHF_LINK_ORDER;
+    }
+  else if (strcmp (name, ELF_STRING_ia64_archext) == 0)
+    hdr->sh_type = SHT_IA_64_EXT;
+
+  if (sec->flags & SEC_SMALL_DATA)
+    hdr->sh_flags |= SHF_IA_64_SHORT;
+
+  return TRUE;
+}
+
+/* Hook called by the linker routine which adds symbols from an object
+   file.  We use it to put .comm items in .sbss, and not .bss.  */
+
+static bfd_boolean
+elf64_ia64_add_symbol_hook (bfd *abfd,
+                           struct bfd_link_info *info,
+                           Elf_Internal_Sym *sym,
+                           const char **namep ATTRIBUTE_UNUSED,
+                           flagword *flagsp ATTRIBUTE_UNUSED,
+                           asection **secp,
+                           bfd_vma *valp)
+{
+  if (sym->st_shndx == SHN_COMMON
+      && !info->relocatable
+      && sym->st_size <= elf_gp_size (abfd))
+    {
+      /* Common symbols less than or equal to -G nn bytes are
+        automatically put into .sbss.  */
+
+      asection *scomm = bfd_get_section_by_name (abfd, ".scommon");
+
+      if (scomm == NULL)
+       {
+         scomm = bfd_make_section_with_flags (abfd, ".scommon",
+                                              (SEC_ALLOC
+                                               | SEC_IS_COMMON
+                                               | SEC_LINKER_CREATED));
+         if (scomm == NULL)
+           return FALSE;
+       }
+
+      *secp = scomm;
+      *valp = sym->st_size;
+    }
+
+  return TRUE;
+}
+
+/* According to the Tahoe assembler spec, all labels starting with a
+   '.' are local.  */
+
+static bfd_boolean
+elf64_ia64_is_local_label_name (bfd *abfd ATTRIBUTE_UNUSED,
+                               const char *name)
+{
+  return name[0] == '.';
+}
+
+/* Should we do dynamic things to this symbol?  */
+
+static bfd_boolean
+elf64_ia64_dynamic_symbol_p (struct elf_link_hash_entry *h)
+{
+  return h != NULL && h->def_dynamic;
+}
+
+static struct bfd_hash_entry*
+elf64_ia64_new_elf_hash_entry (struct bfd_hash_entry *entry,
+                              struct bfd_hash_table *table,
+                              const char *string)
+{
+  struct elf64_ia64_link_hash_entry *ret;
+  ret = (struct elf64_ia64_link_hash_entry *) entry;
+
+  /* Allocate the structure if it has not already been allocated by a
+     subclass.  */
+  if (!ret)
+    ret = bfd_hash_allocate (table, sizeof (*ret));
+
+  if (!ret)
+    return 0;
+
+  /* Call the allocation method of the superclass.  */
+  ret = ((struct elf64_ia64_link_hash_entry *)
+        _bfd_elf_link_hash_newfunc ((struct bfd_hash_entry *) ret,
+                                    table, string));
+
+  ret->info = NULL;
+  ret->count = 0;
+  ret->sorted_count = 0;
+  ret->size = 0;
+  return (struct bfd_hash_entry *) ret;
+}
+
+static void
+elf64_ia64_hash_hide_symbol (struct bfd_link_info *info,
+                            struct elf_link_hash_entry *xh,
+                            bfd_boolean force_local)
+{
+  struct elf64_ia64_link_hash_entry *h;
+  struct elf64_ia64_dyn_sym_info *dyn_i;
+  unsigned int count;
+
+  h = (struct elf64_ia64_link_hash_entry *)xh;
+
+  _bfd_elf_link_hash_hide_symbol (info, &h->root, force_local);
+
+  for (count = h->count, dyn_i = h->info;
+       count != 0;
+       count--, dyn_i++)
+    {
+      dyn_i->want_plt2 = 0;
+      dyn_i->want_plt = 0;
+    }
+}
+
+/* Compute a hash of a local hash entry.  */
+
+static hashval_t
+elf64_ia64_local_htab_hash (const void *ptr)
+{
+  struct elf64_ia64_local_hash_entry *entry
+    = (struct elf64_ia64_local_hash_entry *) ptr;
+
+  return ELF_LOCAL_SYMBOL_HASH (entry->id, entry->r_sym);
+}
+
+/* Compare local hash entries.  */
+
+static int
+elf64_ia64_local_htab_eq (const void *ptr1, const void *ptr2)
+{
+  struct elf64_ia64_local_hash_entry *entry1
+    = (struct elf64_ia64_local_hash_entry *) ptr1;
+  struct elf64_ia64_local_hash_entry *entry2
+    = (struct elf64_ia64_local_hash_entry *) ptr2;
+
+  return entry1->id == entry2->id && entry1->r_sym == entry2->r_sym;
+}
+
+/* Create the derived linker hash table.  The IA-64 ELF port uses this
+   derived hash table to keep information specific to the IA-64 ElF
+   linker (without using static variables).  */
+
+static struct bfd_link_hash_table *
+elf64_ia64_hash_table_create (bfd *abfd)
+{
+  struct elf64_ia64_link_hash_table *ret;
+
+  ret = bfd_zmalloc ((bfd_size_type) sizeof (*ret));
+  if (!ret)
+    return NULL;
+
+  if (!_bfd_elf_link_hash_table_init (&ret->root, abfd,
+                                     elf64_ia64_new_elf_hash_entry,
+                                     sizeof (struct elf64_ia64_link_hash_entry),
+                                     IA64_ELF_DATA))
+    {
+      free (ret);
+      return NULL;
+    }
+
+  ret->loc_hash_table = htab_try_create (1024, elf64_ia64_local_htab_hash,
+                                        elf64_ia64_local_htab_eq, NULL);
+  ret->loc_hash_memory = objalloc_create ();
+  if (!ret->loc_hash_table || !ret->loc_hash_memory)
+    {
+      free (ret);
+      return NULL;
+    }
+
+  return &ret->root.root;
+}
+
+/* Free the global elf64_ia64_dyn_sym_info array.  */
+
+static bfd_boolean
+elf64_ia64_global_dyn_info_free (void **xentry,
+                                void * unused ATTRIBUTE_UNUSED)
+{
+  struct elf64_ia64_link_hash_entry *entry
+    = (struct elf64_ia64_link_hash_entry *) xentry;
+
+  if (entry->root.root.type == bfd_link_hash_warning)
+    entry = (struct elf64_ia64_link_hash_entry *) entry->root.root.u.i.link;
+
+  if (entry->info)
+    {
+      free (entry->info);
+      entry->info = NULL;
+      entry->count = 0;
+      entry->sorted_count = 0;
+      entry->size = 0;
+    }
+
+  return TRUE;
+}
+
+/* Free the local elf64_ia64_dyn_sym_info array.  */
+
+static bfd_boolean
+elf64_ia64_local_dyn_info_free (void **slot,
+                               void * unused ATTRIBUTE_UNUSED)
+{
+  struct elf64_ia64_local_hash_entry *entry
+    = (struct elf64_ia64_local_hash_entry *) *slot;
+
+  if (entry->info)
+    {
+      free (entry->info);
+      entry->info = NULL;
+      entry->count = 0;
+      entry->sorted_count = 0;
+      entry->size = 0;
+    }
+
+  return TRUE;
+}
+
+/* Destroy IA-64 linker hash table.  */
+
+static void
+elf64_ia64_hash_table_free (struct bfd_link_hash_table *hash)
+{
+  struct elf64_ia64_link_hash_table *ia64_info
+    = (struct elf64_ia64_link_hash_table *) hash;
+  if (ia64_info->loc_hash_table)
+    {
+      htab_traverse (ia64_info->loc_hash_table,
+                    elf64_ia64_local_dyn_info_free, NULL);
+      htab_delete (ia64_info->loc_hash_table);
+    }
+  if (ia64_info->loc_hash_memory)
+    objalloc_free ((struct objalloc *) ia64_info->loc_hash_memory);
+  elf_link_hash_traverse (&ia64_info->root,
+                         elf64_ia64_global_dyn_info_free, NULL);
+  _bfd_generic_link_hash_table_free (hash);
+}
+
+/* Traverse both local and global hash tables.  */
+
+struct elf64_ia64_dyn_sym_traverse_data
+{
+  bfd_boolean (*func) (struct elf64_ia64_dyn_sym_info *, void *);
+  void * data;
+};
+
+static bfd_boolean
+elf64_ia64_global_dyn_sym_thunk (struct bfd_hash_entry *xentry,
+                                void * xdata)
+{
+  struct elf64_ia64_link_hash_entry *entry
+    = (struct elf64_ia64_link_hash_entry *) xentry;
+  struct elf64_ia64_dyn_sym_traverse_data *data
+    = (struct elf64_ia64_dyn_sym_traverse_data *) xdata;
+  struct elf64_ia64_dyn_sym_info *dyn_i;
+  unsigned int count;
+
+  if (entry->root.root.type == bfd_link_hash_warning)
+    entry = (struct elf64_ia64_link_hash_entry *) entry->root.root.u.i.link;
+
+  for (count = entry->count, dyn_i = entry->info;
+       count != 0;
+       count--, dyn_i++)
+    if (! (*data->func) (dyn_i, data->data))
+      return FALSE;
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_ia64_local_dyn_sym_thunk (void **slot, void * xdata)
+{
+  struct elf64_ia64_local_hash_entry *entry
+    = (struct elf64_ia64_local_hash_entry *) *slot;
+  struct elf64_ia64_dyn_sym_traverse_data *data
+    = (struct elf64_ia64_dyn_sym_traverse_data *) xdata;
+  struct elf64_ia64_dyn_sym_info *dyn_i;
+  unsigned int count;
+
+  for (count = entry->count, dyn_i = entry->info;
+       count != 0;
+       count--, dyn_i++)
+    if (! (*data->func) (dyn_i, data->data))
+      return FALSE;
+  return TRUE;
+}
+
+static void
+elf64_ia64_dyn_sym_traverse (struct elf64_ia64_link_hash_table *ia64_info,
+                            bfd_boolean (*func) (struct elf64_ia64_dyn_sym_info *, void *),
+                            void * data)
+{
+  struct elf64_ia64_dyn_sym_traverse_data xdata;
+
+  xdata.func = func;
+  xdata.data = data;
+
+  elf_link_hash_traverse (&ia64_info->root,
+                         elf64_ia64_global_dyn_sym_thunk, &xdata);
+  htab_traverse (ia64_info->loc_hash_table,
+                elf64_ia64_local_dyn_sym_thunk, &xdata);
+}
+
+#define NOTE_NAME "IPF/VMS"
+
+static bfd_boolean
+create_ia64_vms_notes (bfd *abfd, struct bfd_link_info *info,
+                       unsigned int time_hi, unsigned int time_lo)
+{
+#define NBR_NOTES 7
+  Elf_Internal_Note notes[NBR_NOTES];
+  char *module_name;
+  int module_name_len;
+  unsigned char cur_time[8];
+  Elf64_External_VMS_ORIG_DYN_Note *orig_dyn;
+  unsigned int orig_dyn_size;
+  unsigned int note_size;
+  int i;
+  unsigned char *noteptr;
+  unsigned char *note_contents;
+  struct elf64_ia64_link_hash_table *ia64_info;
+
+  ia64_info = elf64_ia64_hash_table (info);
+
+  module_name = vms_get_module_name (bfd_get_filename (abfd), TRUE);
+  module_name_len = strlen (module_name) + 1;
+
+  bfd_putl32 (time_lo, cur_time + 0);
+  bfd_putl32 (time_hi, cur_time + 4);
+
+  /* Note 0: IMGNAM.  */
+  notes[0].type = NT_VMS_IMGNAM;
+  notes[0].descdata = module_name;
+  notes[0].descsz = module_name_len;
+
+  /* Note 1: GSTNAM.  */
+  notes[1].type = NT_VMS_GSTNAM;
+  notes[1].descdata = module_name;
+  notes[1].descsz = module_name_len;
+
+  /* Note 2: IMGID.  */
+#define IMG_ID "V1.0"
+  notes[2].type = NT_VMS_IMGID;
+  notes[2].descdata = IMG_ID;
+  notes[2].descsz = sizeof (IMG_ID);
+
+  /* Note 3: Linktime.  */
+  notes[3].type = NT_VMS_LINKTIME;
+  notes[3].descdata = (char *)cur_time;
+  notes[3].descsz = sizeof (cur_time);
+
+  /* Note 4: Linker id.  */
+  notes[4].type = NT_VMS_LINKID;
+  notes[4].descdata = "GNU ld " BFD_VERSION_STRING;
+  notes[4].descsz = strlen (notes[4].descdata) + 1;
+
+  /* Note 5: Original dyn.  */
+  orig_dyn_size = (sizeof (*orig_dyn) + sizeof (IMG_ID) - 1 + 7) & ~7;
+  orig_dyn = bfd_zalloc (abfd, orig_dyn_size);
+  if (orig_dyn == NULL)
+    return FALSE;
+  bfd_putl32 (1, orig_dyn->major_id);
+  bfd_putl32 (3, orig_dyn->minor_id);
+  memcpy (orig_dyn->manipulation_date, cur_time, sizeof (cur_time));
+  bfd_putl64 (VMS_LF_IMGSTA | VMS_LF_MAIN, orig_dyn->link_flags);
+  bfd_putl32 (EF_IA_64_ABI64, orig_dyn->elf_flags);
+  memcpy (orig_dyn->imgid, IMG_ID, sizeof (IMG_ID));
+  notes[5].type = NT_VMS_ORIG_DYN;
+  notes[5].descdata = (char *)orig_dyn;
+  notes[5].descsz = orig_dyn_size;
+
+  /* Note 3: Patchtime.  */
+  notes[6].type = NT_VMS_PATCHTIME;
+  notes[6].descdata = (char *)cur_time;
+  notes[6].descsz = sizeof (cur_time);
+
+  /* Compute notes size.  */
+  note_size = 0;
+  for (i = 0; i < NBR_NOTES; i++)
+    note_size += sizeof (Elf64_External_VMS_Note) - 1
+      + ((sizeof (NOTE_NAME) - 1 + 7) & ~7)
+      + ((notes[i].descsz + 7) & ~7);
+
+  /* Malloc a temporary buffer large enough for most notes */
+  note_contents = (unsigned char *) bfd_zalloc (abfd, note_size);
+  if (note_contents == NULL)
+    return FALSE;
+  noteptr = note_contents;
+
+  /* Fill notes.  */
+  for (i = 0; i < NBR_NOTES; i++)
+    {
+      Elf64_External_VMS_Note *enote = (Elf64_External_VMS_Note *) noteptr;
+
+      bfd_putl64 (sizeof (NOTE_NAME) - 1, enote->namesz);
+      bfd_putl64 (notes[i].descsz, enote->descsz);
+      bfd_putl64 (notes[i].type, enote->type);
+
+      noteptr = (unsigned char *)enote->name;
+      memcpy (noteptr, NOTE_NAME, sizeof (NOTE_NAME) - 1);
+      noteptr += (sizeof (NOTE_NAME) - 1 + 7) & ~7;
+      memcpy (noteptr, notes[i].descdata, notes[i].descsz);
+      noteptr += (notes[i].descsz + 7) & ~7;
+    }
+
+  ia64_info->note_sec->contents = note_contents;
+  ia64_info->note_sec->size = note_size;
+
+  free (module_name);
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_ia64_create_dynamic_sections (bfd *abfd,
+                                   struct bfd_link_info *info)
+{
+  struct elf64_ia64_link_hash_table *ia64_info;
+  asection *s;
+  flagword flags;
+  const struct elf_backend_data *bed;
+
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return FALSE;
+
+  if (elf_hash_table (info)->dynamic_sections_created)
+    return TRUE;
+
+  abfd = elf_hash_table (info)->dynobj;
+  bed = get_elf_backend_data (abfd);
+
+  flags = bed->dynamic_sec_flags;
+
+  s = bfd_make_section_with_flags (abfd, ".dynamic", flags | SEC_READONLY);
+  if (s == NULL
+      || ! bfd_set_section_alignment (abfd, s, bed->s->log_file_align))
+    return FALSE;
+
+  s = bfd_make_section_with_flags (abfd, ".plt", flags | SEC_READONLY);
+  if (s == NULL
+      || ! bfd_set_section_alignment (abfd, s, bed->plt_alignment))
+    return FALSE;
+  ia64_info->root.splt = s;
+
+  if (!get_got (abfd, ia64_info))
+    return FALSE;
+
+  if (!get_pltoff (abfd, ia64_info))
+    return FALSE;
+
+  s = bfd_make_section_with_flags (abfd, ".vmsdynstr",
+                                  (SEC_ALLOC
+                                   | SEC_HAS_CONTENTS
+                                   | SEC_IN_MEMORY
+                                   | SEC_LINKER_CREATED));
+  if (s == NULL
+      || !bfd_set_section_alignment (abfd, s, 0))
+    return FALSE;
+
+  /* Create a fixup section.  */
+  s = bfd_make_section_with_flags (abfd, ".fixups",
+                                   (SEC_ALLOC
+                                    | SEC_HAS_CONTENTS
+                                    | SEC_IN_MEMORY
+                                    | SEC_LINKER_CREATED));
+  if (s == NULL
+      || !bfd_set_section_alignment (abfd, s, 3))
+    return FALSE;
+  ia64_info->fixups_sec = s;
+
+  /* Create the transfer fixup section.  */
+  s = bfd_make_section_with_flags (abfd, ".transfer",
+                                   (SEC_ALLOC
+                                    | SEC_HAS_CONTENTS
+                                    | SEC_IN_MEMORY
+                                    | SEC_LINKER_CREATED));
+  if (s == NULL
+      || !bfd_set_section_alignment (abfd, s, 3))
+    return FALSE;
+  s->size = sizeof (struct elf64_vms_transfer);
+  ia64_info->transfer_sec = s;
+
+  /* Create note section.  */
+  s = bfd_make_section_anyway_with_flags (abfd, ".vms.note",
+                                          (SEC_LINKER_CREATED
+                                           | SEC_HAS_CONTENTS
+                                           | SEC_IN_MEMORY
+                                           | SEC_READONLY));
+  if (s == NULL
+      || !bfd_set_section_alignment (abfd, s, 3))
+    return FALSE;
+  ia64_info->note_sec = s;
+
+  elf_hash_table (info)->dynamic_sections_created = TRUE;
+  return TRUE;
+}
+
+/* Find and/or create a hash entry for local symbol.  */
+static struct elf64_ia64_local_hash_entry *
+get_local_sym_hash (struct elf64_ia64_link_hash_table *ia64_info,
+                   bfd *abfd, const Elf_Internal_Rela *rel,
+                   bfd_boolean create)
+{
+  struct elf64_ia64_local_hash_entry e, *ret;
+  asection *sec = abfd->sections;
+  hashval_t h = ELF_LOCAL_SYMBOL_HASH (sec->id,
+                                      ELF64_R_SYM (rel->r_info));
+  void **slot;
+
+  e.id = sec->id;
+  e.r_sym = ELF64_R_SYM (rel->r_info);
+  slot = htab_find_slot_with_hash (ia64_info->loc_hash_table, &e, h,
+                                  create ? INSERT : NO_INSERT);
+
+  if (!slot)
+    return NULL;
+
+  if (*slot)
+    return (struct elf64_ia64_local_hash_entry *) *slot;
+
+  ret = (struct elf64_ia64_local_hash_entry *)
+       objalloc_alloc ((struct objalloc *) ia64_info->loc_hash_memory,
+                       sizeof (struct elf64_ia64_local_hash_entry));
+  if (ret)
+    {
+      memset (ret, 0, sizeof (*ret));
+      ret->id = sec->id;
+      ret->r_sym = ELF64_R_SYM (rel->r_info);
+      *slot = ret;
+    }
+  return ret;
+}
+
+/* Used to sort elf64_ia64_dyn_sym_info array.  */
+
+static int
+addend_compare (const void *xp, const void *yp)
+{
+  const struct elf64_ia64_dyn_sym_info *x
+    = (const struct elf64_ia64_dyn_sym_info *) xp;
+  const struct elf64_ia64_dyn_sym_info *y
+    = (const struct elf64_ia64_dyn_sym_info *) yp;
+
+  return x->addend < y->addend ? -1 : x->addend > y->addend ? 1 : 0;
+}
+
+/* Sort elf64_ia64_dyn_sym_info array and remove duplicates.  */
+
+static unsigned int
+sort_dyn_sym_info (struct elf64_ia64_dyn_sym_info *info,
+                  unsigned int count)
+{
+  bfd_vma curr, prev, got_offset;
+  unsigned int i, kept, dupes, diff, dest, src, len;
+
+  qsort (info, count, sizeof (*info), addend_compare);
+
+  /* Find the first duplicate.  */
+  prev = info [0].addend;
+  got_offset = info [0].got_offset;
+  for (i = 1; i < count; i++)
+    {
+      curr = info [i].addend;
+      if (curr == prev)
+       {
+         /* For duplicates, make sure that GOT_OFFSET is valid.  */
+         if (got_offset == (bfd_vma) -1)
+           got_offset = info [i].got_offset;
+         break;
+       }
+      got_offset = info [i].got_offset;
+      prev = curr;
+    }
+
+  /* We may move a block of elements to here.  */
+  dest = i++;
+
+  /* Remove duplicates.  */
+  if (i < count)
+    {
+      while (i < count)
+       {
+         /* For duplicates, make sure that the kept one has a valid
+            got_offset.  */
+         kept = dest - 1;
+         if (got_offset != (bfd_vma) -1)
+           info [kept].got_offset = got_offset;
+
+         curr = info [i].addend;
+         got_offset = info [i].got_offset;
+
+         /* Move a block of elements whose first one is different from
+            the previous.  */
+         if (curr == prev)
+           {
+             for (src = i + 1; src < count; src++)
+               {
+                 if (info [src].addend != curr)
+                   break;
+                 /* For duplicates, make sure that GOT_OFFSET is
+                    valid.  */
+                 if (got_offset == (bfd_vma) -1)
+                   got_offset = info [src].got_offset;
+               }
+
+             /* Make sure that the kept one has a valid got_offset.  */
+             if (got_offset != (bfd_vma) -1)
+               info [kept].got_offset = got_offset;
+           }
+         else
+           src = i;
+
+         if (src >= count)
+           break;
+
+         /* Find the next duplicate.  SRC will be kept.  */
+         prev = info [src].addend;
+         got_offset = info [src].got_offset;
+         for (dupes = src + 1; dupes < count; dupes ++)
+           {
+             curr = info [dupes].addend;
+             if (curr == prev)
+               {
+                 /* Make sure that got_offset is valid.  */
+                 if (got_offset == (bfd_vma) -1)
+                   got_offset = info [dupes].got_offset;
+
+                 /* For duplicates, make sure that the kept one has
+                    a valid got_offset.  */
+                 if (got_offset != (bfd_vma) -1)
+                   info [dupes - 1].got_offset = got_offset;
+                 break;
+               }
+             got_offset = info [dupes].got_offset;
+             prev = curr;
+           }
+
+         /* How much to move.  */
+         len = dupes - src;
+         i = dupes + 1;
+
+         if (len == 1 && dupes < count)
+           {
+             /* If we only move 1 element, we combine it with the next
+                one.  There must be at least a duplicate.  Find the
+                next different one.  */
+             for (diff = dupes + 1, src++; diff < count; diff++, src++)
+               {
+                 if (info [diff].addend != curr)
+                   break;
+                 /* Make sure that got_offset is valid.  */
+                 if (got_offset == (bfd_vma) -1)
+                   got_offset = info [diff].got_offset;
+               }
+
+             /* Makre sure that the last duplicated one has an valid
+                offset.  */
+             BFD_ASSERT (curr == prev);
+             if (got_offset != (bfd_vma) -1)
+               info [diff - 1].got_offset = got_offset;
+
+             if (diff < count)
+               {
+                 /* Find the next duplicate.  Track the current valid
+                    offset.  */
+                 prev = info [diff].addend;
+                 got_offset = info [diff].got_offset;
+                 for (dupes = diff + 1; dupes < count; dupes ++)
+                   {
+                     curr = info [dupes].addend;
+                     if (curr == prev)
+                       {
+                         /* For duplicates, make sure that GOT_OFFSET
+                            is valid.  */
+                         if (got_offset == (bfd_vma) -1)
+                           got_offset = info [dupes].got_offset;
+                         break;
+                       }
+                     got_offset = info [dupes].got_offset;
+                     prev = curr;
+                     diff++;
+                   }
+
+                 len = diff - src + 1;
+                 i = diff + 1;
+               }
+           }
+
+         memmove (&info [dest], &info [src], len * sizeof (*info));
+
+         dest += len;
+       }
+
+      count = dest;
+    }
+  else
+    {
+      /* When we get here, either there is no duplicate at all or
+        the only duplicate is the last element.  */
+      if (dest < count)
+       {
+         /* If the last element is a duplicate, make sure that the
+            kept one has a valid got_offset.  We also update count.  */
+         if (got_offset != (bfd_vma) -1)
+           info [dest - 1].got_offset = got_offset;
+         count = dest;
+       }
+    }
+
+  return count;
+}
+
+/* Find and/or create a descriptor for dynamic symbol info.  This will
+   vary based on global or local symbol, and the addend to the reloc.
+
+   We don't sort when inserting.  Also, we sort and eliminate
+   duplicates if there is an unsorted section.  Typically, this will
+   only happen once, because we do all insertions before lookups.  We
+   then use bsearch to do a lookup.  This also allows lookups to be
+   fast.  So we have fast insertion (O(log N) due to duplicate check),
+   fast lookup (O(log N)) and one sort (O(N log N) expected time).
+   Previously, all lookups were O(N) because of the use of the linked
+   list and also all insertions were O(N) because of the check for
+   duplicates.  There are some complications here because the array
+   size grows occasionally, which may add an O(N) factor, but this
+   should be rare.  Also,  we free the excess array allocation, which
+   requires a copy which is O(N), but this only happens once.  */
+
+static struct elf64_ia64_dyn_sym_info *
+get_dyn_sym_info (struct elf64_ia64_link_hash_table *ia64_info,
+                 struct elf_link_hash_entry *h, bfd *abfd,
+                 const Elf_Internal_Rela *rel, bfd_boolean create)
+{
+  struct elf64_ia64_dyn_sym_info **info_p, *info, *dyn_i, key;
+  unsigned int *count_p, *sorted_count_p, *size_p;
+  unsigned int count, sorted_count, size;
+  bfd_vma addend = rel ? rel->r_addend : 0;
+  bfd_size_type amt;
+
+  if (h)
+    {
+      struct elf64_ia64_link_hash_entry *global_h;
+
+      global_h = (struct elf64_ia64_link_hash_entry *) h;
+      info_p = &global_h->info;
+      count_p = &global_h->count;
+      sorted_count_p = &global_h->sorted_count;
+      size_p = &global_h->size;
+    }
+  else
+    {
+      struct elf64_ia64_local_hash_entry *loc_h;
+
+      loc_h = get_local_sym_hash (ia64_info, abfd, rel, create);
+      if (!loc_h)
+       {
+         BFD_ASSERT (!create);
+         return NULL;
+       }
+
+      info_p = &loc_h->info;
+      count_p = &loc_h->count;
+      sorted_count_p = &loc_h->sorted_count;
+      size_p = &loc_h->size;
+    }
+
+  count = *count_p;
+  sorted_count = *sorted_count_p;
+  size = *size_p;
+  info = *info_p;
+  if (create)
+    {
+      /* When we create the array, we don't check for duplicates,
+         except in the previously sorted section if one exists, and
+        against the last inserted entry.  This allows insertions to
+        be fast.  */
+      if (info)
+       {
+         if (sorted_count)
+           {
+             /* Try bsearch first on the sorted section.  */
+             key.addend = addend;
+             dyn_i = bsearch (&key, info, sorted_count,
+                              sizeof (*info), addend_compare);
+
+             if (dyn_i)
+               {
+                 return dyn_i;
+               }
+           }
+
+         /* Do a quick check for the last inserted entry.  */
+         dyn_i = info + count - 1;
+         if (dyn_i->addend == addend)
+           {
+             return dyn_i;
+           }
+       }
+
+      if (size == 0)
+       {
+         /* It is the very first element. We create the array of size
+            1.  */
+         size = 1;
+         amt = size * sizeof (*info);
+         info = bfd_malloc (amt);
+       }
+      else if (size <= count)
+       {
+         /* We double the array size every time when we reach the
+            size limit.  */
+         size += size;
+         amt = size * sizeof (*info);
+         info = bfd_realloc (info, amt);
+       }
+      else
+       goto has_space;
+
+      if (info == NULL)
+       return NULL;
+      *size_p = size;
+      *info_p = info;
+
+has_space:
+      /* Append the new one to the array.  */
+      dyn_i = info + count;
+      memset (dyn_i, 0, sizeof (*dyn_i));
+      dyn_i->got_offset = (bfd_vma) -1;
+      dyn_i->addend = addend;
+
+      /* We increment count only since the new ones are unsorted and
+        may have duplicate.  */
+      (*count_p)++;
+    }
+  else
+    {
+      /* It is a lookup without insertion.  Sort array if part of the
+        array isn't sorted.  */
+      if (count != sorted_count)
+       {
+         count = sort_dyn_sym_info (info, count);
+         *count_p = count;
+         *sorted_count_p = count;
+       }
+
+      /* Free unused memory.  */
+      if (size != count)
+       {
+         amt = count * sizeof (*info);
+         info = bfd_malloc (amt);
+         if (info != NULL)
+           {
+             memcpy (info, *info_p, amt);
+             free (*info_p);
+             *size_p = count;
+             *info_p = info;
+           }
+       }
+
+      key.addend = addend;
+      dyn_i = bsearch (&key, info, count,
+                      sizeof (*info), addend_compare);
+    }
+
+  return dyn_i;
+}
+
+static asection *
+get_got (bfd *abfd, struct elf64_ia64_link_hash_table *ia64_info)
+{
+  asection *got;
+  bfd *dynobj;
+
+  got = ia64_info->root.sgot;
+  if (!got)
+    {
+      flagword flags;
+
+      dynobj = ia64_info->root.dynobj;
+      if (!dynobj)
+       ia64_info->root.dynobj = dynobj = abfd;
+
+      /* The .got section is always aligned at 8 bytes.  */
+      flags = get_elf_backend_data (dynobj)->dynamic_sec_flags;
+      got = bfd_make_section_with_flags (dynobj, ".got", flags | SEC_SMALL_DATA);
+      if (got == NULL
+          || !bfd_set_section_alignment (dynobj, got, 3))
+        return NULL;
+      ia64_info->root.sgot = got;
+    }
+
+  return got;
+}
+
+/* Create function descriptor section (.opd).  This section is called .opd
+   because it contains "official procedure descriptors".  The "official"
+   refers to the fact that these descriptors are used when taking the address
+   of a procedure, thus ensuring a unique address for each procedure.  */
+
+static asection *
+get_fptr (bfd *abfd, struct bfd_link_info *info,
+         struct elf64_ia64_link_hash_table *ia64_info)
+{
+  asection *fptr;
+  bfd *dynobj;
+
+  fptr = ia64_info->fptr_sec;
+  if (!fptr)
+    {
+      dynobj = ia64_info->root.dynobj;
+      if (!dynobj)
+       ia64_info->root.dynobj = dynobj = abfd;
+
+      fptr = bfd_make_section_with_flags (dynobj, ".opd",
+                                         (SEC_ALLOC
+                                          | SEC_LOAD
+                                          | SEC_HAS_CONTENTS
+                                          | SEC_IN_MEMORY
+                                          | (info->pie ? 0 : SEC_READONLY)
+                                          | SEC_LINKER_CREATED));
+      if (!fptr
+         || !bfd_set_section_alignment (dynobj, fptr, 4))
+       {
+         BFD_ASSERT (0);
+         return NULL;
+       }
+
+      ia64_info->fptr_sec = fptr;
+
+      if (info->pie)
+       {
+         asection *fptr_rel;
+         fptr_rel = bfd_make_section_with_flags (dynobj, ".rela.opd",
+                                                 (SEC_ALLOC | SEC_LOAD
+                                                  | SEC_HAS_CONTENTS
+                                                  | SEC_IN_MEMORY
+                                                  | SEC_LINKER_CREATED
+                                                  | SEC_READONLY));
+         if (fptr_rel == NULL
+             || !bfd_set_section_alignment (dynobj, fptr_rel, 3))
+           {
+             BFD_ASSERT (0);
+             return NULL;
+           }
+
+         ia64_info->rel_fptr_sec = fptr_rel;
+       }
+    }
+
+  return fptr;
+}
+
+static asection *
+get_pltoff (bfd *abfd, struct elf64_ia64_link_hash_table *ia64_info)
+{
+  asection *pltoff;
+  bfd *dynobj;
+
+  pltoff = ia64_info->pltoff_sec;
+  if (!pltoff)
+    {
+      dynobj = ia64_info->root.dynobj;
+      if (!dynobj)
+       ia64_info->root.dynobj = dynobj = abfd;
+
+      pltoff = bfd_make_section_with_flags (dynobj,
+                                           ELF_STRING_ia64_pltoff,
+                                           (SEC_ALLOC
+                                            | SEC_LOAD
+                                            | SEC_HAS_CONTENTS
+                                            | SEC_IN_MEMORY
+                                            | SEC_SMALL_DATA
+                                            | SEC_LINKER_CREATED));
+      if (!pltoff
+         || !bfd_set_section_alignment (dynobj, pltoff, 4))
+       {
+         BFD_ASSERT (0);
+         return NULL;
+       }
+
+      ia64_info->pltoff_sec = pltoff;
+    }
+
+  return pltoff;
+}
+
+static asection *
+get_reloc_section (bfd *abfd,
+                  struct elf64_ia64_link_hash_table *ia64_info,
+                  asection *sec, bfd_boolean create)
+{
+  const char *srel_name;
+  asection *srel;
+  bfd *dynobj;
+
+  srel_name = (bfd_elf_string_from_elf_section
+              (abfd, elf_elfheader(abfd)->e_shstrndx,
+               _bfd_elf_single_rel_hdr (sec)->sh_name));
+  if (srel_name == NULL)
+    return NULL;
+
+  BFD_ASSERT ((CONST_STRNEQ (srel_name, ".rela")
+              && strcmp (bfd_get_section_name (abfd, sec),
+                         srel_name+5) == 0)
+             || (CONST_STRNEQ (srel_name, ".rel")
+                 && strcmp (bfd_get_section_name (abfd, sec),
+                            srel_name+4) == 0));
+
+  dynobj = ia64_info->root.dynobj;
+  if (!dynobj)
+    ia64_info->root.dynobj = dynobj = abfd;
+
+  srel = bfd_get_section_by_name (dynobj, srel_name);
+  if (srel == NULL && create)
+    {
+      srel = bfd_make_section_with_flags (dynobj, srel_name,
+                                         (SEC_ALLOC | SEC_LOAD
+                                          | SEC_HAS_CONTENTS
+                                          | SEC_IN_MEMORY
+                                          | SEC_LINKER_CREATED
+                                          | SEC_READONLY));
+      if (srel == NULL
+         || !bfd_set_section_alignment (dynobj, srel, 3))
+       return NULL;
+    }
+
+  return srel;
+}
+
+static bfd_boolean
+count_dyn_reloc (bfd *abfd, struct elf64_ia64_dyn_sym_info *dyn_i,
+                asection *srel, int type)
+{
+  struct elf64_ia64_dyn_reloc_entry *rent;
+
+  for (rent = dyn_i->reloc_entries; rent; rent = rent->next)
+    if (rent->srel == srel && rent->type == type)
+      break;
+
+  if (!rent)
+    {
+      rent = ((struct elf64_ia64_dyn_reloc_entry *)
+             bfd_alloc (abfd, (bfd_size_type) sizeof (*rent)));
+      if (!rent)
+       return FALSE;
+
+      rent->next = dyn_i->reloc_entries;
+      rent->srel = srel;
+      rent->type = type;
+      rent->count = 0;
+      dyn_i->reloc_entries = rent;
+    }
+  rent->count++;
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_ia64_check_relocs (bfd *abfd, struct bfd_link_info *info,
+                        asection *sec,
+                        const Elf_Internal_Rela *relocs)
+{
+  struct elf64_ia64_link_hash_table *ia64_info;
+  const Elf_Internal_Rela *relend;
+  Elf_Internal_Shdr *symtab_hdr;
+  const Elf_Internal_Rela *rel;
+  asection *got, *fptr, *srel, *pltoff;
+  enum {
+    NEED_GOT = 1,
+    NEED_GOTX = 2,
+    NEED_FPTR = 4,
+    NEED_PLTOFF = 8,
+    NEED_MIN_PLT = 16,
+    NEED_FULL_PLT = 32,
+    NEED_DYNREL = 64,
+    NEED_LTOFF_FPTR = 128
+  };
+  int need_entry;
+  struct elf_link_hash_entry *h;
+  unsigned long r_symndx;
+  bfd_boolean maybe_dynamic;
+
+  if (info->relocatable)
+    return TRUE;
+
+  symtab_hdr = &elf_tdata (abfd)->symtab_hdr;
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return FALSE;
+
+  got = fptr = srel = pltoff = NULL;
+
+  relend = relocs + sec->reloc_count;
+
+  /* We scan relocations first to create dynamic relocation arrays.  We
+     modified get_dyn_sym_info to allow fast insertion and support fast
+     lookup in the next loop.  */
+  for (rel = relocs; rel < relend; ++rel)
+    {
+      r_symndx = ELF64_R_SYM (rel->r_info);
+      if (r_symndx >= symtab_hdr->sh_info)
+       {
+         long indx = r_symndx - symtab_hdr->sh_info;
+         h = elf_sym_hashes (abfd)[indx];
+         while (h->root.type == bfd_link_hash_indirect
+                || h->root.type == bfd_link_hash_warning)
+           h = (struct elf_link_hash_entry *) h->root.u.i.link;
+       }
+      else
+       h = NULL;
+
+      /* We can only get preliminary data on whether a symbol is
+        locally or externally defined, as not all of the input files
+        have yet been processed.  Do something with what we know, as
+        this may help reduce memory usage and processing time later.  */
+      maybe_dynamic = (h && ((!info->executable
+                             && (!SYMBOLIC_BIND (info, h)
+                                 || info->unresolved_syms_in_shared_libs == RM_IGNORE))
+                            || !h->def_regular
+                            || h->root.type == bfd_link_hash_defweak));
+
+      need_entry = 0;
+      switch (ELF64_R_TYPE (rel->r_info))
+       {
+       case R_IA64_TPREL64MSB:
+       case R_IA64_TPREL64LSB:
+       case R_IA64_LTOFF_TPREL22:
+       case R_IA64_DTPREL32MSB:
+       case R_IA64_DTPREL32LSB:
+       case R_IA64_DTPREL64MSB:
+       case R_IA64_DTPREL64LSB:
+       case R_IA64_LTOFF_DTPREL22:
+       case R_IA64_DTPMOD64MSB:
+       case R_IA64_DTPMOD64LSB:
+       case R_IA64_LTOFF_DTPMOD22:
+          abort ();
+         break;
+
+       case R_IA64_IPLTMSB:
+       case R_IA64_IPLTLSB:
+          break;
+
+       case R_IA64_LTOFF_FPTR22:
+       case R_IA64_LTOFF_FPTR64I:
+       case R_IA64_LTOFF_FPTR32MSB:
+       case R_IA64_LTOFF_FPTR32LSB:
+       case R_IA64_LTOFF_FPTR64MSB:
+       case R_IA64_LTOFF_FPTR64LSB:
+         need_entry = NEED_FPTR | NEED_GOT | NEED_LTOFF_FPTR;
+         break;
+
+       case R_IA64_FPTR64I:
+       case R_IA64_FPTR32MSB:
+       case R_IA64_FPTR32LSB:
+       case R_IA64_FPTR64MSB:
+       case R_IA64_FPTR64LSB:
+         if (info->shared || h)
+           need_entry = NEED_FPTR | NEED_DYNREL;
+         else
+           need_entry = NEED_FPTR;
+         break;
+
+       case R_IA64_LTOFF22:
+       case R_IA64_LTOFF64I:
+         need_entry = NEED_GOT;
+         break;
+
+       case R_IA64_LTOFF22X:
+         need_entry = NEED_GOTX;
+         break;
+
+       case R_IA64_PLTOFF22:
+       case R_IA64_PLTOFF64I:
+       case R_IA64_PLTOFF64MSB:
+       case R_IA64_PLTOFF64LSB:
+         need_entry = NEED_PLTOFF;
+         if (h)
+           {
+             if (maybe_dynamic)
+               need_entry |= NEED_MIN_PLT;
+           }
+         else
+           {
+             (*info->callbacks->warning)
+               (info, _("@pltoff reloc against local symbol"), 0,
+                abfd, 0, (bfd_vma) 0);
+           }
+         break;
+
+       case R_IA64_PCREL21B:
+        case R_IA64_PCREL60B:
+         /* Depending on where this symbol is defined, we may or may not
+            need a full plt entry.  Only skip if we know we'll not need
+            the entry -- static or symbolic, and the symbol definition
+            has already been seen.  */
+         if (maybe_dynamic && rel->r_addend == 0)
+           need_entry = NEED_FULL_PLT;
+         break;
+
+       case R_IA64_IMM14:
+       case R_IA64_IMM22:
+       case R_IA64_IMM64:
+       case R_IA64_DIR32MSB:
+       case R_IA64_DIR32LSB:
+       case R_IA64_DIR64MSB:
+       case R_IA64_DIR64LSB:
+         /* Shared objects will always need at least a REL relocation.  */
+         if (info->shared || maybe_dynamic)
+           need_entry = NEED_DYNREL;
+         break;
+
+       case R_IA64_PCREL22:
+       case R_IA64_PCREL64I:
+       case R_IA64_PCREL32MSB:
+       case R_IA64_PCREL32LSB:
+       case R_IA64_PCREL64MSB:
+       case R_IA64_PCREL64LSB:
+         if (maybe_dynamic)
+           need_entry = NEED_DYNREL;
+         break;
+       }
+
+      if (!need_entry)
+       continue;
+
+      if ((need_entry & NEED_FPTR) != 0
+         && rel->r_addend)
+       {
+         (*info->callbacks->warning)
+           (info, _("non-zero addend in @fptr reloc"), 0,
+            abfd, 0, (bfd_vma) 0);
+       }
+
+      if (get_dyn_sym_info (ia64_info, h, abfd, rel, TRUE) == NULL)
+       return FALSE;
+    }
+
+  /* Now, we only do lookup without insertion, which is very fast
+     with the modified get_dyn_sym_info.  */
+  for (rel = relocs; rel < relend; ++rel)
+    {
+      struct elf64_ia64_dyn_sym_info *dyn_i;
+      int dynrel_type = R_IA64_NONE;
+
+      r_symndx = ELF64_R_SYM (rel->r_info);
+      if (r_symndx >= symtab_hdr->sh_info)
+       {
+         /* We're dealing with a global symbol -- find its hash entry
+            and mark it as being referenced.  */
+         long indx = r_symndx - symtab_hdr->sh_info;
+         h = elf_sym_hashes (abfd)[indx];
+         while (h->root.type == bfd_link_hash_indirect
+                || h->root.type == bfd_link_hash_warning)
+           h = (struct elf_link_hash_entry *) h->root.u.i.link;
+
+         h->ref_regular = 1;
+       }
+      else
+       h = NULL;
+
+      /* We can only get preliminary data on whether a symbol is
+        locally or externally defined, as not all of the input files
+        have yet been processed.  Do something with what we know, as
+        this may help reduce memory usage and processing time later.  */
+      maybe_dynamic = (h && ((!info->executable
+                             && (!SYMBOLIC_BIND (info, h)
+                                 || info->unresolved_syms_in_shared_libs == RM_IGNORE))
+                            || !h->def_regular
+                            || h->root.type == bfd_link_hash_defweak));
+
+      need_entry = 0;
+      switch (ELF64_R_TYPE (rel->r_info))
+       {
+       case R_IA64_TPREL64MSB:
+       case R_IA64_TPREL64LSB:
+       case R_IA64_LTOFF_TPREL22:
+       case R_IA64_DTPREL32MSB:
+       case R_IA64_DTPREL32LSB:
+       case R_IA64_DTPREL64MSB:
+       case R_IA64_DTPREL64LSB:
+       case R_IA64_LTOFF_DTPREL22:
+       case R_IA64_DTPMOD64MSB:
+       case R_IA64_DTPMOD64LSB:
+       case R_IA64_LTOFF_DTPMOD22:
+          abort ();
+         break;
+
+       case R_IA64_LTOFF_FPTR22:
+       case R_IA64_LTOFF_FPTR64I:
+       case R_IA64_LTOFF_FPTR32MSB:
+       case R_IA64_LTOFF_FPTR32LSB:
+       case R_IA64_LTOFF_FPTR64MSB:
+       case R_IA64_LTOFF_FPTR64LSB:
+         need_entry = NEED_FPTR | NEED_GOT | NEED_LTOFF_FPTR;
+         break;
+
+       case R_IA64_FPTR64I:
+       case R_IA64_FPTR32MSB:
+       case R_IA64_FPTR32LSB:
+       case R_IA64_FPTR64MSB:
+       case R_IA64_FPTR64LSB:
+         if (info->shared || h)
+           need_entry = NEED_FPTR | NEED_DYNREL;
+         else
+           need_entry = NEED_FPTR;
+         dynrel_type = R_IA64_FPTR64LSB;
+         break;
+
+       case R_IA64_LTOFF22:
+       case R_IA64_LTOFF64I:
+         need_entry = NEED_GOT;
+         break;
+
+       case R_IA64_LTOFF22X:
+         need_entry = NEED_GOTX;
+         break;
+
+       case R_IA64_PLTOFF22:
+       case R_IA64_PLTOFF64I:
+       case R_IA64_PLTOFF64MSB:
+       case R_IA64_PLTOFF64LSB:
+         need_entry = NEED_PLTOFF;
+         if (h)
+           {
+             if (maybe_dynamic)
+               need_entry |= NEED_MIN_PLT;
+           }
+         break;
+
+       case R_IA64_PCREL21B:
+        case R_IA64_PCREL60B:
+         /* Depending on where this symbol is defined, we may or may not
+            need a full plt entry.  Only skip if we know we'll not need
+            the entry -- static or symbolic, and the symbol definition
+            has already been seen.  */
+         if (maybe_dynamic && rel->r_addend == 0)
+           need_entry = NEED_FULL_PLT;
+         break;
+
+       case R_IA64_IMM14:
+       case R_IA64_IMM22:
+       case R_IA64_IMM64:
+       case R_IA64_DIR32MSB:
+       case R_IA64_DIR32LSB:
+       case R_IA64_DIR64MSB:
+       case R_IA64_DIR64LSB:
+         /* Shared objects will always need at least a REL relocation.  */
+         if (info->shared || maybe_dynamic)
+           need_entry = NEED_DYNREL;
+         dynrel_type = R_IA64_DIR64LSB;
+         break;
+
+       case R_IA64_IPLTMSB:
+       case R_IA64_IPLTLSB:
+         break;
+
+       case R_IA64_PCREL22:
+       case R_IA64_PCREL64I:
+       case R_IA64_PCREL32MSB:
+       case R_IA64_PCREL32LSB:
+       case R_IA64_PCREL64MSB:
+       case R_IA64_PCREL64LSB:
+         if (maybe_dynamic)
+           need_entry = NEED_DYNREL;
+         dynrel_type = R_IA64_PCREL64LSB;
+         break;
+       }
+
+      if (!need_entry)
+       continue;
+
+      dyn_i = get_dyn_sym_info (ia64_info, h, abfd, rel, FALSE);
+
+      /* Record whether or not this is a local symbol.  */
+      dyn_i->h = h;
+
+      /* Create what's needed.  */
+      if (need_entry & (NEED_GOT | NEED_GOTX))
+       {
+         if (!got)
+           {
+             got = get_got (abfd, ia64_info);
+             if (!got)
+               return FALSE;
+           }
+         if (need_entry & NEED_GOT)
+           dyn_i->want_got = 1;
+         if (need_entry & NEED_GOTX)
+           dyn_i->want_gotx = 1;
+       }
+      if (need_entry & NEED_FPTR)
+       {
+          /* Create the .opd section.  */
+         if (!fptr)
+           {
+             fptr = get_fptr (abfd, info, ia64_info);
+             if (!fptr)
+               return FALSE;
+           }
+         dyn_i->want_fptr = 1;
+       }
+      if (need_entry & NEED_LTOFF_FPTR)
+       dyn_i->want_ltoff_fptr = 1;
+      if (need_entry & (NEED_MIN_PLT | NEED_FULL_PLT))
+       {
+          if (!ia64_info->root.dynobj)
+           ia64_info->root.dynobj = abfd;
+         h->needs_plt = 1;
+         dyn_i->want_plt = 1;
+       }
+      if (need_entry & NEED_FULL_PLT)
+       dyn_i->want_plt2 = 1;
+      if (need_entry & NEED_PLTOFF)
+       {
+         /* This is needed here, in case @pltoff is used in a non-shared
+            link.  */
+         if (!pltoff)
+           {
+             pltoff = get_pltoff (abfd, ia64_info);
+             if (!pltoff)
+               return FALSE;
+           }
+
+         dyn_i->want_pltoff = 1;
+       }
+      if ((need_entry & NEED_DYNREL) && (sec->flags & SEC_ALLOC))
+       {
+         if (!srel)
+           {
+             srel = get_reloc_section (abfd, ia64_info, sec, TRUE);
+             if (!srel)
+               return FALSE;
+           }
+         if (!count_dyn_reloc (abfd, dyn_i, srel, dynrel_type))
+           return FALSE;
+       }
+    }
+
+  return TRUE;
+}
+
+/* For cleanliness, and potentially faster dynamic loading, allocate
+   external GOT entries first.  */
+
+static bfd_boolean
+allocate_global_data_got (struct elf64_ia64_dyn_sym_info *dyn_i,
+                         void * data)
+{
+  struct elf64_ia64_allocate_data *x = (struct elf64_ia64_allocate_data *)data;
+
+  if ((dyn_i->want_got || dyn_i->want_gotx)
+      && ! dyn_i->want_fptr
+      && elf64_ia64_dynamic_symbol_p (dyn_i->h))
+     {
+       /* GOT entry with FPTR is done by allocate_global_fptr_got.  */
+       dyn_i->got_offset = x->ofs;
+       x->ofs += 8;
+     }
+  return TRUE;
+}
+
+/* Next, allocate all the GOT entries used by LTOFF_FPTR relocs.  */
+
+static bfd_boolean
+allocate_global_fptr_got (struct elf64_ia64_dyn_sym_info *dyn_i,
+                         void * data)
+{
+  struct elf64_ia64_allocate_data *x = (struct elf64_ia64_allocate_data *)data;
+
+  if (dyn_i->want_got
+      && dyn_i->want_fptr
+      && elf64_ia64_dynamic_symbol_p (dyn_i->h))
+    {
+      dyn_i->got_offset = x->ofs;
+      x->ofs += 8;
+    }
+  return TRUE;
+}
+
+/* Lastly, allocate all the GOT entries for local data.  */
+
+static bfd_boolean
+allocate_local_got (struct elf64_ia64_dyn_sym_info *dyn_i,
+                   void * data)
+{
+  struct elf64_ia64_allocate_data *x = (struct elf64_ia64_allocate_data *) data;
+
+  if ((dyn_i->want_got || dyn_i->want_gotx)
+      && !elf64_ia64_dynamic_symbol_p (dyn_i->h))
+    {
+      dyn_i->got_offset = x->ofs;
+      x->ofs += 8;
+    }
+  return TRUE;
+}
+
+/* Allocate function descriptors.  We can do these for every function
+   in a main executable that is not exported.  */
+
+static bfd_boolean
+allocate_fptr (struct elf64_ia64_dyn_sym_info *dyn_i, void * data)
+{
+  struct elf64_ia64_allocate_data *x = (struct elf64_ia64_allocate_data *) data;
+
+  if (dyn_i->want_fptr)
+    {
+      struct elf_link_hash_entry *h = dyn_i->h;
+
+      if (h)
+       while (h->root.type == bfd_link_hash_indirect
+              || h->root.type == bfd_link_hash_warning)
+         h = (struct elf_link_hash_entry *) h->root.u.i.link;
+
+      if (h == NULL || !h->def_dynamic)
+       {
+          /*  A non dynamic symbol.  */
+         dyn_i->fptr_offset = x->ofs;
+         x->ofs += 16;
+       }
+      else
+       dyn_i->want_fptr = 0;
+    }
+  return TRUE;
+}
+
+/* Allocate all the minimal PLT entries.  */
+
+static bfd_boolean
+allocate_plt_entries (struct elf64_ia64_dyn_sym_info *dyn_i,
+                     void * data ATTRIBUTE_UNUSED)
+{
+  if (dyn_i->want_plt)
+    {
+      struct elf_link_hash_entry *h = dyn_i->h;
+
+      if (h)
+       while (h->root.type == bfd_link_hash_indirect
+              || h->root.type == bfd_link_hash_warning)
+         h = (struct elf_link_hash_entry *) h->root.u.i.link;
+
+      /* ??? Versioned symbols seem to lose NEEDS_PLT.  */
+      if (elf64_ia64_dynamic_symbol_p (h))
+       {
+         dyn_i->want_pltoff = 1;
+       }
+      else
+       {
+         dyn_i->want_plt = 0;
+         dyn_i->want_plt2 = 0;
+       }
+    }
+  return TRUE;
+}
+
+/* Allocate all the full PLT entries.  */
+
+static bfd_boolean
+allocate_plt2_entries (struct elf64_ia64_dyn_sym_info *dyn_i,
+                      void * data)
+{
+  struct elf64_ia64_allocate_data *x = (struct elf64_ia64_allocate_data *)data;
+
+  if (dyn_i->want_plt2)
+    {
+      struct elf_link_hash_entry *h = dyn_i->h;
+      bfd_size_type ofs = x->ofs;
+
+      dyn_i->plt2_offset = ofs;
+      x->ofs = ofs + PLT_FULL_ENTRY_SIZE;
+
+      while (h->root.type == bfd_link_hash_indirect
+            || h->root.type == bfd_link_hash_warning)
+       h = (struct elf_link_hash_entry *) h->root.u.i.link;
+      dyn_i->h->plt.offset = ofs;
+    }
+  return TRUE;
+}
+
+/* Allocate all the PLTOFF entries requested by relocations and
+   plt entries.  We can't share space with allocated FPTR entries,
+   because the latter are not necessarily addressable by the GP.
+   ??? Relaxation might be able to determine that they are.  */
+
+static bfd_boolean
+allocate_pltoff_entries (struct elf64_ia64_dyn_sym_info *dyn_i,
+                        void * data)
+{
+  struct elf64_ia64_allocate_data *x = (struct elf64_ia64_allocate_data *)data;
+
+  if (dyn_i->want_pltoff)
+    {
+      dyn_i->pltoff_offset = x->ofs;
+      x->ofs += 16;
+    }
+  return TRUE;
+}
+
+/* Allocate dynamic relocations for those symbols that turned out
+   to be dynamic.  */
+
+static bfd_boolean
+allocate_dynrel_entries (struct elf64_ia64_dyn_sym_info *dyn_i,
+                        void * data)
+{
+  struct elf64_ia64_allocate_data *x = (struct elf64_ia64_allocate_data *)data;
+  struct elf64_ia64_link_hash_table *ia64_info;
+  struct elf64_ia64_dyn_reloc_entry *rent;
+  bfd_boolean dynamic_symbol, shared, resolved_zero;
+  struct elf64_ia64_link_hash_entry *h_ia64;
+
+  ia64_info = elf64_ia64_hash_table (x->info);
+  if (ia64_info == NULL)
+    return FALSE;
+
+  /* Note that this can't be used in relation to FPTR relocs below.  */
+  dynamic_symbol = elf64_ia64_dynamic_symbol_p (dyn_i->h);
+
+  shared = x->info->shared;
+  resolved_zero = (dyn_i->h
+                  && ELF_ST_VISIBILITY (dyn_i->h->other)
+                  && dyn_i->h->root.type == bfd_link_hash_undefweak);
+
+  /* Take care of the GOT and PLT relocations.  */
+
+  if ((!resolved_zero
+       && (dynamic_symbol || shared)
+       && (dyn_i->want_got || dyn_i->want_gotx))
+      || (dyn_i->want_ltoff_fptr
+         && dyn_i->h
+         && dyn_i->h->def_dynamic))
+    {
+      /* VMS: FIX64.  */
+      if (dyn_i->h != NULL && dyn_i->h->def_dynamic)
+        {
+          h_ia64 = (struct elf64_ia64_link_hash_entry *) dyn_i->h;
+          elf_ia64_vms_tdata (h_ia64->shl)->fixups_off +=
+            sizeof (Elf64_External_VMS_IMAGE_FIXUP);
+          ia64_info->fixups_sec->size +=
+            sizeof (Elf64_External_VMS_IMAGE_FIXUP);
+        }
+    }
+
+  if (ia64_info->rel_fptr_sec && dyn_i->want_fptr)
+    {
+      /* VMS: only image reloc.  */
+      if (dyn_i->h == NULL || dyn_i->h->root.type != bfd_link_hash_undefweak)
+       ia64_info->rel_fptr_sec->size += sizeof (Elf64_External_Rela);
+    }
+
+  if (!resolved_zero && dyn_i->want_pltoff)
+    {
+      /* VMS: FIXFD.  */
+      if (dyn_i->h != NULL && dyn_i->h->def_dynamic)
+        {
+          h_ia64 = (struct elf64_ia64_link_hash_entry *) dyn_i->h;
+          elf_ia64_vms_tdata (h_ia64->shl)->fixups_off +=
+            sizeof (Elf64_External_VMS_IMAGE_FIXUP);
+          ia64_info->fixups_sec->size +=
+            sizeof (Elf64_External_VMS_IMAGE_FIXUP);
+        }
+    }
+
+  /* Take care of the normal data relocations.  */
+
+  for (rent = dyn_i->reloc_entries; rent; rent = rent->next)
+    {
+      int count = rent->count;
+
+      switch (rent->type)
+       {
+       case R_IA64_FPTR32LSB:
+       case R_IA64_FPTR64LSB:
+         /* Allocate one iff !want_fptr and not PIE, which by this point
+            will be true only if we're actually allocating one statically
+            in the main executable.  Position independent executables
+            need a relative reloc.  */
+         if (dyn_i->want_fptr && !x->info->pie)
+           continue;
+         break;
+       case R_IA64_PCREL32LSB:
+       case R_IA64_PCREL64LSB:
+         if (!dynamic_symbol)
+           continue;
+         break;
+       case R_IA64_DIR32LSB:
+       case R_IA64_DIR64LSB:
+         if (!dynamic_symbol && !shared)
+           continue;
+         break;
+       case R_IA64_IPLTLSB:
+         if (!dynamic_symbol && !shared)
+           continue;
+         /* Use two REL relocations for IPLT relocations
+            against local symbols.  */
+         if (!dynamic_symbol)
+           count *= 2;
+         break;
+       case R_IA64_DTPREL32LSB:
+       case R_IA64_TPREL64LSB:
+       case R_IA64_DTPREL64LSB:
+       case R_IA64_DTPMOD64LSB:
+         break;
+       default:
+         abort ();
+       }
+
+      /* Add a fixup.  */
+      if (!dynamic_symbol)
+        abort ();
+
+      h_ia64 = (struct elf64_ia64_link_hash_entry *) dyn_i->h;
+      elf_ia64_vms_tdata (h_ia64->shl)->fixups_off +=
+        sizeof (Elf64_External_VMS_IMAGE_FIXUP);
+      ia64_info->fixups_sec->size +=
+        sizeof (Elf64_External_VMS_IMAGE_FIXUP);
+    }
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_ia64_adjust_dynamic_symbol (struct bfd_link_info *info ATTRIBUTE_UNUSED,
+                                 struct elf_link_hash_entry *h)
+{
+  /* ??? Undefined symbols with PLT entries should be re-defined
+     to be the PLT entry.  */
+
+  /* If this is a weak symbol, and there is a real definition, the
+     processor independent code will have arranged for us to see the
+     real definition first, and we can just use the same value.  */
+  if (h->u.weakdef != NULL)
+    {
+      BFD_ASSERT (h->u.weakdef->root.type == bfd_link_hash_defined
+                  || h->u.weakdef->root.type == bfd_link_hash_defweak);
+      h->root.u.def.section = h->u.weakdef->root.u.def.section;
+      h->root.u.def.value = h->u.weakdef->root.u.def.value;
+      return TRUE;
+    }
+
+  /* If this is a reference to a symbol defined by a dynamic object which
+     is not a function, we might allocate the symbol in our .dynbss section
+     and allocate a COPY dynamic relocation.
+
+     But IA-64 code is canonically PIC, so as a rule we can avoid this sort
+     of hackery.  */
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_ia64_size_dynamic_sections (bfd *output_bfd ATTRIBUTE_UNUSED,
+                                 struct bfd_link_info *info)
+{
+  struct elf64_ia64_allocate_data data;
+  struct elf64_ia64_link_hash_table *ia64_info;
+  asection *sec;
+  bfd *dynobj;
+  struct elf_link_hash_table *hash_table;
+
+  hash_table = elf_hash_table (info);
+  dynobj = hash_table->dynobj;
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return FALSE;
+  BFD_ASSERT(dynobj != NULL);
+  data.info = info;
+
+  /* Allocate the GOT entries.  */
+
+  if (ia64_info->root.sgot)
+    {
+      data.ofs = 0;
+      elf64_ia64_dyn_sym_traverse (ia64_info, allocate_global_data_got, &data);
+      elf64_ia64_dyn_sym_traverse (ia64_info, allocate_global_fptr_got, &data);
+      elf64_ia64_dyn_sym_traverse (ia64_info, allocate_local_got, &data);
+      ia64_info->root.sgot->size = data.ofs;
+    }
+
+  /* Allocate the FPTR entries.  */
+
+  if (ia64_info->fptr_sec)
+    {
+      data.ofs = 0;
+      elf64_ia64_dyn_sym_traverse (ia64_info, allocate_fptr, &data);
+      ia64_info->fptr_sec->size = data.ofs;
+    }
+
+  /* Now that we've seen all of the input files, we can decide which
+     symbols need plt entries.  Allocate the minimal PLT entries first.
+     We do this even though dynamic_sections_created may be FALSE, because
+     this has the side-effect of clearing want_plt and want_plt2.  */
+
+  data.ofs = 0;
+  elf64_ia64_dyn_sym_traverse (ia64_info, allocate_plt_entries, &data);
+
+  /* Align the pointer for the plt2 entries.  */
+  data.ofs = (data.ofs + 31) & (bfd_vma) -32;
+
+  elf64_ia64_dyn_sym_traverse (ia64_info, allocate_plt2_entries, &data);
+  if (data.ofs != 0 || ia64_info->root.dynamic_sections_created)
+    {
+      /* FIXME: we always reserve the memory for dynamic linker even if
+        there are no PLT entries since dynamic linker may assume the
+        reserved memory always exists.  */
+
+      BFD_ASSERT (ia64_info->root.dynamic_sections_created);
+
+      ia64_info->root.splt->size = data.ofs;
+    }
+
+  /* Allocate the PLTOFF entries.  */
+
+  if (ia64_info->pltoff_sec)
+    {
+      data.ofs = 0;
+      elf64_ia64_dyn_sym_traverse (ia64_info, allocate_pltoff_entries, &data);
+      ia64_info->pltoff_sec->size = data.ofs;
+    }
+
+  if (ia64_info->root.dynamic_sections_created)
+    {
+      /* Allocate space for the dynamic relocations that turned out to be
+        required.  */
+      elf64_ia64_dyn_sym_traverse (ia64_info, allocate_dynrel_entries, &data);
+    }
+
+  /* We have now determined the sizes of the various dynamic sections.
+     Allocate memory for them.  */
+  for (sec = dynobj->sections; sec != NULL; sec = sec->next)
+    {
+      bfd_boolean strip;
+
+      if (!(sec->flags & SEC_LINKER_CREATED))
+       continue;
+
+      /* If we don't need this section, strip it from the output file.
+        There were several sections primarily related to dynamic
+        linking that must be create before the linker maps input
+        sections to output sections.  The linker does that before
+        bfd_elf_size_dynamic_sections is called, and it is that
+        function which decides whether anything needs to go into
+        these sections.  */
+
+      strip = (sec->size == 0);
+
+      if (sec == ia64_info->root.sgot)
+       strip = FALSE;
+      else if (sec == ia64_info->root.srelgot)
+       {
+         if (strip)
+           ia64_info->root.srelgot = NULL;
+         else
+           /* We use the reloc_count field as a counter if we need to
+              copy relocs into the output file.  */
+           sec->reloc_count = 0;
+       }
+      else if (sec == ia64_info->fptr_sec)
+       {
+         if (strip)
+           ia64_info->fptr_sec = NULL;
+       }
+      else if (sec == ia64_info->rel_fptr_sec)
+       {
+         if (strip)
+           ia64_info->rel_fptr_sec = NULL;
+         else
+           /* We use the reloc_count field as a counter if we need to
+              copy relocs into the output file.  */
+           sec->reloc_count = 0;
+       }
+      else if (sec == ia64_info->root.splt)
+       {
+         if (strip)
+           ia64_info->root.splt = NULL;
+       }
+      else if (sec == ia64_info->pltoff_sec)
+       {
+         if (strip)
+           ia64_info->pltoff_sec = NULL;
+       }
+      else if (sec == ia64_info->fixups_sec)
+       {
+          if (strip)
+            ia64_info->fixups_sec = NULL;
+       }
+      else if (sec == ia64_info->transfer_sec)
+        {
+          ;
+        }
+      else
+       {
+         const char *name;
+
+         /* It's OK to base decisions on the section name, because none
+            of the dynobj section names depend upon the input files.  */
+         name = bfd_get_section_name (dynobj, sec);
+
+         if (strcmp (name, ".got.plt") == 0)
+           strip = FALSE;
+         else if (CONST_STRNEQ (name, ".rel"))
+           {
+             if (!strip)
+               {
+                 /* We use the reloc_count field as a counter if we need to
+                    copy relocs into the output file.  */
+                 sec->reloc_count = 0;
+               }
+           }
+         else
+           continue;
+       }
+
+      if (strip)
+       sec->flags |= SEC_EXCLUDE;
+      else
+       {
+         /* Allocate memory for the section contents.  */
+         sec->contents = (bfd_byte *) bfd_zalloc (dynobj, sec->size);
+         if (sec->contents == NULL && sec->size != 0)
+           return FALSE;
+       }
+    }
+
+  if (elf_hash_table (info)->dynamic_sections_created)
+    {
+      bfd *abfd;
+      asection *dynsec;
+      asection *dynstrsec;
+      Elf_Internal_Dyn dyn;
+      const struct elf_backend_data *bed;
+      unsigned int shl_num = 0;
+      bfd_vma fixups_off = 0;
+      bfd_vma strdyn_off;
+      unsigned int time_hi, time_lo;
+
+      /* The .dynamic section must exist and be empty.  */
+      dynsec = bfd_get_section_by_name (hash_table->dynobj, ".dynamic");
+      BFD_ASSERT (dynsec != NULL);
+      BFD_ASSERT (dynsec->size == 0);
+
+      dynstrsec = bfd_get_section_by_name (hash_table->dynobj, ".vmsdynstr");
+      BFD_ASSERT (dynstrsec != NULL);
+      BFD_ASSERT (dynstrsec->size == 0);
+      dynstrsec->size = 1;     /* Initial blank.  */
+
+      /* Ident + link time.  */
+      vms_get_time (&time_hi, &time_lo);
+
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_IDENT, 0))
+        return FALSE;
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_LINKTIME,
+                                       (((bfd_uint64_t)time_hi) << 32)
+                                       + time_lo))
+        return FALSE;
+
+      /* Strtab.  */
+      strdyn_off = dynsec->size;
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_STRTAB_OFFSET, 0))
+        return FALSE;
+      if (!_bfd_elf_add_dynamic_entry (info, DT_STRSZ, 0))
+        return FALSE;
+
+      /* PLTGOT  */
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_PLTGOT_SEG, 0))
+        return FALSE;
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_PLTGOT_OFFSET, 0))
+        return FALSE;
+
+      /* Misc.  */
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_FPMODE, 0x9800000))
+        return FALSE;
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_LNKFLAGS,
+                                       VMS_LF_IMGSTA | VMS_LF_MAIN))
+        return FALSE;
+
+      /* Add entries for shared libraries.  */
+      for (abfd = info->input_bfds; abfd; abfd = abfd->link_next)
+        {
+          char *soname;
+          size_t soname_len;
+          bfd_size_type strindex;
+          bfd_byte *newcontents;
+          bfd_vma fixups_shl_off;
+
+          if (!(abfd->flags & DYNAMIC))
+            continue;
+          BFD_ASSERT (abfd->xvec == output_bfd->xvec);
+
+          if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_NEEDED_IDENT,
+                                           elf_ia64_vms_ident (abfd)))
+            return FALSE;
+
+          soname = vms_get_module_name (abfd->filename, TRUE);
+          if (soname == NULL)
+            return FALSE;
+          strindex = dynstrsec->size;
+          soname_len = strlen (soname) + 1;
+          newcontents = (bfd_byte *) bfd_realloc (dynstrsec->contents,
+                                                  strindex + soname_len);
+          if (newcontents == NULL)
+            return FALSE;
+          memcpy (newcontents + strindex, soname, soname_len);
+          dynstrsec->size += soname_len;
+          dynstrsec->contents = newcontents;
+
+          if (!_bfd_elf_add_dynamic_entry (info, DT_NEEDED, strindex))
+            return FALSE;
+
+          if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_FIXUP_NEEDED,
+                                           shl_num))
+            return FALSE;
+          shl_num++;
+
+          /* The fixups_off was in fact containing the size of the fixup
+             section.  Remap into the offset.  */
+          fixups_shl_off = elf_ia64_vms_tdata (abfd)->fixups_off;
+          elf_ia64_vms_tdata (abfd)->fixups_off = fixups_off;
+
+          if (!_bfd_elf_add_dynamic_entry
+              (info, DT_IA_64_VMS_FIXUP_RELA_CNT,
+               fixups_shl_off / sizeof (Elf64_External_VMS_IMAGE_FIXUP)))
+            return FALSE;
+          if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_FIXUP_RELA_OFF,
+                                           fixups_off))
+            return FALSE;
+          fixups_off += fixups_shl_off;
+        }
+
+      /* Unwind.  */
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_UNWINDSZ, 0))
+        return FALSE;
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_UNWIND_CODSEG, 0))
+        return FALSE;
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_UNWIND_INFOSEG, 0))
+        return FALSE;
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_UNWIND_OFFSET, 0))
+        return FALSE;
+      if (!_bfd_elf_add_dynamic_entry (info, DT_IA_64_VMS_UNWIND_SEG, 0))
+        return FALSE;
+
+      if (!_bfd_elf_add_dynamic_entry (info, DT_NULL, 0xdead))
+            return FALSE;
+
+      /* Fix the strtab entries.  */
+      bed = get_elf_backend_data (hash_table->dynobj);
+
+      if (dynstrsec->size > 1)
+        dynstrsec->contents[0] = 0;
+      else
+        dynstrsec->size = 0;
+
+      /* Note: one 'spare' (ie DT_NULL) entry is added by
+         bfd_elf_size_dynsym_hash_dynstr.  */
+      dyn.d_tag = DT_IA_64_VMS_STRTAB_OFFSET;
+      dyn.d_un.d_val = dynsec->size /* + sizeof (Elf64_External_Dyn) */;
+      bed->s->swap_dyn_out (hash_table->dynobj, &dyn,
+                            dynsec->contents + strdyn_off);
+
+      dyn.d_tag = DT_STRSZ;
+      dyn.d_un.d_val = dynstrsec->size;
+      bed->s->swap_dyn_out (hash_table->dynobj, &dyn,
+                            dynsec->contents + strdyn_off + bed->s->sizeof_dyn);
+
+      elf_ia64_vms_tdata (output_bfd)->needed_count = shl_num;
+
+      /* Note section.  */
+      if (!create_ia64_vms_notes (output_bfd, info, time_hi, time_lo))
+        return FALSE;
+    }
+
+  /* ??? Perhaps force __gp local.  */
+
+  return TRUE;
+}
+
+static void
+elf64_ia64_install_fixup (bfd *output_bfd,
+                          struct elf64_ia64_link_hash_table *ia64_info,
+                          struct elf_link_hash_entry *h,
+                          unsigned int type, asection *sec, bfd_vma offset,
+                          bfd_vma addend)
+{
+  asection *relsec;
+  Elf64_External_VMS_IMAGE_FIXUP *fixup;
+  struct elf64_ia64_link_hash_entry *h_ia64;
+  bfd_vma fixoff;
+  Elf_Internal_Phdr *phdr;
+
+  if (h == NULL || !h->def_dynamic)
+    abort ();
+
+  h_ia64 = (struct elf64_ia64_link_hash_entry *) h;
+  fixoff = elf_ia64_vms_tdata (h_ia64->shl)->fixups_off;
+  elf_ia64_vms_tdata (h_ia64->shl)->fixups_off +=
+    sizeof (Elf64_External_VMS_IMAGE_FIXUP);
+  relsec = ia64_info->fixups_sec;
+
+  fixup = (Elf64_External_VMS_IMAGE_FIXUP *)(relsec->contents + fixoff);
+  offset += sec->output_section->vma + sec->output_offset;
+
+  /* FIXME: this is slow.  We should cache the last one used, or create a
+     map.  */
+  phdr = _bfd_elf_find_segment_containing_section
+    (output_bfd, sec->output_section);
+  BFD_ASSERT (phdr != NULL);
+
+  bfd_putl64 (offset - phdr->p_vaddr, fixup->fixup_offset);
+  bfd_putl32 (type, fixup->type);
+  bfd_putl32 (phdr - elf_tdata (output_bfd)->phdr, fixup->fixup_seg);
+  bfd_putl64 (addend, fixup->addend);
+  bfd_putl32 (h->root.u.def.value, fixup->symvec_index);
+  bfd_putl32 (2, fixup->data_type);
+}
+
+/* Store an entry for target address TARGET_ADDR in the linkage table
+   and return the gp-relative address of the linkage table entry.  */
+
+static bfd_vma
+set_got_entry (bfd *abfd, struct bfd_link_info *info,
+              struct elf64_ia64_dyn_sym_info *dyn_i,
+              bfd_vma addend, bfd_vma value, unsigned int dyn_r_type)
+{
+  struct elf64_ia64_link_hash_table *ia64_info;
+  asection *got_sec;
+  bfd_boolean done;
+  bfd_vma got_offset;
+
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return 0;
+
+  got_sec = ia64_info->root.sgot;
+
+  switch (dyn_r_type)
+    {
+    case R_IA64_TPREL64LSB:
+    case R_IA64_DTPMOD64LSB:
+    case R_IA64_DTPREL32LSB:
+    case R_IA64_DTPREL64LSB:
+      abort ();
+      break;
+    default:
+      done = dyn_i->got_done;
+      dyn_i->got_done = TRUE;
+      got_offset = dyn_i->got_offset;
+      break;
+    }
+
+  BFD_ASSERT ((got_offset & 7) == 0);
+
+  if (! done)
+    {
+      /* Store the target address in the linkage table entry.  */
+      bfd_put_64 (abfd, value, got_sec->contents + got_offset);
+
+      /* Install a dynamic relocation if needed.  */
+      if (((info->shared
+           && (!dyn_i->h
+               || ELF_ST_VISIBILITY (dyn_i->h->other) == STV_DEFAULT
+               || dyn_i->h->root.type != bfd_link_hash_undefweak))
+           || elf64_ia64_dynamic_symbol_p (dyn_i->h))
+         && (!dyn_i->want_ltoff_fptr
+             || !info->pie
+             || !dyn_i->h
+             || dyn_i->h->root.type != bfd_link_hash_undefweak))
+       {
+         if (!dyn_i->h || !dyn_i->h->def_dynamic)
+           {
+             dyn_r_type = R_IA64_REL64LSB;
+             addend = value;
+           }
+
+          /* VMS: install a FIX32 or FIX64.  */
+          switch (dyn_r_type)
+            {
+            case R_IA64_DIR32LSB:
+            case R_IA64_FPTR32LSB:
+              dyn_r_type = R_IA64_VMS_FIX32;
+              break;
+            case R_IA64_DIR64LSB:
+            case R_IA64_FPTR64LSB:
+              dyn_r_type = R_IA64_VMS_FIX64;
+              break;
+            default:
+              BFD_ASSERT (FALSE);
+              break;
+            }
+          elf64_ia64_install_fixup
+            (info->output_bfd, ia64_info, dyn_i->h,
+             dyn_r_type, got_sec, got_offset, addend);
+        }
+    }
+
+  /* Return the address of the linkage table entry.  */
+  value = (got_sec->output_section->vma
+          + got_sec->output_offset
+          + got_offset);
+
+  return value;
+}
+
+/* Fill in a function descriptor consisting of the function's code
+   address and its global pointer.  Return the descriptor's address.  */
+
+static bfd_vma
+set_fptr_entry (bfd *abfd, struct bfd_link_info *info,
+               struct elf64_ia64_dyn_sym_info *dyn_i,
+               bfd_vma value)
+{
+  struct elf64_ia64_link_hash_table *ia64_info;
+  asection *fptr_sec;
+
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return 0;
+
+  fptr_sec = ia64_info->fptr_sec;
+
+  if (!dyn_i->fptr_done)
+    {
+      dyn_i->fptr_done = 1;
+
+      /* Fill in the function descriptor.  */
+      bfd_put_64 (abfd, value, fptr_sec->contents + dyn_i->fptr_offset);
+      bfd_put_64 (abfd, _bfd_get_gp_value (abfd),
+                 fptr_sec->contents + dyn_i->fptr_offset + 8);
+    }
+
+  /* Return the descriptor's address.  */
+  value = (fptr_sec->output_section->vma
+          + fptr_sec->output_offset
+          + dyn_i->fptr_offset);
+
+  return value;
+}
+
+/* Fill in a PLTOFF entry consisting of the function's code address
+   and its global pointer.  Return the descriptor's address.  */
+
+static bfd_vma
+set_pltoff_entry (bfd *abfd, struct bfd_link_info *info,
+                 struct elf64_ia64_dyn_sym_info *dyn_i,
+                 bfd_vma value, bfd_boolean is_plt)
+{
+  struct elf64_ia64_link_hash_table *ia64_info;
+  asection *pltoff_sec;
+
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return 0;
+
+  pltoff_sec = ia64_info->pltoff_sec;
+
+  /* Don't do anything if this symbol uses a real PLT entry.  In
+     that case, we'll fill this in during finish_dynamic_symbol.  */
+  if ((! dyn_i->want_plt || is_plt)
+      && !dyn_i->pltoff_done)
+    {
+      bfd_vma gp = _bfd_get_gp_value (abfd);
+
+      /* Fill in the function descriptor.  */
+      bfd_put_64 (abfd, value, pltoff_sec->contents + dyn_i->pltoff_offset);
+      bfd_put_64 (abfd, gp, pltoff_sec->contents + dyn_i->pltoff_offset + 8);
+
+      /* Install dynamic relocations if needed.  */
+      if (!is_plt
+         && info->shared
+         && (!dyn_i->h
+             || ELF_ST_VISIBILITY (dyn_i->h->other) == STV_DEFAULT
+             || dyn_i->h->root.type != bfd_link_hash_undefweak))
+       {
+          /* VMS:  */
+          abort ();
+       }
+
+      dyn_i->pltoff_done = 1;
+    }
+
+  /* Return the descriptor's address.  */
+  value = (pltoff_sec->output_section->vma
+          + pltoff_sec->output_offset
+          + dyn_i->pltoff_offset);
+
+  return value;
+}
+
+/* Called through qsort to sort the .IA_64.unwind section during a
+   non-relocatable link.  Set elf64_ia64_unwind_entry_compare_bfd
+   to the output bfd so we can do proper endianness frobbing.  */
+
+static bfd *elf64_ia64_unwind_entry_compare_bfd;
+
+static int
+elf64_ia64_unwind_entry_compare (const void * a, const void * b)
+{
+  bfd_vma av, bv;
+
+  av = bfd_get_64 (elf64_ia64_unwind_entry_compare_bfd, a);
+  bv = bfd_get_64 (elf64_ia64_unwind_entry_compare_bfd, b);
+
+  return (av < bv ? -1 : av > bv ? 1 : 0);
+}
+
+/* Make sure we've got ourselves a nice fat __gp value.  */
+static bfd_boolean
+elf64_ia64_choose_gp (bfd *abfd, struct bfd_link_info *info, bfd_boolean final)
+{
+  bfd_vma min_vma = (bfd_vma) -1, max_vma = 0;
+  bfd_vma min_short_vma = min_vma, max_short_vma = 0;
+  struct elf_link_hash_entry *gp;
+  bfd_vma gp_val;
+  asection *os;
+  struct elf64_ia64_link_hash_table *ia64_info;
+
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return FALSE;
+
+  /* Find the min and max vma of all sections marked short.  Also collect
+     min and max vma of any type, for use in selecting a nice gp.  */
+  for (os = abfd->sections; os ; os = os->next)
+    {
+      bfd_vma lo, hi;
+
+      if ((os->flags & SEC_ALLOC) == 0)
+       continue;
+
+      lo = os->vma;
+      /* When this function is called from elfNN_ia64_final_link
+        the correct value to use is os->size.  When called from
+        elfNN_ia64_relax_section we are in the middle of section
+        sizing; some sections will already have os->size set, others
+        will have os->size zero and os->rawsize the previous size.  */
+      hi = os->vma + (!final && os->rawsize ? os->rawsize : os->size);
+      if (hi < lo)
+       hi = (bfd_vma) -1;
+
+      if (min_vma > lo)
+       min_vma = lo;
+      if (max_vma < hi)
+       max_vma = hi;
+      if (os->flags & SEC_SMALL_DATA)
+       {
+         if (min_short_vma > lo)
+           min_short_vma = lo;
+         if (max_short_vma < hi)
+           max_short_vma = hi;
+       }
+    }
+
+  if (ia64_info->min_short_sec)
+    {
+      if (min_short_vma
+         > (ia64_info->min_short_sec->vma
+            + ia64_info->min_short_offset))
+       min_short_vma = (ia64_info->min_short_sec->vma
+                        + ia64_info->min_short_offset);
+      if (max_short_vma
+         < (ia64_info->max_short_sec->vma
+            + ia64_info->max_short_offset))
+       max_short_vma = (ia64_info->max_short_sec->vma
+                        + ia64_info->max_short_offset);
+    }
+
+  /* See if the user wants to force a value.  */
+  gp = elf_link_hash_lookup (elf_hash_table (info), "__gp", FALSE,
+                            FALSE, FALSE);
+
+  if (gp
+      && (gp->root.type == bfd_link_hash_defined
+         || gp->root.type == bfd_link_hash_defweak))
+    {
+      asection *gp_sec = gp->root.u.def.section;
+      gp_val = (gp->root.u.def.value
+               + gp_sec->output_section->vma
+               + gp_sec->output_offset);
+    }
+  else
+    {
+      /* Pick a sensible value.  */
+
+      if (ia64_info->min_short_sec)
+       {
+         bfd_vma short_range = max_short_vma - min_short_vma;
+
+         /* If min_short_sec is set, pick one in the middle bewteen
+            min_short_vma and max_short_vma.  */
+         if (short_range >= 0x400000)
+           goto overflow;
+         gp_val = min_short_vma + short_range / 2;
+       }
+      else
+       {
+         asection *got_sec = ia64_info->root.sgot;
+
+         /* Start with just the address of the .got.  */
+         if (got_sec)
+           gp_val = got_sec->output_section->vma;
+         else if (max_short_vma != 0)
+           gp_val = min_short_vma;
+         else if (max_vma - min_vma < 0x200000)
+           gp_val = min_vma;
+         else
+           gp_val = max_vma - 0x200000 + 8;
+       }
+
+      /* If it is possible to address the entire image, but we
+        don't with the choice above, adjust.  */
+      if (max_vma - min_vma < 0x400000
+         && (max_vma - gp_val >= 0x200000
+             || gp_val - min_vma > 0x200000))
+       gp_val = min_vma + 0x200000;
+      else if (max_short_vma != 0)
+       {
+         /* If we don't cover all the short data, adjust.  */
+         if (max_short_vma - gp_val >= 0x200000)
+           gp_val = min_short_vma + 0x200000;
+
+         /* If we're addressing stuff past the end, adjust back.  */
+         if (gp_val > max_vma)
+           gp_val = max_vma - 0x200000 + 8;
+       }
+    }
+
+  /* Validate whether all SHF_IA_64_SHORT sections are within
+     range of the chosen GP.  */
+
+  if (max_short_vma != 0)
+    {
+      if (max_short_vma - min_short_vma >= 0x400000)
+       {
+overflow:
+         (*_bfd_error_handler)
+           (_("%s: short data segment overflowed (0x%lx >= 0x400000)"),
+            bfd_get_filename (abfd),
+            (unsigned long) (max_short_vma - min_short_vma));
+         return FALSE;
+       }
+      else if ((gp_val > min_short_vma
+               && gp_val - min_short_vma > 0x200000)
+              || (gp_val < max_short_vma
+                  && max_short_vma - gp_val >= 0x200000))
+       {
+         (*_bfd_error_handler)
+           (_("%s: __gp does not cover short data segment"),
+            bfd_get_filename (abfd));
+         return FALSE;
+       }
+    }
+
+  _bfd_set_gp_value (abfd, gp_val);
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_ia64_final_link (bfd *abfd, struct bfd_link_info *info)
+{
+  struct elf64_ia64_link_hash_table *ia64_info;
+  asection *unwind_output_sec;
+
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return FALSE;
+
+  /* Make sure we've got ourselves a nice fat __gp value.  */
+  if (!info->relocatable)
+    {
+      bfd_vma gp_val;
+      struct elf_link_hash_entry *gp;
+
+      /* We assume after gp is set, section size will only decrease. We
+        need to adjust gp for it.  */
+      _bfd_set_gp_value (abfd, 0);
+      if (! elf64_ia64_choose_gp (abfd, info, TRUE))
+       return FALSE;
+      gp_val = _bfd_get_gp_value (abfd);
+
+      gp = elf_link_hash_lookup (elf_hash_table (info), "__gp", FALSE,
+                                FALSE, FALSE);
+      if (gp)
+       {
+         gp->root.type = bfd_link_hash_defined;
+         gp->root.u.def.value = gp_val;
+         gp->root.u.def.section = bfd_abs_section_ptr;
+       }
+    }
+
+  /* If we're producing a final executable, we need to sort the contents
+     of the .IA_64.unwind section.  Force this section to be relocated
+     into memory rather than written immediately to the output file.  */
+  unwind_output_sec = NULL;
+  if (!info->relocatable)
+    {
+      asection *s = bfd_get_section_by_name (abfd, ELF_STRING_ia64_unwind);
+      if (s)
+       {
+         unwind_output_sec = s->output_section;
+         unwind_output_sec->contents
+           = bfd_malloc (unwind_output_sec->size);
+         if (unwind_output_sec->contents == NULL)
+           return FALSE;
+       }
+    }
+
+  /* Invoke the regular ELF backend linker to do all the work.  */
+  if (!bfd_elf_final_link (abfd, info))
+    return FALSE;
+
+  if (unwind_output_sec)
+    {
+      elf64_ia64_unwind_entry_compare_bfd = abfd;
+      qsort (unwind_output_sec->contents,
+            (size_t) (unwind_output_sec->size / 24),
+            24,
+            elf64_ia64_unwind_entry_compare);
+
+      if (! bfd_set_section_contents (abfd, unwind_output_sec,
+                                     unwind_output_sec->contents, (bfd_vma) 0,
+                                     unwind_output_sec->size))
+       return FALSE;
+    }
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_ia64_relocate_section (bfd *output_bfd,
+                            struct bfd_link_info *info,
+                            bfd *input_bfd,
+                            asection *input_section,
+                            bfd_byte *contents,
+                            Elf_Internal_Rela *relocs,
+                            Elf_Internal_Sym *local_syms,
+                            asection **local_sections)
+{
+  struct elf64_ia64_link_hash_table *ia64_info;
+  Elf_Internal_Shdr *symtab_hdr;
+  Elf_Internal_Rela *rel;
+  Elf_Internal_Rela *relend;
+  bfd_boolean ret_val = TRUE;  /* for non-fatal errors */
+  bfd_vma gp_val;
+
+  symtab_hdr = &elf_tdata (input_bfd)->symtab_hdr;
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return FALSE;
+
+  /* Infect various flags from the input section to the output section.  */
+  if (info->relocatable)
+    {
+      bfd_vma flags;
+
+      flags = elf_section_data(input_section)->this_hdr.sh_flags;
+      flags &= SHF_IA_64_NORECOV;
+
+      elf_section_data(input_section->output_section)
+       ->this_hdr.sh_flags |= flags;
+    }
+
+  gp_val = _bfd_get_gp_value (output_bfd);
+
+  rel = relocs;
+  relend = relocs + input_section->reloc_count;
+  for (; rel < relend; ++rel)
+    {
+      struct elf_link_hash_entry *h;
+      struct elf64_ia64_dyn_sym_info *dyn_i;
+      bfd_reloc_status_type r;
+      reloc_howto_type *howto;
+      unsigned long r_symndx;
+      Elf_Internal_Sym *sym;
+      unsigned int r_type;
+      bfd_vma value;
+      asection *sym_sec;
+      bfd_byte *hit_addr;
+      bfd_boolean dynamic_symbol_p;
+      bfd_boolean undef_weak_ref;
+
+      r_type = ELF64_R_TYPE (rel->r_info);
+      if (r_type > R_IA64_MAX_RELOC_CODE)
+       {
+         (*_bfd_error_handler)
+           (_("%B: unknown relocation type %d"),
+            input_bfd, (int) r_type);
+         bfd_set_error (bfd_error_bad_value);
+         ret_val = FALSE;
+         continue;
+       }
+
+      howto = ia64_elf_lookup_howto (r_type);
+      r_symndx = ELF64_R_SYM (rel->r_info);
+      h = NULL;
+      sym = NULL;
+      sym_sec = NULL;
+      undef_weak_ref = FALSE;
+
+      if (r_symndx < symtab_hdr->sh_info)
+       {
+         /* Reloc against local symbol.  */
+         asection *msec;
+         sym = local_syms + r_symndx;
+         sym_sec = local_sections[r_symndx];
+         msec = sym_sec;
+         value = _bfd_elf_rela_local_sym (output_bfd, sym, &msec, rel);
+         if (!info->relocatable
+             && (sym_sec->flags & SEC_MERGE) != 0
+             && ELF_ST_TYPE (sym->st_info) == STT_SECTION
+             && sym_sec->sec_info_type == ELF_INFO_TYPE_MERGE)
+           {
+             struct elf64_ia64_local_hash_entry *loc_h;
+
+             loc_h = get_local_sym_hash (ia64_info, input_bfd, rel, FALSE);
+             if (loc_h && ! loc_h->sec_merge_done)
+               {
+                 struct elf64_ia64_dyn_sym_info *dynent;
+                 unsigned int count;
+
+                 for (count = loc_h->count, dynent = loc_h->info;
+                      count != 0;
+                      count--, dynent++)
+                   {
+                     msec = sym_sec;
+                     dynent->addend =
+                       _bfd_merged_section_offset (output_bfd, &msec,
+                                                   elf_section_data (msec)->
+                                                   sec_info,
+                                                   sym->st_value
+                                                   + dynent->addend);
+                     dynent->addend -= sym->st_value;
+                     dynent->addend += msec->output_section->vma
+                                       + msec->output_offset
+                                       - sym_sec->output_section->vma
+                                       - sym_sec->output_offset;
+                   }
+
+                 /* We may have introduced duplicated entries. We need
+                    to remove them properly.  */
+                 count = sort_dyn_sym_info (loc_h->info, loc_h->count);
+                 if (count != loc_h->count)
+                   {
+                     loc_h->count = count;
+                     loc_h->sorted_count = count;
+                   }
+
+                 loc_h->sec_merge_done = 1;
+               }
+           }
+       }
+      else
+       {
+         bfd_boolean unresolved_reloc;
+         bfd_boolean warned;
+         struct elf_link_hash_entry **sym_hashes = elf_sym_hashes (input_bfd);
+
+         RELOC_FOR_GLOBAL_SYMBOL (info, input_bfd, input_section, rel,
+                                  r_symndx, symtab_hdr, sym_hashes,
+                                  h, sym_sec, value,
+                                  unresolved_reloc, warned);
+
+         if (h->root.type == bfd_link_hash_undefweak)
+           undef_weak_ref = TRUE;
+         else if (warned)
+           continue;
+       }
+
+      /* For relocs against symbols from removed linkonce sections,
+        or sections discarded by a linker script, we just want the
+        section contents zeroed.  Avoid any special processing.  */
+      if (sym_sec != NULL && elf_discarded_section (sym_sec))
+       RELOC_AGAINST_DISCARDED_SECTION (info, input_bfd, input_section,
+                                        rel, relend, howto, contents);
+
+      if (info->relocatable)
+       continue;
+
+      hit_addr = contents + rel->r_offset;
+      value += rel->r_addend;
+      dynamic_symbol_p = elf64_ia64_dynamic_symbol_p (h);
+
+      switch (r_type)
+       {
+       case R_IA64_NONE:
+       case R_IA64_LDXMOV:
+         continue;
+
+       case R_IA64_IMM14:
+       case R_IA64_IMM22:
+       case R_IA64_IMM64:
+       case R_IA64_DIR32MSB:
+       case R_IA64_DIR32LSB:
+       case R_IA64_DIR64MSB:
+       case R_IA64_DIR64LSB:
+         /* Install a dynamic relocation for this reloc.  */
+         if ((dynamic_symbol_p || info->shared)
+             && r_symndx != 0
+             && (input_section->flags & SEC_ALLOC) != 0)
+           {
+             unsigned int dyn_r_type;
+             bfd_vma addend;
+
+             switch (r_type)
+               {
+               case R_IA64_IMM14:
+               case R_IA64_IMM22:
+               case R_IA64_IMM64:
+                 /* ??? People shouldn't be doing non-pic code in
+                    shared libraries nor dynamic executables.  */
+                 (*_bfd_error_handler)
+                   (_("%B: non-pic code with imm relocation against dynamic symbol `%s'"),
+                    input_bfd,
+                    h ? h->root.root.string
+                      : bfd_elf_sym_name (input_bfd, symtab_hdr, sym,
+                                          sym_sec));
+                 ret_val = FALSE;
+                 continue;
+
+               default:
+                 break;
+               }
+
+             /* If we don't need dynamic symbol lookup, find a
+                matching RELATIVE relocation.  */
+             dyn_r_type = r_type;
+             if (dynamic_symbol_p)
+               {
+                 addend = rel->r_addend;
+                 value = 0;
+               }
+             else
+               {
+                 addend = value;
+               }
+
+              /* VMS: install a FIX64.  */
+              switch (dyn_r_type)
+                {
+                case R_IA64_DIR32LSB:
+                  dyn_r_type = R_IA64_VMS_FIX32;
+                  break;
+                case R_IA64_DIR64LSB:
+                  dyn_r_type = R_IA64_VMS_FIX64;
+                  break;
+                default:
+                  BFD_ASSERT (FALSE);
+                  break;
+                }
+              elf64_ia64_install_fixup
+                (output_bfd, ia64_info, h,
+                 dyn_r_type, input_section, rel->r_offset, addend);
+              r = bfd_reloc_ok;
+              break;
+           }
+         /* Fall through.  */
+
+       case R_IA64_LTV32MSB:
+       case R_IA64_LTV32LSB:
+       case R_IA64_LTV64MSB:
+       case R_IA64_LTV64LSB:
+         r = ia64_elf_install_value (hit_addr, value, r_type);
+         break;
+
+       case R_IA64_GPREL22:
+       case R_IA64_GPREL64I:
+       case R_IA64_GPREL32MSB:
+       case R_IA64_GPREL32LSB:
+       case R_IA64_GPREL64MSB:
+       case R_IA64_GPREL64LSB:
+         if (dynamic_symbol_p)
+           {
+             (*_bfd_error_handler)
+               (_("%B: @gprel relocation against dynamic symbol %s"),
+                input_bfd,
+                h ? h->root.root.string
+                  : bfd_elf_sym_name (input_bfd, symtab_hdr, sym,
+                                      sym_sec));
+             ret_val = FALSE;
+             continue;
+           }
+         value -= gp_val;
+         r = ia64_elf_install_value (hit_addr, value, r_type);
+         break;
+
+       case R_IA64_LTOFF22:
+       case R_IA64_LTOFF22X:
+       case R_IA64_LTOFF64I:
+          dyn_i = get_dyn_sym_info (ia64_info, h, input_bfd, rel, FALSE);
+         value = set_got_entry (input_bfd, info, dyn_i,
+                                rel->r_addend, value, R_IA64_DIR64LSB);
+         value -= gp_val;
+         r = ia64_elf_install_value (hit_addr, value, r_type);
+         break;
+
+       case R_IA64_PLTOFF22:
+       case R_IA64_PLTOFF64I:
+       case R_IA64_PLTOFF64MSB:
+       case R_IA64_PLTOFF64LSB:
+          dyn_i = get_dyn_sym_info (ia64_info, h, input_bfd, rel, FALSE);
+         value = set_pltoff_entry (output_bfd, info, dyn_i, value, FALSE);
+         value -= gp_val;
+         r = ia64_elf_install_value (hit_addr, value, r_type);
+         break;
+
+       case R_IA64_FPTR64I:
+       case R_IA64_FPTR32MSB:
+       case R_IA64_FPTR32LSB:
+       case R_IA64_FPTR64MSB:
+       case R_IA64_FPTR64LSB:
+          dyn_i = get_dyn_sym_info (ia64_info, h, input_bfd, rel, FALSE);
+         if (dyn_i->want_fptr)
+           {
+             if (!undef_weak_ref)
+               value = set_fptr_entry (output_bfd, info, dyn_i, value);
+           }
+         if (!dyn_i->want_fptr || info->pie)
+           {
+             /* Otherwise, we expect the dynamic linker to create
+                the entry.  */
+
+             if (dyn_i->want_fptr)
+               {
+                 if (r_type == R_IA64_FPTR64I)
+                   {
+                     /* We can't represent this without a dynamic symbol.
+                        Adjust the relocation to be against an output
+                        section symbol, which are always present in the
+                        dynamic symbol table.  */
+                     /* ??? People shouldn't be doing non-pic code in
+                        shared libraries.  Hork.  */
+                     (*_bfd_error_handler)
+                       (_("%B: linking non-pic code in a position independent executable"),
+                        input_bfd);
+                     ret_val = FALSE;
+                     continue;
+                   }
+               }
+             else
+               {
+                 value = 0;
+               }
+
+              /* VMS: FIXFD.  */
+              elf64_ia64_install_fixup
+                (output_bfd, ia64_info, h, R_IA64_VMS_FIXFD,
+                 input_section, rel->r_offset, 0);
+              r = bfd_reloc_ok;
+              break;
+           }
+
+         r = ia64_elf_install_value (hit_addr, value, r_type);
+         break;
+
+       case R_IA64_LTOFF_FPTR22:
+       case R_IA64_LTOFF_FPTR64I:
+       case R_IA64_LTOFF_FPTR32MSB:
+       case R_IA64_LTOFF_FPTR32LSB:
+       case R_IA64_LTOFF_FPTR64MSB:
+       case R_IA64_LTOFF_FPTR64LSB:
+          dyn_i = get_dyn_sym_info (ia64_info, h, input_bfd, rel, FALSE);
+          if (dyn_i->want_fptr)
+            {
+              BFD_ASSERT (h == NULL || !h->def_dynamic);
+              if (!undef_weak_ref)
+                value = set_fptr_entry (output_bfd, info, dyn_i, value);
+            }
+          else
+            value = 0;
+
+          value = set_got_entry (output_bfd, info, dyn_i,
+                                 rel->r_addend, value, R_IA64_FPTR64LSB);
+          value -= gp_val;
+          r = ia64_elf_install_value (hit_addr, value, r_type);
+         break;
+
+       case R_IA64_PCREL32MSB:
+       case R_IA64_PCREL32LSB:
+       case R_IA64_PCREL64MSB:
+       case R_IA64_PCREL64LSB:
+         /* Install a dynamic relocation for this reloc.  */
+         if (dynamic_symbol_p && r_symndx != 0)
+           {
+              /* VMS: doesn't exist ???  */
+              abort ();
+           }
+         goto finish_pcrel;
+
+       case R_IA64_PCREL21B:
+       case R_IA64_PCREL60B:
+         /* We should have created a PLT entry for any dynamic symbol.  */
+         dyn_i = NULL;
+         if (h)
+           dyn_i = get_dyn_sym_info (ia64_info, h, NULL, NULL, FALSE);
+
+         if (dyn_i && dyn_i->want_plt2)
+           {
+             /* Should have caught this earlier.  */
+             BFD_ASSERT (rel->r_addend == 0);
+
+             value = (ia64_info->root.splt->output_section->vma
+                      + ia64_info->root.splt->output_offset
+                      + dyn_i->plt2_offset);
+           }
+         else
+           {
+             /* Since there's no PLT entry, Validate that this is
+                locally defined.  */
+             BFD_ASSERT (undef_weak_ref || sym_sec->output_section != NULL);
+
+             /* If the symbol is undef_weak, we shouldn't be trying
+                to call it.  There's every chance that we'd wind up
+                with an out-of-range fixup here.  Don't bother setting
+                any value at all.  */
+             if (undef_weak_ref)
+               continue;
+           }
+         goto finish_pcrel;
+
+       case R_IA64_PCREL21BI:
+       case R_IA64_PCREL21F:
+       case R_IA64_PCREL21M:
+       case R_IA64_PCREL22:
+       case R_IA64_PCREL64I:
+         /* The PCREL21BI reloc is specifically not intended for use with
+            dynamic relocs.  PCREL21F and PCREL21M are used for speculation
+            fixup code, and thus probably ought not be dynamic.  The
+            PCREL22 and PCREL64I relocs aren't emitted as dynamic relocs.  */
+         if (dynamic_symbol_p)
+           {
+             const char *msg;
+
+             if (r_type == R_IA64_PCREL21BI)
+               msg = _("%B: @internal branch to dynamic symbol %s");
+             else if (r_type == R_IA64_PCREL21F || r_type == R_IA64_PCREL21M)
+               msg = _("%B: speculation fixup to dynamic symbol %s");
+             else
+               msg = _("%B: @pcrel relocation against dynamic symbol %s");
+             (*_bfd_error_handler) (msg, input_bfd,
+                                    h ? h->root.root.string
+                                      : bfd_elf_sym_name (input_bfd,
+                                                          symtab_hdr,
+                                                          sym,
+                                                          sym_sec));
+             ret_val = FALSE;
+             continue;
+           }
+         goto finish_pcrel;
+
+       finish_pcrel:
+         /* Make pc-relative.  */
+         value -= (input_section->output_section->vma
+                   + input_section->output_offset
+                   + rel->r_offset) & ~ (bfd_vma) 0x3;
+         r = ia64_elf_install_value (hit_addr, value, r_type);
+         break;
+
+       case R_IA64_SEGREL32MSB:
+       case R_IA64_SEGREL32LSB:
+       case R_IA64_SEGREL64MSB:
+       case R_IA64_SEGREL64LSB:
+           {
+             /* Find the segment that contains the output_section.  */
+             Elf_Internal_Phdr *p = _bfd_elf_find_segment_containing_section
+               (output_bfd, sym_sec->output_section);
+
+             if (p == NULL)
+               {
+                 r = bfd_reloc_notsupported;
+               }
+             else
+               {
+                 /* The VMA of the segment is the vaddr of the associated
+                    program header.  */
+                 if (value > p->p_vaddr)
+                   value -= p->p_vaddr;
+                 else
+                   value = 0;
+                 r = ia64_elf_install_value (hit_addr, value, r_type);
+               }
+             break;
+           }
+
+       case R_IA64_SECREL32MSB:
+       case R_IA64_SECREL32LSB:
+       case R_IA64_SECREL64MSB:
+       case R_IA64_SECREL64LSB:
+         /* Make output-section relative to section where the symbol
+            is defined. PR 475  */
+         if (sym_sec)
+           value -= sym_sec->output_section->vma;
+         r = ia64_elf_install_value (hit_addr, value, r_type);
+         break;
+
+       case R_IA64_IPLTMSB:
+       case R_IA64_IPLTLSB:
+         /* Install a dynamic relocation for this reloc.  */
+         if ((dynamic_symbol_p || info->shared)
+             && (input_section->flags & SEC_ALLOC) != 0)
+           {
+              /* VMS: FIXFD ??  */
+              abort ();
+           }
+
+         if (r_type == R_IA64_IPLTMSB)
+           r_type = R_IA64_DIR64MSB;
+         else
+           r_type = R_IA64_DIR64LSB;
+         ia64_elf_install_value (hit_addr, value, r_type);
+         r = ia64_elf_install_value (hit_addr + 8, gp_val, r_type);
+         break;
+
+       case R_IA64_TPREL14:
+       case R_IA64_TPREL22:
+       case R_IA64_TPREL64I:
+         r = bfd_reloc_notsupported;
+         break;
+
+       case R_IA64_DTPREL14:
+       case R_IA64_DTPREL22:
+       case R_IA64_DTPREL64I:
+       case R_IA64_DTPREL32LSB:
+       case R_IA64_DTPREL32MSB:
+       case R_IA64_DTPREL64LSB:
+       case R_IA64_DTPREL64MSB:
+         r = bfd_reloc_notsupported;
+         break;
+
+       case R_IA64_LTOFF_TPREL22:
+       case R_IA64_LTOFF_DTPMOD22:
+       case R_IA64_LTOFF_DTPREL22:
+         r = bfd_reloc_notsupported;
+         break;
+
+       default:
+         r = bfd_reloc_notsupported;
+         break;
+       }
+
+      switch (r)
+       {
+       case bfd_reloc_ok:
+         break;
+
+       case bfd_reloc_undefined:
+         /* This can happen for global table relative relocs if
+            __gp is undefined.  This is a panic situation so we
+            don't try to continue.  */
+         (*info->callbacks->undefined_symbol)
+           (info, "__gp", input_bfd, input_section, rel->r_offset, 1);
+         return FALSE;
+
+       case bfd_reloc_notsupported:
+         {
+           const char *name;
+
+           if (h)
+             name = h->root.root.string;
+           else
+             name = bfd_elf_sym_name (input_bfd, symtab_hdr, sym,
+                                      sym_sec);
+           if (!(*info->callbacks->warning) (info, _("unsupported reloc"),
+                                             name, input_bfd,
+                                             input_section, rel->r_offset))
+             return FALSE;
+           ret_val = FALSE;
+         }
+         break;
+
+       case bfd_reloc_dangerous:
+       case bfd_reloc_outofrange:
+       case bfd_reloc_overflow:
+       default:
+         {
+           const char *name;
+
+           if (h)
+             name = h->root.root.string;
+           else
+             name = bfd_elf_sym_name (input_bfd, symtab_hdr, sym,
+                                      sym_sec);
+
+           switch (r_type)
+             {
+             case R_IA64_TPREL14:
+             case R_IA64_TPREL22:
+             case R_IA64_TPREL64I:
+             case R_IA64_DTPREL14:
+             case R_IA64_DTPREL22:
+             case R_IA64_DTPREL64I:
+             case R_IA64_DTPREL32LSB:
+             case R_IA64_DTPREL32MSB:
+             case R_IA64_DTPREL64LSB:
+             case R_IA64_DTPREL64MSB:
+             case R_IA64_LTOFF_TPREL22:
+             case R_IA64_LTOFF_DTPMOD22:
+             case R_IA64_LTOFF_DTPREL22:
+               (*_bfd_error_handler)
+                 (_("%B: missing TLS section for relocation %s against `%s' at 0x%lx in section `%A'."),
+                  input_bfd, input_section, howto->name, name,
+                  rel->r_offset);
+               break;
+
+             case R_IA64_PCREL21B:
+             case R_IA64_PCREL21BI:
+             case R_IA64_PCREL21M:
+             case R_IA64_PCREL21F:
+               if (is_elf_hash_table (info->hash))
+                 {
+                   /* Relaxtion is always performed for ELF output.
+                      Overflow failures for those relocations mean
+                      that the section is too big to relax.  */
+                   (*_bfd_error_handler)
+                     (_("%B: Can't relax br (%s) to `%s' at 0x%lx in section `%A' with size 0x%lx (> 0x1000000)."),
+                      input_bfd, input_section, howto->name, name,
+                      rel->r_offset, input_section->size);
+                   break;
+                 }
+             default:
+               if (!(*info->callbacks->reloc_overflow) (info,
+                                                        &h->root,
+                                                        name,
+                                                        howto->name,
+                                                        (bfd_vma) 0,
+                                                        input_bfd,
+                                                        input_section,
+                                                        rel->r_offset))
+                 return FALSE;
+               break;
+             }
+
+           ret_val = FALSE;
+         }
+         break;
+       }
+    }
+
+  return ret_val;
+}
+
+static bfd_boolean
+elf64_ia64_finish_dynamic_symbol (bfd *output_bfd,
+                                 struct bfd_link_info *info,
+                                 struct elf_link_hash_entry *h,
+                                 Elf_Internal_Sym *sym)
+{
+  struct elf64_ia64_link_hash_table *ia64_info;
+  struct elf64_ia64_dyn_sym_info *dyn_i;
+
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return FALSE;
+
+  dyn_i = get_dyn_sym_info (ia64_info, h, NULL, NULL, FALSE);
+
+  /* Fill in the PLT data, if required.  */
+  if (dyn_i && dyn_i->want_plt)
+    {
+      bfd_byte *loc;
+      asection *plt_sec;
+      bfd_vma plt_addr, pltoff_addr, gp_val;
+
+      gp_val = _bfd_get_gp_value (output_bfd);
+
+      plt_sec = ia64_info->root.splt;
+      plt_addr = 0;  /* Not used as overriden by FIXUPs.  */
+      pltoff_addr = set_pltoff_entry (output_bfd, info, dyn_i, plt_addr, TRUE);
+
+      /* Initialize the FULL PLT entry, if needed.  */
+      if (dyn_i->want_plt2)
+       {
+         loc = plt_sec->contents + dyn_i->plt2_offset;
+
+         memcpy (loc, plt_full_entry, PLT_FULL_ENTRY_SIZE);
+         ia64_elf_install_value (loc, pltoff_addr - gp_val, R_IA64_IMM22);
+
+         /* Mark the symbol as undefined, rather than as defined in the
+            plt section.  Leave the value alone.  */
+         /* ??? We didn't redefine it in adjust_dynamic_symbol in the
+            first place.  But perhaps elflink.c did some for us.  */
+         if (!h->def_regular)
+           sym->st_shndx = SHN_UNDEF;
+       }
+
+      /* VMS: FIXFD.  */
+      elf64_ia64_install_fixup
+        (output_bfd, ia64_info, h, R_IA64_VMS_FIXFD, ia64_info->pltoff_sec,
+         pltoff_addr - (ia64_info->pltoff_sec->output_section->vma
+                        + ia64_info->pltoff_sec->output_offset), 0);
+    }
+
+  /* Mark some specially defined symbols as absolute.  */
+  if (strcmp (h->root.root.string, "_DYNAMIC") == 0
+      || h == ia64_info->root.hgot
+      || h == ia64_info->root.hplt)
+    sym->st_shndx = SHN_ABS;
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_ia64_finish_dynamic_sections (bfd *abfd,
+                                   struct bfd_link_info *info)
+{
+  struct elf64_ia64_link_hash_table *ia64_info;
+  bfd *dynobj;
+
+  ia64_info = elf64_ia64_hash_table (info);
+  if (ia64_info == NULL)
+    return FALSE;
+
+  dynobj = ia64_info->root.dynobj;
+
+  if (elf_hash_table (info)->dynamic_sections_created)
+    {
+      Elf64_External_Dyn *dyncon, *dynconend;
+      asection *sdyn;
+      asection *unwind_sec;
+      bfd_vma gp_val;
+      unsigned int gp_seg;
+      bfd_vma gp_off;
+      Elf_Internal_Phdr *phdr;
+      Elf_Internal_Phdr *base_phdr;
+      unsigned int unwind_seg = 0;
+      unsigned int code_seg = 0;
+
+      sdyn = bfd_get_section_by_name (dynobj, ".dynamic");
+      BFD_ASSERT (sdyn != NULL);
+      dyncon = (Elf64_External_Dyn *) sdyn->contents;
+      dynconend = (Elf64_External_Dyn *) (sdyn->contents + sdyn->size);
+
+      gp_val = _bfd_get_gp_value (abfd);
+      phdr = _bfd_elf_find_segment_containing_section
+        (info->output_bfd, ia64_info->pltoff_sec->output_section);
+      BFD_ASSERT (phdr != NULL);
+      base_phdr = elf_tdata (info->output_bfd)->phdr;
+      gp_seg = phdr - base_phdr;
+      gp_off = gp_val - phdr->p_vaddr;
+
+      unwind_sec = bfd_get_section_by_name (abfd, ELF_STRING_ia64_unwind);
+      if (unwind_sec != NULL)
+        {
+          asection *code_sec;
+
+          phdr = _bfd_elf_find_segment_containing_section (abfd, unwind_sec);
+          BFD_ASSERT (phdr != NULL);
+          unwind_seg = phdr - base_phdr;
+
+          code_sec = bfd_get_section_by_name (abfd, "$CODE$");
+          phdr = _bfd_elf_find_segment_containing_section (abfd, code_sec);
+          BFD_ASSERT (phdr != NULL);
+          code_seg = phdr - base_phdr;
+        }
+
+      for (; dyncon < dynconend; dyncon++)
+       {
+         Elf_Internal_Dyn dyn;
+
+         bfd_elf64_swap_dyn_in (dynobj, dyncon, &dyn);
+
+         switch (dyn.d_tag)
+           {
+            case DT_IA_64_VMS_FIXUP_RELA_OFF:
+              dyn.d_un.d_val +=
+                (ia64_info->fixups_sec->output_section->vma
+                 + ia64_info->fixups_sec->output_offset)
+                - (sdyn->output_section->vma + sdyn->output_offset);
+              break;
+
+            case DT_IA_64_VMS_PLTGOT_OFFSET:
+              dyn.d_un.d_val = gp_off;
+              break;
+
+            case DT_IA_64_VMS_PLTGOT_SEG:
+              dyn.d_un.d_val = gp_seg;
+              break;
+
+            case DT_IA_64_VMS_UNWINDSZ:
+              if (unwind_sec == NULL)
+                {
+                  dyn.d_tag = DT_NULL;
+                  dyn.d_un.d_val = 0xdead;
+                }
+              else
+                dyn.d_un.d_val = unwind_sec->size;
+              break;
+
+            case DT_IA_64_VMS_UNWIND_CODSEG:
+              dyn.d_un.d_val = code_seg;
+              break;
+
+            case DT_IA_64_VMS_UNWIND_INFOSEG:
+            case DT_IA_64_VMS_UNWIND_SEG:
+              dyn.d_un.d_val = unwind_seg;
+              break;
+
+            case DT_IA_64_VMS_UNWIND_OFFSET:
+              break;
+
+            default:
+              /* No need to rewrite the entry.  */
+              continue;
+           }
+
+         bfd_elf64_swap_dyn_out (abfd, &dyn, dyncon);
+       }
+    }
+
+  /* Handle transfer addresses.  */
+  {
+    asection *tfr_sec = ia64_info->transfer_sec;
+    struct elf64_vms_transfer *tfr;
+    struct elf_link_hash_entry *tfr3;
+
+    tfr = (struct elf64_vms_transfer *)tfr_sec->contents;
+    bfd_putl32 (6 * 8, tfr->size);
+    bfd_putl64 (tfr_sec->output_section->vma
+                + tfr_sec->output_offset
+                + 6 * 8, tfr->tfradr3);
+
+    tfr3 = elf_link_hash_lookup (elf_hash_table (info), "ELF$TFRADR", FALSE,
+                                 FALSE, FALSE);
+
+    if (tfr3
+        && (tfr3->root.type == bfd_link_hash_defined
+            || tfr3->root.type == bfd_link_hash_defweak))
+      {
+        asection *tfr3_sec = tfr3->root.u.def.section;
+        bfd_vma tfr3_val;
+
+        tfr3_val = (tfr3->root.u.def.value
+                    + tfr3_sec->output_section->vma
+                    + tfr3_sec->output_offset);
+
+        bfd_putl64 (tfr3_val, tfr->tfr3_func);
+        bfd_putl64 (_bfd_get_gp_value (info->output_bfd), tfr->tfr3_gp);
+      }
+
+    /* FIXME: set linker flags,
+       handle lib$initialize.  */
+  }
+
+  return TRUE;
+}
+
+/* ELF file flag handling:  */
+
+/* Function to keep IA-64 specific file flags.  */
+static bfd_boolean
+elf64_ia64_set_private_flags (bfd *abfd, flagword flags)
+{
+  BFD_ASSERT (!elf_flags_init (abfd)
+             || elf_elfheader (abfd)->e_flags == flags);
+
+  elf_elfheader (abfd)->e_flags = flags;
+  elf_flags_init (abfd) = TRUE;
+  return TRUE;
+}
+
+/* Merge backend specific data from an object file to the output
+   object file when linking.  */
+static bfd_boolean
+elf64_ia64_merge_private_bfd_data (bfd *ibfd, bfd *obfd)
+{
+  flagword out_flags;
+  flagword in_flags;
+  bfd_boolean ok = TRUE;
+
+  /* Don't even pretend to support mixed-format linking.  */
+  if (bfd_get_flavour (ibfd) != bfd_target_elf_flavour
+      || bfd_get_flavour (obfd) != bfd_target_elf_flavour)
+    return FALSE;
+
+  in_flags  = elf_elfheader (ibfd)->e_flags;
+  out_flags = elf_elfheader (obfd)->e_flags;
+
+  if (! elf_flags_init (obfd))
+    {
+      elf_flags_init (obfd) = TRUE;
+      elf_elfheader (obfd)->e_flags = in_flags;
+
+      if (bfd_get_arch (obfd) == bfd_get_arch (ibfd)
+         && bfd_get_arch_info (obfd)->the_default)
+       {
+         return bfd_set_arch_mach (obfd, bfd_get_arch (ibfd),
+                                   bfd_get_mach (ibfd));
+       }
+
+      return TRUE;
+    }
+
+  /* Check flag compatibility.  */
+  if (in_flags == out_flags)
+    return TRUE;
+
+  /* Output has EF_IA_64_REDUCEDFP set only if all inputs have it set.  */
+  if (!(in_flags & EF_IA_64_REDUCEDFP) && (out_flags & EF_IA_64_REDUCEDFP))
+    elf_elfheader (obfd)->e_flags &= ~EF_IA_64_REDUCEDFP;
+
+  if ((in_flags & EF_IA_64_TRAPNIL) != (out_flags & EF_IA_64_TRAPNIL))
+    {
+      (*_bfd_error_handler)
+       (_("%B: linking trap-on-NULL-dereference with non-trapping files"),
+        ibfd);
+
+      bfd_set_error (bfd_error_bad_value);
+      ok = FALSE;
+    }
+  if ((in_flags & EF_IA_64_BE) != (out_flags & EF_IA_64_BE))
+    {
+      (*_bfd_error_handler)
+       (_("%B: linking big-endian files with little-endian files"),
+        ibfd);
+
+      bfd_set_error (bfd_error_bad_value);
+      ok = FALSE;
+    }
+  if ((in_flags & EF_IA_64_ABI64) != (out_flags & EF_IA_64_ABI64))
+    {
+      (*_bfd_error_handler)
+       (_("%B: linking 64-bit files with 32-bit files"),
+        ibfd);
+
+      bfd_set_error (bfd_error_bad_value);
+      ok = FALSE;
+    }
+  if ((in_flags & EF_IA_64_CONS_GP) != (out_flags & EF_IA_64_CONS_GP))
+    {
+      (*_bfd_error_handler)
+       (_("%B: linking constant-gp files with non-constant-gp files"),
+        ibfd);
+
+      bfd_set_error (bfd_error_bad_value);
+      ok = FALSE;
+    }
+  if ((in_flags & EF_IA_64_NOFUNCDESC_CONS_GP)
+      != (out_flags & EF_IA_64_NOFUNCDESC_CONS_GP))
+    {
+      (*_bfd_error_handler)
+       (_("%B: linking auto-pic files with non-auto-pic files"),
+        ibfd);
+
+      bfd_set_error (bfd_error_bad_value);
+      ok = FALSE;
+    }
+
+  return ok;
+}
+
+static bfd_boolean
+elf64_ia64_print_private_bfd_data (bfd *abfd, void * ptr)
+{
+  FILE *file = (FILE *) ptr;
+  flagword flags = elf_elfheader (abfd)->e_flags;
+
+  BFD_ASSERT (abfd != NULL && ptr != NULL);
+
+  fprintf (file, "private flags = %s%s%s%s%s%s%s%s\n",
+          (flags & EF_IA_64_TRAPNIL) ? "TRAPNIL, " : "",
+          (flags & EF_IA_64_EXT) ? "EXT, " : "",
+          (flags & EF_IA_64_BE) ? "BE, " : "LE, ",
+          (flags & EF_IA_64_REDUCEDFP) ? "REDUCEDFP, " : "",
+          (flags & EF_IA_64_CONS_GP) ? "CONS_GP, " : "",
+          (flags & EF_IA_64_NOFUNCDESC_CONS_GP) ? "NOFUNCDESC_CONS_GP, " : "",
+          (flags & EF_IA_64_ABSOLUTE) ? "ABSOLUTE, " : "",
+          (flags & EF_IA_64_ABI64) ? "ABI64" : "ABI32");
+
+  _bfd_elf_print_private_bfd_data (abfd, ptr);
+  return TRUE;
+}
+
+static enum elf_reloc_type_class
+elf64_ia64_reloc_type_class (const Elf_Internal_Rela *rela)
+{
+  switch ((int) ELF64_R_TYPE (rela->r_info))
+    {
+    case R_IA64_REL32MSB:
+    case R_IA64_REL32LSB:
+    case R_IA64_REL64MSB:
+    case R_IA64_REL64LSB:
+      return reloc_class_relative;
+    case R_IA64_IPLTMSB:
+    case R_IA64_IPLTLSB:
+      return reloc_class_plt;
+    case R_IA64_COPY:
+      return reloc_class_copy;
+    default:
+      return reloc_class_normal;
+    }
+}
+
+static const struct bfd_elf_special_section elf64_ia64_special_sections[] =
+{
+  { STRING_COMMA_LEN (".sbss"),  -1, SHT_NOBITS,   SHF_ALLOC + SHF_WRITE + SHF_IA_64_SHORT },
+  { STRING_COMMA_LEN (".sdata"), -1, SHT_PROGBITS, SHF_ALLOC + SHF_WRITE + SHF_IA_64_SHORT },
+  { NULL,                    0,   0, 0,            0 }
+};
+
+static bfd_boolean
+elf64_ia64_object_p (bfd *abfd)
+{
+  asection *sec;
+  asection *group, *unwi, *unw;
+  flagword flags;
+  const char *name;
+  char *unwi_name, *unw_name;
+  bfd_size_type amt;
+
+  if (abfd->flags & DYNAMIC)
+    return TRUE;
+
+  /* Flags for fake group section.  */
+  flags = (SEC_LINKER_CREATED | SEC_GROUP | SEC_LINK_ONCE
+          | SEC_EXCLUDE);
+
+  /* We add a fake section group for each .gnu.linkonce.t.* section,
+     which isn't in a section group, and its unwind sections.  */
+  for (sec = abfd->sections; sec != NULL; sec = sec->next)
+    {
+      if (elf_sec_group (sec) == NULL
+         && ((sec->flags & (SEC_LINK_ONCE | SEC_CODE | SEC_GROUP))
+             == (SEC_LINK_ONCE | SEC_CODE))
+         && CONST_STRNEQ (sec->name, ".gnu.linkonce.t."))
+       {
+         name = sec->name + 16;
+
+         amt = strlen (name) + sizeof (".gnu.linkonce.ia64unwi.");
+         unwi_name = bfd_alloc (abfd, amt);
+         if (!unwi_name)
+           return FALSE;
+
+         strcpy (stpcpy (unwi_name, ".gnu.linkonce.ia64unwi."), name);
+         unwi = bfd_get_section_by_name (abfd, unwi_name);
+
+         amt = strlen (name) + sizeof (".gnu.linkonce.ia64unw.");
+         unw_name = bfd_alloc (abfd, amt);
+         if (!unw_name)
+           return FALSE;
+
+         strcpy (stpcpy (unw_name, ".gnu.linkonce.ia64unw."), name);
+         unw = bfd_get_section_by_name (abfd, unw_name);
+
+         /* We need to create a fake group section for it and its
+            unwind sections.  */
+         group = bfd_make_section_anyway_with_flags (abfd, name,
+                                                     flags);
+         if (group == NULL)
+           return FALSE;
+
+         /* Move the fake group section to the beginning.  */
+         bfd_section_list_remove (abfd, group);
+         bfd_section_list_prepend (abfd, group);
+
+         elf_next_in_group (group) = sec;
+
+         elf_group_name (sec) = name;
+         elf_next_in_group (sec) = sec;
+         elf_sec_group (sec) = group;
+
+         if (unwi)
+           {
+             elf_group_name (unwi) = name;
+             elf_next_in_group (unwi) = sec;
+             elf_next_in_group (sec) = unwi;
+             elf_sec_group (unwi) = group;
+           }
+
+          if (unw)
+            {
+              elf_group_name (unw) = name;
+              if (unwi)
+                {
+                  elf_next_in_group (unw) = elf_next_in_group (unwi);
+                  elf_next_in_group (unwi) = unw;
+                }
+              else
+                {
+                  elf_next_in_group (unw) = sec;
+                  elf_next_in_group (sec) = unw;
+                }
+              elf_sec_group (unw) = group;
+            }
+
+          /* Fake SHT_GROUP section header.  */
+         elf_section_data (group)->this_hdr.bfd_section = group;
+         elf_section_data (group)->this_hdr.sh_type = SHT_GROUP;
+       }
+    }
+  return TRUE;
+}
+
+/* Handle an IA-64 specific section when reading an object file.  This
+   is called when bfd_section_from_shdr finds a section with an unknown
+   type.  */
+
+static bfd_boolean
+elf64_vms_section_from_shdr (bfd *abfd,
+                            Elf_Internal_Shdr *hdr,
+                            const char *name,
+                            int shindex)
+{
+  flagword secflags = 0;
+
+  switch (hdr->sh_type)
+    {
+    case SHT_IA_64_VMS_TRACE:
+    case SHT_IA_64_VMS_DEBUG:
+    case SHT_IA_64_VMS_DEBUG_STR:
+      secflags = SEC_DEBUGGING;
+      break;
+
+    case SHT_IA_64_UNWIND:
+    case SHT_IA_64_HP_OPT_ANOT:
+      break;
+
+    case SHT_IA_64_EXT:
+      if (strcmp (name, ELF_STRING_ia64_archext) != 0)
+       return FALSE;
+      break;
+
+    default:
+      return FALSE;
+    }
+
+  if (! _bfd_elf_make_section_from_shdr (abfd, hdr, name, shindex))
+    return FALSE;
+
+  if (secflags != 0)
+    {
+      asection *newsect = hdr->bfd_section;
+
+      if (! bfd_set_section_flags
+          (abfd, newsect, bfd_get_section_flags (abfd, newsect) | secflags))
+       return FALSE;
+    }
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_vms_object_p (bfd *abfd)
+{
+  Elf_Internal_Ehdr *i_ehdrp = elf_elfheader (abfd);
+  Elf_Internal_Phdr *i_phdr = elf_tdata (abfd)->phdr;
+  unsigned int i;
+  unsigned int num_text = 0;
+  unsigned int num_data = 0;
+  unsigned int num_rodata = 0;
+  char name[16];
+
+  if (!elf64_ia64_object_p (abfd))
+    return FALSE;
+
+  /* Many VMS compilers do not generate sections for the corresponding
+     segment.  This is boring as binutils tools won't be able to disassemble
+     the code.  So we simply create all the missing sections.  */
+  for (i = 0; i < i_ehdrp->e_phnum; i++, i_phdr++)
+    {
+      /* Is there a section for this segment?  */
+      bfd_vma base_vma = i_phdr->p_vaddr;
+      bfd_vma limit_vma = base_vma + i_phdr->p_filesz;
+
+      if (i_phdr->p_type != PT_LOAD)
+       continue;
+
+      /* We need to cover from base_vms to limit_vma.  */
+    again:
+      while (base_vma < limit_vma)
+       {
+         bfd_vma next_vma = limit_vma;
+         asection *nsec;
+         asection *sec;
+         flagword flags;
+         char *nname = NULL;
+
+         /* Find a section covering [base_vma;limit_vma)  */
+         for (sec = abfd->sections; sec != NULL; sec = sec->next)
+           {
+             /* Skip uninteresting sections (either not in memory or
+                below base_vma.  */
+             if ((sec->flags & (SEC_ALLOC | SEC_LOAD)) == 0
+                 || sec->vma + sec->size <= base_vma)
+               continue;
+             if (sec->vma <= base_vma)
+               {
+                 /* This section covers (maybe partially) the beginning
+                    of the range.  */
+                 base_vma = sec->vma + sec->size;
+                 goto again;
+               }
+             if (sec->vma < next_vma)
+               {
+                 /* This section partially covers the end of the range.
+                    Used to compute the size of the hole.  */
+                 next_vma = sec->vma;
+               }
+           }
+
+         /* No section covering [base_vma; next_vma).  Create a fake one.  */
+         flags = SEC_ALLOC | SEC_LOAD | SEC_HAS_CONTENTS;
+         if (i_phdr->p_flags & PF_X)
+           {
+             flags |= SEC_CODE;
+             if (num_text++ == 0)
+               nname = ".text";
+             else
+               sprintf (name, ".text$%u", num_text);
+           }
+         else if ((i_phdr->p_flags & (PF_R | PF_W)) == PF_R)
+           {
+             flags |= SEC_READONLY;
+             sprintf (name, ".rodata$%u", num_rodata++);
+           }
+         else
+           {
+             flags |= SEC_DATA;
+             sprintf (name, ".data$%u", num_data++);
+           }
+
+         /* Allocate name.  */
+         if (nname == NULL)
+           {
+             size_t name_len = strlen (name) + 1;
+             nname = bfd_alloc (abfd, name_len);
+             if (nname == NULL)
+               return FALSE;
+             memcpy (nname, name, name_len);
+           }
+
+         /* Create and fill new section.  */
+         nsec = bfd_make_section_anyway_with_flags (abfd, nname, flags);
+         if (nsec == NULL)
+           return FALSE;
+         nsec->vma = base_vma;
+         nsec->size = next_vma - base_vma;
+         nsec->filepos = i_phdr->p_offset + (base_vma - i_phdr->p_vaddr);
+
+         base_vma = next_vma;
+       }
+    }
+  return TRUE;
+}
+
+static void
+elf64_vms_post_process_headers (bfd *abfd,
+                               struct bfd_link_info *info ATTRIBUTE_UNUSED)
+{
+  Elf_Internal_Ehdr *i_ehdrp = elf_elfheader (abfd);
+
+  i_ehdrp->e_ident[EI_OSABI] = ELFOSABI_OPENVMS;
+  i_ehdrp->e_ident[EI_ABIVERSION] = 2;
+}
+
+static bfd_boolean
+elf64_vms_section_processing (bfd *abfd ATTRIBUTE_UNUSED,
+                             Elf_Internal_Shdr *hdr)
+{
+  if (hdr->bfd_section != NULL)
+    {
+      const char *name = bfd_get_section_name (abfd, hdr->bfd_section);
+
+      if (strcmp (name, ".text") == 0)
+       hdr->sh_flags |= SHF_IA_64_VMS_SHARED;
+      else if ((strcmp (name, ".debug") == 0)
+           || (strcmp (name, ".debug_abbrev") == 0)
+           || (strcmp (name, ".debug_aranges") == 0)
+           || (strcmp (name, ".debug_frame") == 0)
+           || (strcmp (name, ".debug_info") == 0)
+           || (strcmp (name, ".debug_loc") == 0)
+           || (strcmp (name, ".debug_macinfo") == 0)
+           || (strcmp (name, ".debug_pubnames") == 0)
+           || (strcmp (name, ".debug_pubtypes") == 0))
+       hdr->sh_type = SHT_IA_64_VMS_DEBUG;
+      else if ((strcmp (name, ".debug_line") == 0)
+           || (strcmp (name, ".debug_ranges") == 0)
+           || (strcmp (name, ".trace_info") == 0)
+           || (strcmp (name, ".trace_abbrev") == 0)
+           || (strcmp (name, ".trace_aranges") == 0))
+       hdr->sh_type = SHT_IA_64_VMS_TRACE;
+      else if (strcmp (name, ".debug_str") == 0)
+       hdr->sh_type = SHT_IA_64_VMS_DEBUG_STR;
+    }
+
+  return TRUE;
+}
+
+/* The final processing done just before writing out a VMS IA-64 ELF
+   object file.  */
+
+static void
+elf64_vms_final_write_processing (bfd *abfd,
+                                 bfd_boolean linker ATTRIBUTE_UNUSED)
+{
+  Elf_Internal_Shdr *hdr;
+  asection *s;
+  int unwind_info_sect_idx = 0;
+
+  for (s = abfd->sections; s; s = s->next)
+    {
+      hdr = &elf_section_data (s)->this_hdr;
+
+      if (strcmp (bfd_get_section_name (abfd, hdr->bfd_section),
+                 ".IA_64.unwind_info") == 0)
+       unwind_info_sect_idx = elf_section_data (s)->this_idx;
+
+      switch (hdr->sh_type)
+       {
+       case SHT_IA_64_UNWIND:
+         /* VMS requires sh_info to point to the unwind info section.  */
+          hdr->sh_info = unwind_info_sect_idx;
+         break;
+       }
+    }
+
+  if (! elf_flags_init (abfd))
+    {
+      unsigned long flags = 0;
+
+      if (abfd->xvec->byteorder == BFD_ENDIAN_BIG)
+       flags |= EF_IA_64_BE;
+      if (bfd_get_mach (abfd) == bfd_mach_ia64_elf64)
+       flags |= EF_IA_64_ABI64;
+
+      elf_elfheader (abfd)->e_flags = flags;
+      elf_flags_init (abfd) = TRUE;
+    }
+}
+
+static bfd_boolean
+elf64_vms_write_shdrs_and_ehdr (bfd *abfd)
+{
+  unsigned char needed_count[8];
+
+  if (!bfd_elf64_write_shdrs_and_ehdr (abfd))
+    return FALSE;
+
+  bfd_putl64 (elf_ia64_vms_tdata (abfd)->needed_count, needed_count);
+
+  if (bfd_seek (abfd, sizeof (Elf64_External_Ehdr), SEEK_SET) != 0
+      || bfd_bwrite (needed_count, 8, abfd) != 8)
+    return FALSE;
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_vms_close_and_cleanup (bfd *abfd)
+{
+  if (bfd_get_format (abfd) == bfd_object)
+    {
+      long isize;
+
+      /* Pad to 8 byte boundary for IPF/VMS.  */
+      isize = bfd_get_size (abfd);
+      if ((isize & 7) != 0)
+       {
+         int ishort = 8 - (isize & 7);
+          bfd_uint64_t pad = 0;
+
+         bfd_seek (abfd, isize, SEEK_SET);
+         bfd_bwrite (&pad, ishort, abfd);
+       }
+    }
+
+  return _bfd_elf_close_and_cleanup (abfd);
+}
+
+/* Add symbols from an ELF object file to the linker hash table.  */
+
+static bfd_boolean
+elf64_vms_link_add_object_symbols (bfd *abfd, struct bfd_link_info *info)
+{
+  Elf_Internal_Shdr *hdr;
+  bfd_size_type symcount;
+  bfd_size_type extsymcount;
+  bfd_size_type extsymoff;
+  struct elf_link_hash_entry **sym_hash;
+  bfd_boolean dynamic;
+  Elf_Internal_Sym *isymbuf = NULL;
+  Elf_Internal_Sym *isym;
+  Elf_Internal_Sym *isymend;
+  const struct elf_backend_data *bed;
+  struct elf_link_hash_table *htab;
+  bfd_size_type amt;
+
+  htab = elf_hash_table (info);
+  bed = get_elf_backend_data (abfd);
+
+  if ((abfd->flags & DYNAMIC) == 0)
+    dynamic = FALSE;
+  else
+    {
+      dynamic = TRUE;
+
+      /* You can't use -r against a dynamic object.  Also, there's no
+        hope of using a dynamic object which does not exactly match
+        the format of the output file.  */
+      if (info->relocatable
+         || !is_elf_hash_table (htab)
+         || info->output_bfd->xvec != abfd->xvec)
+       {
+         if (info->relocatable)
+           bfd_set_error (bfd_error_invalid_operation);
+         else
+           bfd_set_error (bfd_error_wrong_format);
+         goto error_return;
+       }
+    }
+
+  if (! dynamic)
+    {
+      /* If we are creating a shared library, create all the dynamic
+        sections immediately.  We need to attach them to something,
+        so we attach them to this BFD, provided it is the right
+        format.  FIXME: If there are no input BFD's of the same
+        format as the output, we can't make a shared library.  */
+      if (info->shared
+         && is_elf_hash_table (htab)
+         && info->output_bfd->xvec == abfd->xvec
+         && !htab->dynamic_sections_created)
+       {
+         if (! elf64_ia64_create_dynamic_sections (abfd, info))
+           goto error_return;
+       }
+    }
+  else if (!is_elf_hash_table (htab))
+    goto error_return;
+  else
+    {
+      asection *s;
+      bfd_byte *dynbuf;
+      bfd_byte *extdyn;
+
+      /* ld --just-symbols and dynamic objects don't mix very well.
+        ld shouldn't allow it.  */
+      if ((s = abfd->sections) != NULL
+         && s->sec_info_type == ELF_INFO_TYPE_JUST_SYMS)
+       abort ();
+
+      /* Be sure there are dynamic sections.  */
+      if (! elf64_ia64_create_dynamic_sections (htab->dynobj, info))
+        goto error_return;
+
+      s = bfd_get_section_by_name (abfd, ".dynamic");
+      if (s == NULL)
+        {
+          /* VMS libraries do not have dynamic sections.  Create one from
+             the segment.  */
+          Elf_Internal_Phdr *phdr;
+          unsigned int i, phnum;
+
+          phdr = elf_tdata (abfd)->phdr;
+          if (phdr == NULL)
+            goto error_return;
+          phnum = elf_elfheader (abfd)->e_phnum;
+          for (i = 0; i < phnum; phdr++)
+            if (phdr->p_type == PT_DYNAMIC)
+              {
+                s = bfd_make_section (abfd, ".dynamic");
+                if (s == NULL)
+                  goto error_return;
+                s->vma = phdr->p_vaddr;
+                s->lma = phdr->p_paddr;
+                s->size = phdr->p_filesz;
+                s->filepos = phdr->p_offset;
+                s->flags |= SEC_HAS_CONTENTS;
+                s->alignment_power = bfd_log2 (phdr->p_align);
+                break;
+              }
+          if (s == NULL)
+            goto error_return;
+        }
+
+      /* Extract IDENT.  */
+      if (!bfd_malloc_and_get_section (abfd, s, &dynbuf))
+        {
+error_free_dyn:
+          free (dynbuf);
+          goto error_return;
+        }
+
+      for (extdyn = dynbuf;
+           extdyn < dynbuf + s->size;
+           extdyn += bed->s->sizeof_dyn)
+        {
+          Elf_Internal_Dyn dyn;
+
+          bed->s->swap_dyn_in (abfd, extdyn, &dyn);
+          if (dyn.d_tag == DT_IA_64_VMS_IDENT)
+            {
+              bfd_uint64_t tagv = dyn.d_un.d_val;
+              elf_ia64_vms_ident (abfd) = tagv;
+              break;
+            }
+        }
+      if (extdyn >= dynbuf + s->size)
+        {
+          /* Ident not found.  */
+          goto error_free_dyn;
+        }
+      free (dynbuf);
+
+      /* We do not want to include any of the sections in a dynamic
+        object in the output file.  We hack by simply clobbering the
+        list of sections in the BFD.  This could be handled more
+        cleanly by, say, a new section flag; the existing
+        SEC_NEVER_LOAD flag is not the one we want, because that one
+        still implies that the section takes up space in the output
+        file.  */
+      bfd_section_list_clear (abfd);
+
+      /* FIXME: should we detect if this library is already included ?
+         This should be harmless and shouldn't happen in practice.  */
+    }
+
+  hdr = &elf_tdata (abfd)->symtab_hdr;
+  symcount = hdr->sh_size / bed->s->sizeof_sym;
+
+  /* The sh_info field of the symtab header tells us where the
+     external symbols start.  We don't care about the local symbols at
+     this point.  */
+  extsymcount = symcount - hdr->sh_info;
+  extsymoff = hdr->sh_info;
+
+  sym_hash = NULL;
+  if (extsymcount != 0)
+    {
+      isymbuf = bfd_elf_get_elf_syms (abfd, hdr, extsymcount, extsymoff,
+                                     NULL, NULL, NULL);
+      if (isymbuf == NULL)
+       goto error_return;
+
+      /* We store a pointer to the hash table entry for each external
+        symbol.  */
+      amt = extsymcount * sizeof (struct elf_link_hash_entry *);
+      sym_hash = (struct elf_link_hash_entry **) bfd_alloc (abfd, amt);
+      if (sym_hash == NULL)
+       goto error_free_sym;
+      elf_sym_hashes (abfd) = sym_hash;
+    }
+
+  for (isym = isymbuf, isymend = isymbuf + extsymcount;
+       isym < isymend;
+       isym++, sym_hash++)
+    {
+      int bind;
+      bfd_vma value;
+      asection *sec, *new_sec;
+      flagword flags;
+      const char *name;
+      struct elf_link_hash_entry *h;
+      bfd_boolean definition;
+      bfd_boolean size_change_ok;
+      bfd_boolean type_change_ok;
+      bfd_boolean common;
+      unsigned int old_alignment;
+      bfd *old_bfd;
+
+      flags = BSF_NO_FLAGS;
+      sec = NULL;
+      value = isym->st_value;
+      *sym_hash = NULL;
+      common = bed->common_definition (isym);
+
+      bind = ELF_ST_BIND (isym->st_info);
+      switch (bind)
+       {
+       case STB_LOCAL:
+         /* This should be impossible, since ELF requires that all
+            global symbols follow all local symbols, and that sh_info
+            point to the first global symbol.  Unfortunately, Irix 5
+            screws this up.  */
+         continue;
+
+       case STB_GLOBAL:
+         if (isym->st_shndx != SHN_UNDEF && !common)
+           flags = BSF_GLOBAL;
+         break;
+
+       case STB_WEAK:
+         flags = BSF_WEAK;
+         break;
+
+       case STB_GNU_UNIQUE:
+         flags = BSF_GNU_UNIQUE;
+         break;
+
+       default:
+         /* Leave it up to the processor backend.  */
+         break;
+       }
+
+      if (isym->st_shndx == SHN_UNDEF)
+       sec = bfd_und_section_ptr;
+      else if (isym->st_shndx == SHN_ABS)
+       sec = bfd_abs_section_ptr;
+      else if (isym->st_shndx == SHN_COMMON)
+       {
+         sec = bfd_com_section_ptr;
+         /* What ELF calls the size we call the value.  What ELF
+            calls the value we call the alignment.  */
+         value = isym->st_size;
+       }
+      else
+       {
+         sec = bfd_section_from_elf_index (abfd, isym->st_shndx);
+         if (sec == NULL)
+           sec = bfd_abs_section_ptr;
+         else if (sec->kept_section)
+           {
+             /* Symbols from discarded section are undefined.  We keep
+                its visibility.  */
+             sec = bfd_und_section_ptr;
+             isym->st_shndx = SHN_UNDEF;
+           }
+         else if ((abfd->flags & (EXEC_P | DYNAMIC)) != 0)
+           value -= sec->vma;
+       }
+
+      name = bfd_elf_string_from_elf_section (abfd, hdr->sh_link,
+                                             isym->st_name);
+      if (name == NULL)
+       goto error_free_vers;
+
+      if (bed->elf_add_symbol_hook)
+       {
+         if (! (*bed->elf_add_symbol_hook) (abfd, info, isym, &name, &flags,
+                                            &sec, &value))
+           goto error_free_vers;
+
+         /* The hook function sets the name to NULL if this symbol
+            should be skipped for some reason.  */
+         if (name == NULL)
+           continue;
+       }
+
+      /* Sanity check that all possibilities were handled.  */
+      if (sec == NULL)
+       {
+         bfd_set_error (bfd_error_bad_value);
+         goto error_free_vers;
+       }
+
+      if (bfd_is_und_section (sec)
+         || bfd_is_com_section (sec))
+       definition = FALSE;
+      else
+       definition = TRUE;
+
+      size_change_ok = FALSE;
+      type_change_ok = bed->type_change_ok;
+      old_alignment = 0;
+      old_bfd = NULL;
+      new_sec = sec;
+
+      if (! bfd_is_und_section (sec))
+        h = elf_link_hash_lookup (htab, name, TRUE, FALSE, FALSE);
+      else
+        h = ((struct elf_link_hash_entry *) bfd_wrapped_link_hash_lookup
+             (abfd, info, name, TRUE, FALSE, FALSE));
+      if (h == NULL)
+        goto error_free_sym;
+
+      *sym_hash = h;
+
+      if (is_elf_hash_table (htab))
+       {
+         while (h->root.type == bfd_link_hash_indirect
+                || h->root.type == bfd_link_hash_warning)
+           h = (struct elf_link_hash_entry *) h->root.u.i.link;
+
+         /* Remember the old alignment if this is a common symbol, so
+            that we don't reduce the alignment later on.  We can't
+            check later, because _bfd_generic_link_add_one_symbol
+            will set a default for the alignment which we want to
+            override. We also remember the old bfd where the existing
+            definition comes from.  */
+         switch (h->root.type)
+           {
+           default:
+             break;
+
+           case bfd_link_hash_defined:
+              if (abfd->selective_search)
+                continue;
+              /* Fall-through.  */
+           case bfd_link_hash_defweak:
+             old_bfd = h->root.u.def.section->owner;
+             break;
+
+           case bfd_link_hash_common:
+             old_bfd = h->root.u.c.p->section->owner;
+             old_alignment = h->root.u.c.p->alignment_power;
+             break;
+           }
+       }
+
+      if (! (_bfd_generic_link_add_one_symbol
+            (info, abfd, name, flags, sec, value, NULL, FALSE, bed->collect,
+             (struct bfd_link_hash_entry **) sym_hash)))
+       goto error_free_vers;
+
+      h = *sym_hash;
+      while (h->root.type == bfd_link_hash_indirect
+            || h->root.type == bfd_link_hash_warning)
+       h = (struct elf_link_hash_entry *) h->root.u.i.link;
+
+      *sym_hash = h;
+      h->unique_global = (flags & BSF_GNU_UNIQUE) != 0;
+
+      /* Set the alignment of a common symbol.  */
+      if ((common || bfd_is_com_section (sec))
+         && h->root.type == bfd_link_hash_common)
+       {
+         unsigned int align;
+
+         if (common)
+           align = bfd_log2 (isym->st_value);
+         else
+           {
+             /* The new symbol is a common symbol in a shared object.
+                We need to get the alignment from the section.  */
+             align = new_sec->alignment_power;
+           }
+         if (align > old_alignment
+             /* Permit an alignment power of zero if an alignment of one
+                is specified and no other alignments have been specified.  */
+             || (isym->st_value == 1 && old_alignment == 0))
+           h->root.u.c.p->alignment_power = align;
+         else
+           h->root.u.c.p->alignment_power = old_alignment;
+       }
+
+      if (is_elf_hash_table (htab))
+       {
+         /* Check the alignment when a common symbol is involved. This
+            can change when a common symbol is overridden by a normal
+            definition or a common symbol is ignored due to the old
+            normal definition. We need to make sure the maximum
+            alignment is maintained.  */
+         if ((old_alignment || common)
+             && h->root.type != bfd_link_hash_common)
+           {
+             unsigned int common_align;
+             unsigned int normal_align;
+             unsigned int symbol_align;
+             bfd *normal_bfd;
+             bfd *common_bfd;
+
+             symbol_align = ffs (h->root.u.def.value) - 1;
+             if (h->root.u.def.section->owner != NULL
+                 && (h->root.u.def.section->owner->flags & DYNAMIC) == 0)
+               {
+                 normal_align = h->root.u.def.section->alignment_power;
+                 if (normal_align > symbol_align)
+                   normal_align = symbol_align;
+               }
+             else
+               normal_align = symbol_align;
+
+             if (old_alignment)
+               {
+                 common_align = old_alignment;
+                 common_bfd = old_bfd;
+                 normal_bfd = abfd;
+               }
+             else
+               {
+                 common_align = bfd_log2 (isym->st_value);
+                 common_bfd = abfd;
+                 normal_bfd = old_bfd;
+               }
+
+             if (normal_align < common_align)
+               {
+                 /* PR binutils/2735 */
+                 if (normal_bfd == NULL)
+                   (*_bfd_error_handler)
+                     (_("Warning: alignment %u of common symbol `%s' in %B"
+                        " is greater than the alignment (%u) of its section %A"),
+                      common_bfd, h->root.u.def.section,
+                      1 << common_align, name, 1 << normal_align);
+                 else
+                   (*_bfd_error_handler)
+                     (_("Warning: alignment %u of symbol `%s' in %B"
+                        " is smaller than %u in %B"),
+                      normal_bfd, common_bfd,
+                      1 << normal_align, name, 1 << common_align);
+               }
+           }
+
+         /* Remember the symbol size if it isn't undefined.  */
+         if ((isym->st_size != 0 && isym->st_shndx != SHN_UNDEF)
+             && (definition || h->size == 0))
+           {
+             if (h->size != 0
+                 && h->size != isym->st_size
+                 && ! size_change_ok)
+               (*_bfd_error_handler)
+                 (_("Warning: size of symbol `%s' changed"
+                    " from %lu in %B to %lu in %B"),
+                  old_bfd, abfd,
+                  name, (unsigned long) h->size,
+                  (unsigned long) isym->st_size);
+
+             h->size = isym->st_size;
+           }
+
+         /* If this is a common symbol, then we always want H->SIZE
+            to be the size of the common symbol.  The code just above
+            won't fix the size if a common symbol becomes larger.  We
+            don't warn about a size change here, because that is
+            covered by --warn-common.  Allow changed between different
+            function types.  */
+         if (h->root.type == bfd_link_hash_common)
+           h->size = h->root.u.c.size;
+
+         if (ELF_ST_TYPE (isym->st_info) != STT_NOTYPE
+             && (definition || h->type == STT_NOTYPE))
+           {
+             unsigned int type = ELF_ST_TYPE (isym->st_info);
+
+             if (h->type != type)
+               {
+                 if (h->type != STT_NOTYPE && ! type_change_ok)
+                   (*_bfd_error_handler)
+                     (_("Warning: type of symbol `%s' changed"
+                        " from %d to %d in %B"),
+                      abfd, name, h->type, type);
+
+                 h->type = type;
+               }
+           }
+
+         /* Set a flag in the hash table entry indicating the type of
+            reference or definition we just found.  Keep a count of
+            the number of dynamic symbols we find.  A dynamic symbol
+            is one which is referenced or defined by both a regular
+            object and a shared object.  */
+         if (! dynamic)
+           {
+             if (! definition)
+               {
+                 h->ref_regular = 1;
+                 if (bind != STB_WEAK)
+                   h->ref_regular_nonweak = 1;
+               }
+             else
+               {
+                  BFD_ASSERT (!h->def_dynamic);
+                 h->def_regular = 1;
+               }
+           }
+         else
+           {
+             BFD_ASSERT (definition);
+              h->def_dynamic = 1;
+              h->dynindx = -2;
+              ((struct elf64_ia64_link_hash_entry *)h)->shl = abfd;
+           }
+       }
+    }
+
+  if (isymbuf != NULL)
+    {
+      free (isymbuf);
+      isymbuf = NULL;
+    }
+
+  /* If this object is the same format as the output object, and it is
+     not a shared library, then let the backend look through the
+     relocs.
+
+     This is required to build global offset table entries and to
+     arrange for dynamic relocs.  It is not required for the
+     particular common case of linking non PIC code, even when linking
+     against shared libraries, but unfortunately there is no way of
+     knowing whether an object file has been compiled PIC or not.
+     Looking through the relocs is not particularly time consuming.
+     The problem is that we must either (1) keep the relocs in memory,
+     which causes the linker to require additional runtime memory or
+     (2) read the relocs twice from the input file, which wastes time.
+     This would be a good case for using mmap.
+
+     I have no idea how to handle linking PIC code into a file of a
+     different format.  It probably can't be done.  */
+  if (! dynamic
+      && is_elf_hash_table (htab)
+      && bed->check_relocs != NULL
+      && (*bed->relocs_compatible) (abfd->xvec, info->output_bfd->xvec))
+    {
+      asection *o;
+
+      for (o = abfd->sections; o != NULL; o = o->next)
+       {
+         Elf_Internal_Rela *internal_relocs;
+         bfd_boolean ok;
+
+         if ((o->flags & SEC_RELOC) == 0
+             || o->reloc_count == 0
+             || ((info->strip == strip_all || info->strip == strip_debugger)
+                 && (o->flags & SEC_DEBUGGING) != 0)
+             || bfd_is_abs_section (o->output_section))
+           continue;
+
+         internal_relocs = _bfd_elf_link_read_relocs (abfd, o, NULL, NULL,
+                                                      info->keep_memory);
+         if (internal_relocs == NULL)
+           goto error_return;
+
+         ok = (*bed->check_relocs) (abfd, info, o, internal_relocs);
+
+         if (elf_section_data (o)->relocs != internal_relocs)
+           free (internal_relocs);
+
+         if (! ok)
+           goto error_return;
+       }
+    }
+
+  return TRUE;
+
+ error_free_vers:
+ error_free_sym:
+  if (isymbuf != NULL)
+    free (isymbuf);
+ error_return:
+  return FALSE;
+}
+
+static bfd_boolean
+elf64_vms_link_add_archive_symbols (bfd *abfd, struct bfd_link_info *info)
+{
+  int pass;
+  struct bfd_link_hash_entry **pundef;
+  struct bfd_link_hash_entry **next_pundef;
+
+  /* We only accept VMS libraries.  */
+  if (info->output_bfd->xvec != abfd->xvec)
+    {
+      bfd_set_error (bfd_error_wrong_format);
+      return FALSE;
+    }
+
+  /* The archive_pass field in the archive itself is used to
+     initialize PASS, since we may search the same archive multiple
+     times.  */
+  pass = ++abfd->archive_pass;
+
+  /* Look through the list of undefined symbols.  */
+  for (pundef = &info->hash->undefs; *pundef != NULL; pundef = next_pundef)
+    {
+      struct bfd_link_hash_entry *h;
+      symindex symidx;
+      bfd *element;
+      bfd *orig_element;
+
+      h = *pundef;
+      next_pundef = &(*pundef)->u.undef.next;
+
+      /* When a symbol is defined, it is not necessarily removed from
+        the list.  */
+      if (h->type != bfd_link_hash_undefined
+         && h->type != bfd_link_hash_common)
+       {
+         /* Remove this entry from the list, for general cleanliness
+            and because we are going to look through the list again
+            if we search any more libraries.  We can't remove the
+            entry if it is the tail, because that would lose any
+            entries we add to the list later on.  */
+         if (*pundef != info->hash->undefs_tail)
+            {
+              *pundef = *next_pundef;
+              next_pundef = pundef;
+            }
+         continue;
+       }
+
+      /* Look for this symbol in the archive hash table.  */
+      symidx = _bfd_vms_lib_find_symbol (abfd, h->root.string);
+      if (symidx == BFD_NO_MORE_SYMBOLS)
+       {
+         /* Nothing in this slot.  */
+         continue;
+       }
+
+      element = bfd_get_elt_at_index (abfd, symidx);
+      if (element == NULL)
+       return FALSE;
+
+      if (element->archive_pass == -1 || element->archive_pass == pass)
+        {
+          /* Next symbol if this archive is wrong or already handled.  */
+          continue;
+        }
+
+      orig_element = element;
+      if (bfd_is_thin_archive (abfd))
+        {
+          element = _bfd_vms_lib_get_imagelib_file (element);
+          if (element == NULL || !bfd_check_format (element, bfd_object))
+            {
+              orig_element->archive_pass = -1;
+              return FALSE;
+            }
+        }
+      else if (! bfd_check_format (element, bfd_object))
+        {
+          element->archive_pass = -1;
+          return FALSE;
+        }
+
+      /* Unlike the generic linker, we know that this element provides
+        a definition for an undefined symbol and we know that we want
+        to include it.  We don't need to check anything.  */
+      if (! (*info->callbacks->add_archive_element) (info, element,
+                                                     h->root.string, &element))
+       return FALSE;
+      if (! elf64_vms_link_add_object_symbols (element, info))
+       return FALSE;
+
+      orig_element->archive_pass = pass;
+    }
+
+  return TRUE;
+}
+
+static bfd_boolean
+elf64_vms_bfd_link_add_symbols (bfd *abfd, struct bfd_link_info *info)
+{
+  switch (bfd_get_format (abfd))
+    {
+    case bfd_object:
+      return elf64_vms_link_add_object_symbols (abfd, info);
+      break;
+    case bfd_archive:
+      return elf64_vms_link_add_archive_symbols (abfd, info);
+      break;
+    default:
+      bfd_set_error (bfd_error_wrong_format);
+      return FALSE;
+    }
+}
+
+static bfd_boolean
+elf64_ia64_vms_mkobject (bfd *abfd)
+{
+  return bfd_elf_allocate_object
+    (abfd, sizeof (struct elf64_ia64_vms_obj_tdata), IA64_ELF_DATA);
+}
+
+
+/* Size-dependent data and functions.  */
+static const struct elf_size_info elf64_ia64_vms_size_info = {
+  sizeof (Elf64_External_VMS_Ehdr),
+  sizeof (Elf64_External_Phdr),
+  sizeof (Elf64_External_Shdr),
+  sizeof (Elf64_External_Rel),
+  sizeof (Elf64_External_Rela),
+  sizeof (Elf64_External_Sym),
+  sizeof (Elf64_External_Dyn),
+  sizeof (Elf_External_Note),
+  4,
+  1,
+  64, 3, /* ARCH_SIZE, LOG_FILE_ALIGN */
+  ELFCLASS64, EV_CURRENT,
+  bfd_elf64_write_out_phdrs,
+  elf64_vms_write_shdrs_and_ehdr,
+  bfd_elf64_checksum_contents,
+  bfd_elf64_write_relocs,
+  bfd_elf64_swap_symbol_in,
+  bfd_elf64_swap_symbol_out,
+  bfd_elf64_slurp_reloc_table,
+  bfd_elf64_slurp_symbol_table,
+  bfd_elf64_swap_dyn_in,
+  bfd_elf64_swap_dyn_out,
+  bfd_elf64_swap_reloc_in,
+  bfd_elf64_swap_reloc_out,
+  bfd_elf64_swap_reloca_in,
+  bfd_elf64_swap_reloca_out
+};
+
+#define ELF_ARCH                       bfd_arch_ia64
+#define ELF_MACHINE_CODE               EM_IA_64
+#define ELF_MAXPAGESIZE                        0x10000 /* 64KB */
+#define ELF_COMMONPAGESIZE             0x200   /* 16KB */
+
+#define elf_backend_section_from_shdr \
+       elf64_ia64_section_from_shdr
+#define elf_backend_section_flags \
+       elf64_ia64_section_flags
+#define elf_backend_fake_sections \
+       elf64_ia64_fake_sections
+#define elf_backend_final_write_processing \
+       elf64_ia64_final_write_processing
+#define elf_backend_add_symbol_hook \
+       elf64_ia64_add_symbol_hook
+#define elf_info_to_howto \
+       elf64_ia64_info_to_howto
+
+#define bfd_elf64_bfd_reloc_type_lookup \
+       ia64_elf_reloc_type_lookup
+#define bfd_elf64_bfd_reloc_name_lookup \
+       ia64_elf_reloc_name_lookup
+#define bfd_elf64_bfd_is_local_label_name \
+       elf64_ia64_is_local_label_name
+#define bfd_elf64_bfd_relax_section \
+       elf64_ia64_relax_section
+
+#define elf_backend_object_p \
+       elf64_ia64_object_p
+
+/* Stuff for the BFD linker: */
+#define bfd_elf64_bfd_link_hash_table_create \
+       elf64_ia64_hash_table_create
+#define bfd_elf64_bfd_link_hash_table_free \
+       elf64_ia64_hash_table_free
+#define elf_backend_create_dynamic_sections \
+       elf64_ia64_create_dynamic_sections
+#define elf_backend_check_relocs \
+       elf64_ia64_check_relocs
+#define elf_backend_adjust_dynamic_symbol \
+       elf64_ia64_adjust_dynamic_symbol
+#define elf_backend_size_dynamic_sections \
+       elf64_ia64_size_dynamic_sections
+#define elf_backend_omit_section_dynsym \
+  ((bfd_boolean (*) (bfd *, struct bfd_link_info *, asection *)) bfd_true)
+#define elf_backend_relocate_section \
+       elf64_ia64_relocate_section
+#define elf_backend_finish_dynamic_symbol \
+       elf64_ia64_finish_dynamic_symbol
+#define elf_backend_finish_dynamic_sections \
+       elf64_ia64_finish_dynamic_sections
+#define bfd_elf64_bfd_final_link \
+       elf64_ia64_final_link
+
+#define bfd_elf64_bfd_merge_private_bfd_data \
+       elf64_ia64_merge_private_bfd_data
+#define bfd_elf64_bfd_set_private_flags \
+       elf64_ia64_set_private_flags
+#define bfd_elf64_bfd_print_private_bfd_data \
+       elf64_ia64_print_private_bfd_data
+
+#define elf_backend_plt_readonly       1
+#define elf_backend_want_plt_sym       0
+#define elf_backend_plt_alignment      5
+#define elf_backend_got_header_size    0
+#define elf_backend_want_got_plt       1
+#define elf_backend_may_use_rel_p      1
+#define elf_backend_may_use_rela_p     1
+#define elf_backend_default_use_rela_p 1
+#define elf_backend_want_dynbss                0
+#define elf_backend_hide_symbol                elf64_ia64_hash_hide_symbol
+#define elf_backend_fixup_symbol       _bfd_elf_link_hash_fixup_symbol
+#define elf_backend_reloc_type_class   elf64_ia64_reloc_type_class
+#define elf_backend_rela_normal                1
+#define elf_backend_special_sections   elf64_ia64_special_sections
+#define elf_backend_default_execstack  0
+
+/* FIXME: PR 290: The Intel C compiler generates SHT_IA_64_UNWIND with
+   SHF_LINK_ORDER. But it doesn't set the sh_link or sh_info fields.
+   We don't want to flood users with so many error messages. We turn
+   off the warning for now. It will be turned on later when the Intel
+   compiler is fixed.   */
+#define elf_backend_link_order_error_handler NULL
+
+/* VMS-specific vectors.  */
+
+#undef  TARGET_LITTLE_SYM
+#define TARGET_LITTLE_SYM              bfd_elf64_ia64_vms_vec
+#undef  TARGET_LITTLE_NAME
+#define TARGET_LITTLE_NAME             "elf64-ia64-vms"
+#undef  TARGET_BIG_SYM
+#undef  TARGET_BIG_NAME
+
+/* These are VMS specific functions.  */
+
+#undef  elf_backend_object_p
+#define elf_backend_object_p elf64_vms_object_p
+
+#undef  elf_backend_section_from_shdr
+#define elf_backend_section_from_shdr elf64_vms_section_from_shdr
+
+#undef  elf_backend_post_process_headers
+#define elf_backend_post_process_headers elf64_vms_post_process_headers
+
+#undef  elf_backend_section_processing
+#define elf_backend_section_processing elf64_vms_section_processing
+
+#undef  elf_backend_final_write_processing
+#define elf_backend_final_write_processing elf64_vms_final_write_processing
+
+#undef  bfd_elf64_close_and_cleanup
+#define bfd_elf64_close_and_cleanup elf64_vms_close_and_cleanup
+
+#undef  elf_backend_section_from_bfd_section
+
+#undef  elf_backend_symbol_processing
+
+#undef  elf_backend_want_p_paddr_set_to_zero
+
+#undef  ELF_OSABI
+#define ELF_OSABI                      ELFOSABI_OPENVMS
+
+#undef  ELF_MAXPAGESIZE
+#define ELF_MAXPAGESIZE                        0x10000 /* 64KB */
+
+#undef  elf64_bed
+#define elf64_bed elf64_ia64_vms_bed
+
+#define elf_backend_size_info elf64_ia64_vms_size_info
+
+/* Use VMS-style archives (in particular, don't use the standard coff
+   archive format).  */
+#define bfd_elf64_archive_functions
+
+#undef bfd_elf64_archive_p
+#define bfd_elf64_archive_p _bfd_vms_lib_ia64_archive_p
+#undef bfd_elf64_write_archive_contents
+#define bfd_elf64_write_archive_contents _bfd_vms_lib_write_archive_contents
+#undef bfd_elf64_mkarchive
+#define bfd_elf64_mkarchive _bfd_vms_lib_ia64_mkarchive
+
+#define bfd_elf64_archive_slurp_armap \
+  _bfd_vms_lib_slurp_armap
+#define bfd_elf64_archive_slurp_extended_name_table \
+  _bfd_vms_lib_slurp_extended_name_table
+#define bfd_elf64_archive_construct_extended_name_table \
+  _bfd_vms_lib_construct_extended_name_table
+#define bfd_elf64_archive_truncate_arname \
+  _bfd_vms_lib_truncate_arname
+#define bfd_elf64_archive_write_armap \
+  _bfd_vms_lib_write_armap
+#define bfd_elf64_archive_read_ar_hdr \
+  _bfd_vms_lib_read_ar_hdr
+#define bfd_elf64_archive_write_ar_hdr \
+  _bfd_vms_lib_write_ar_hdr
+#define bfd_elf64_archive_openr_next_archived_file \
+  _bfd_vms_lib_openr_next_archived_file
+#define bfd_elf64_archive_get_elt_at_index \
+  _bfd_vms_lib_get_elt_at_index
+#define bfd_elf64_archive_generic_stat_arch_elt \
+  _bfd_vms_lib_generic_stat_arch_elt
+#define bfd_elf64_archive_update_armap_timestamp \
+  _bfd_vms_lib_update_armap_timestamp
+
+/* VMS link methods.  */
+#undef  bfd_elf64_bfd_link_add_symbols
+#define bfd_elf64_bfd_link_add_symbols         elf64_vms_bfd_link_add_symbols
+
+#undef  elf_backend_want_got_sym
+#define elf_backend_want_got_sym       0
+
+#undef  bfd_elf64_mkobject
+#define bfd_elf64_mkobject             elf64_ia64_vms_mkobject
+
+/* Redefine to align segments on block size.  */
+#undef  ELF_MAXPAGESIZE
+#define ELF_MAXPAGESIZE                        0x200 /* 512B  */
+
+#undef  elf_backend_want_got_plt
+#define elf_backend_want_got_plt       0
+
+#include "elf64-target.h"
index 12025a3b053745df1a8e29a9d88f91ae66eded95..0ed52081999d43d5237185dd36e229ffe43ae9ff 100644 (file)
@@ -8908,7 +8908,8 @@ elf_link_output_extsym (struct bfd_hash_entry *bh, void *data)
   /* If this symbol should be put in the .dynsym section, then put it
      there now.  We already know the symbol index.  We also fill in
      the entry in the .hash section.  */
-  if (h->dynindx != -1
+  if (finfo->dynsym_sec != NULL
+      && h->dynindx != -1
       && elf_hash_table (finfo->info)->dynamic_sections_created)
     {
       bfd_byte *esym;
@@ -10305,7 +10306,7 @@ bfd_elf_final_link (bfd *abfd, struct bfd_link_info *info)
     {
       finfo.dynsym_sec = bfd_get_section_by_name (dynobj, ".dynsym");
       finfo.hash_sec = bfd_get_section_by_name (dynobj, ".hash");
-      BFD_ASSERT (finfo.dynsym_sec != NULL);
+      /* Note that dynsym_sec can be NULL (on VMS).  */
       finfo.symver_sec = bfd_get_section_by_name (dynobj, ".gnu.version");
       /* Note that it is OK if symver_sec is NULL.  */
     }
@@ -10825,6 +10826,7 @@ bfd_elf_final_link (bfd *abfd, struct bfd_link_info *info)
   symtab_hdr->sh_info = bfd_get_symcount (abfd);
 
   if (dynamic
+      && finfo.dynsym_sec != NULL
       && finfo.dynsym_sec->output_section != bfd_abs_section_ptr)
     {
       Elf_Internal_Sym sym;
index d70b28e9d33fc2ba15de5f9663ef8c62f9904e1f..7ea2981e380e9c9c2bcf91dc0d93bcb937726e2c 100644 (file)
@@ -1,6 +1,6 @@
 /* IA-64 support for 64-bit ELF
    Copyright 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007,
-   2008, 2009, 2010, 2011  Free Software Foundation, Inc.
+   2008, 2009, 2010, 2011, 2012  Free Software Foundation, Inc.
    Contributed by David Mosberger-Tang <davidm@hpl.hp.com>
 
    This file is part of BFD, the Binary File Descriptor library.
 #define        LOG_SECTION_ALIGN       2
 #endif
 
-/* Only add code for vms when the vms target is enabled.  This is required
-   because it depends on vms-lib.c for its archive format and we don't want
-   to compile that code if it is not used.  */
-#if ARCH_SIZE == 64 && \
-  (defined (HAVE_bfd_elf64_ia64_vms_vec) || defined (HAVE_all_vecs))
-#define INCLUDE_IA64_VMS
-#endif
-
 typedef struct bfd_hash_entry *(*new_hash_entry_func)
   (struct bfd_hash_entry *, struct bfd_hash_table *, const char *);
 
@@ -4978,277 +4970,6 @@ elfNN_hpux_backend_symbol_processing (bfd *abfd ATTRIBUTE_UNUSED,
       break;
     }
 }
-
-#ifdef INCLUDE_IA64_VMS
-
-static bfd_boolean
-elfNN_vms_section_from_shdr (bfd *abfd,
-                            Elf_Internal_Shdr *hdr,
-                            const char *name,
-                            int shindex)
-{
-  switch (hdr->sh_type)
-    {
-    case SHT_IA_64_VMS_TRACE:
-    case SHT_IA_64_VMS_DEBUG:
-    case SHT_IA_64_VMS_DEBUG_STR:
-      break;
-
-    default:
-      return elfNN_ia64_section_from_shdr (abfd, hdr, name, shindex);
-    }
-
-  if (! _bfd_elf_make_section_from_shdr (abfd, hdr, name, shindex))
-    return FALSE;
-
-  return TRUE;
-}
-
-static bfd_boolean
-elfNN_vms_object_p (bfd *abfd)
-{
-  Elf_Internal_Ehdr *i_ehdrp = elf_elfheader (abfd);
-  Elf_Internal_Phdr *i_phdr = elf_tdata (abfd)->phdr;
-  unsigned int i;
-  unsigned int num_text = 0;
-  unsigned int num_data = 0;
-  unsigned int num_rodata = 0;
-  char name[16];
-
-  if (!elfNN_ia64_object_p (abfd))
-    return FALSE;
-
-  for (i = 0; i < i_ehdrp->e_phnum; i++, i_phdr++)
-    {
-      /* Is there a section for this segment?  */
-      bfd_vma base_vma = i_phdr->p_vaddr;
-      bfd_vma limit_vma = base_vma + i_phdr->p_filesz;
-
-      if (i_phdr->p_type != PT_LOAD)
-       continue;
-
-    again:
-      while (base_vma < limit_vma)
-       {
-         bfd_vma next_vma = limit_vma;
-         asection *nsec;
-         asection *sec;
-         flagword flags;
-         char *nname = NULL;
-
-         /* Find a section covering [base_vma;limit_vma)  */
-         for (sec = abfd->sections; sec != NULL; sec = sec->next)
-           {
-             /* Skip uninteresting sections (either not in memory or
-                below base_vma.  */
-             if ((sec->flags & (SEC_ALLOC | SEC_LOAD)) == 0
-                 || sec->vma + sec->size <= base_vma)
-               continue;
-             if (sec->vma <= base_vma)
-               {
-                 /* This section covers (maybe partially) the beginning
-                    of the range.  */
-                 base_vma = sec->vma + sec->size;
-                 goto again;
-               }
-             if (sec->vma < next_vma)
-               {
-                 /* This section partially covers the end of the range.
-                    Used to compute the size of the hole.  */
-                 next_vma = sec->vma;
-               }
-           }
-
-         /* No section covering [base_vma; next_vma).  Create a fake one.  */
-         flags = SEC_ALLOC | SEC_LOAD | SEC_HAS_CONTENTS;
-         if (i_phdr->p_flags & PF_X)
-           {
-             flags |= SEC_CODE;
-             if (num_text++ == 0)
-               nname = ".text";
-             else
-               sprintf (name, ".text$%u", num_text);
-           }
-         else if ((i_phdr->p_flags & (PF_R | PF_W)) == PF_R)
-           {
-             flags |= SEC_READONLY;
-             sprintf (name, ".rodata$%u", num_rodata++);
-           }
-         else
-           {
-             flags |= SEC_DATA;
-             sprintf (name, ".data$%u", num_data++);
-           }
-
-         /* Allocate name.  */
-         if (nname == NULL)
-           {
-             size_t name_len = strlen (name) + 1;
-             nname = bfd_alloc (abfd, name_len);
-             if (nname == NULL)
-               return FALSE;
-             memcpy (nname, name, name_len);
-           }
-
-         /* Create and fill new section.  */
-         nsec = bfd_make_section_anyway_with_flags (abfd, nname, flags);
-         if (nsec == NULL)
-           return FALSE;
-         nsec->vma = base_vma;
-         nsec->size = next_vma - base_vma;
-         nsec->filepos = i_phdr->p_offset + (base_vma - i_phdr->p_vaddr);
-
-         base_vma = next_vma;
-       }
-    }
-  return TRUE;
-}
-
-static void
-elfNN_vms_post_process_headers (bfd *abfd,
-                               struct bfd_link_info *info ATTRIBUTE_UNUSED)
-{
-  Elf_Internal_Ehdr *i_ehdrp = elf_elfheader (abfd);
-
-  i_ehdrp->e_ident[EI_OSABI] = ELFOSABI_OPENVMS;
-  i_ehdrp->e_ident[EI_ABIVERSION] = 2;
-}
-
-static bfd_boolean
-elfNN_vms_section_processing (bfd *abfd ATTRIBUTE_UNUSED,
-                             Elf_Internal_Shdr *hdr)
-{
-  if (hdr->bfd_section != NULL)
-    {
-      const char *name = bfd_get_section_name (abfd, hdr->bfd_section);
-
-      if (strcmp (name, ".text") == 0)
-       hdr->sh_flags |= SHF_IA_64_VMS_SHARED;
-      else if ((strcmp (name, ".debug") == 0)
-           || (strcmp (name, ".debug_abbrev") == 0)
-           || (strcmp (name, ".debug_aranges") == 0)
-           || (strcmp (name, ".debug_frame") == 0)
-           || (strcmp (name, ".debug_info") == 0)
-           || (strcmp (name, ".debug_loc") == 0)
-           || (strcmp (name, ".debug_macinfo") == 0)
-           || (strcmp (name, ".debug_pubnames") == 0)
-           || (strcmp (name, ".debug_pubtypes") == 0))
-       hdr->sh_type = SHT_IA_64_VMS_DEBUG;
-      else if ((strcmp (name, ".debug_line") == 0)
-           || (strcmp (name, ".debug_ranges") == 0))
-       hdr->sh_type = SHT_IA_64_VMS_TRACE;
-      else if (strcmp (name, ".debug_str") == 0)
-       hdr->sh_type = SHT_IA_64_VMS_DEBUG_STR;
-      else if (strcmp (name, ".vms_display_name_info") == 0)
-       {
-         int idx, symcount;
-         asymbol **syms;
-         struct elf_obj_tdata *t = elf_tdata (abfd);
-         int buf[2];
-         int demangler_sym_idx = -1;
-
-         symcount = bfd_get_symcount (abfd);
-         syms = bfd_get_outsymbols (abfd);
-         for (idx = 0; idx < symcount; idx++)
-           {
-             asymbol *sym;
-             sym = syms[idx];
-             if ((sym->flags & (BSF_DEBUGGING | BSF_DYNAMIC))
-                 && strchr (sym->name, '@')
-                 && (strcmp (sym->section->name, BFD_ABS_SECTION_NAME) == 0))
-               {
-                 demangler_sym_idx = sym->udata.i;
-                 break;
-               }
-           }
-
-         hdr->sh_type = SHT_IA_64_VMS_DISPLAY_NAME_INFO;
-         hdr->sh_entsize = 4;
-         hdr->sh_addralign = 0;
-         hdr->sh_link = t->symtab_section;
-
-         /* Find symtab index of demangler routine and stuff it in
-            the second long word of section data.  */
-
-         if (demangler_sym_idx > -1)
-           {
-             bfd_seek (abfd, hdr->sh_offset, SEEK_SET);
-             bfd_bread (buf, hdr->sh_size, abfd);
-             buf [1] = demangler_sym_idx;
-             bfd_seek (abfd, hdr->sh_offset, SEEK_SET);
-             bfd_bwrite (buf, hdr->sh_size, abfd);
-           }
-       }
-    }
-
-  return TRUE;
-}
-
-/* The final processing done just before writing out a VMS IA-64 ELF
-   object file.  */
-
-static void
-elfNN_vms_final_write_processing (bfd *abfd,
-                                 bfd_boolean linker ATTRIBUTE_UNUSED)
-{
-  Elf_Internal_Shdr *hdr;
-  asection *s;
-  int unwind_info_sect_idx = 0;
-
-  for (s = abfd->sections; s; s = s->next)
-    {
-      hdr = &elf_section_data (s)->this_hdr;
-
-      if (strcmp (bfd_get_section_name (abfd, hdr->bfd_section),
-                 ".IA_64.unwind_info") == 0)
-       unwind_info_sect_idx = elf_section_data (s)->this_idx;
-
-      switch (hdr->sh_type)
-       {
-       case SHT_IA_64_UNWIND:
-         /* VMS requires sh_info to point to the unwind info section.  */
-          hdr->sh_info = unwind_info_sect_idx;
-         break;
-       }
-    }
-
-  if (! elf_flags_init (abfd))
-    {
-      unsigned long flags = 0;
-
-      if (abfd->xvec->byteorder == BFD_ENDIAN_BIG)
-       flags |= EF_IA_64_BE;
-      if (bfd_get_mach (abfd) == bfd_mach_ia64_elf64)
-       flags |= EF_IA_64_ABI64;
-
-      elf_elfheader(abfd)->e_flags = flags;
-      elf_flags_init (abfd) = TRUE;
-    }
-}
-
-static bfd_boolean
-elfNN_vms_close_and_cleanup (bfd *abfd)
-{
-  if (bfd_get_format (abfd) == bfd_object)
-    {
-      long isize, irsize;
-
-      if (elf_shstrtab (abfd) != NULL)
-       _bfd_elf_strtab_free (elf_shstrtab (abfd));
-
-      /* Pad to 8 byte boundary for IPF/VMS.  */
-      isize = bfd_get_size (abfd);
-      if ((irsize = isize/8*8) < isize)
-       {
-         int ishort = (irsize + 8) - isize;
-         bfd_seek (abfd, isize, SEEK_SET);
-         bfd_bwrite (bfd_zmalloc (ishort), ishort, abfd);
-       }
-    }
-
-  return _bfd_generic_close_and_cleanup (abfd);
-}
-#endif /* INCLUDE_IA64_VMS */
 \f
 #define TARGET_LITTLE_SYM              bfd_elfNN_ia64_little_vec
 #define TARGET_LITTLE_NAME             "elfNN-ia64-little"
@@ -5381,86 +5102,3 @@ elfNN_vms_close_and_cleanup (bfd *abfd)
 #define elfNN_bed elfNN_ia64_hpux_bed
 
 #include "elfNN-target.h"
-
-/* VMS-specific vectors.  */
-#ifdef INCLUDE_IA64_VMS
-
-#undef  TARGET_LITTLE_SYM
-#define TARGET_LITTLE_SYM              bfd_elfNN_ia64_vms_vec
-#undef  TARGET_LITTLE_NAME
-#define TARGET_LITTLE_NAME             "elfNN-ia64-vms"
-#undef  TARGET_BIG_SYM
-#undef  TARGET_BIG_NAME
-
-/* These are VMS specific functions.  */
-
-#undef  elf_backend_object_p
-#define elf_backend_object_p elfNN_vms_object_p
-
-#undef  elf_backend_section_from_shdr
-#define elf_backend_section_from_shdr elfNN_vms_section_from_shdr
-
-#undef  elf_backend_post_process_headers
-#define elf_backend_post_process_headers elfNN_vms_post_process_headers
-
-#undef  elf_backend_section_processing
-#define elf_backend_section_processing elfNN_vms_section_processing
-
-#undef  elf_backend_final_write_processing
-#define elf_backend_final_write_processing elfNN_vms_final_write_processing
-
-#undef  bfd_elfNN_close_and_cleanup
-#define bfd_elfNN_close_and_cleanup elfNN_vms_close_and_cleanup
-
-#undef  elf_backend_section_from_bfd_section
-
-#undef  elf_backend_symbol_processing
-
-#undef  elf_backend_want_p_paddr_set_to_zero
-
-#undef ELF_OSABI
-#define ELF_OSABI                      ELFOSABI_OPENVMS
-
-#undef  ELF_MAXPAGESIZE
-#define ELF_MAXPAGESIZE                        0x10000 /* 64KB */
-
-#undef  elfNN_bed
-#define elfNN_bed elfNN_ia64_vms_bed
-
-/* Use VMS-style archives (in particular, don't use the standard coff
-   archive format).  */
-#define bfd_elfNN_archive_functions
-
-#undef bfd_elfNN_archive_p
-#define bfd_elfNN_archive_p _bfd_vms_lib_ia64_archive_p
-#undef bfd_elfNN_write_archive_contents
-#define bfd_elfNN_write_archive_contents _bfd_vms_lib_write_archive_contents
-#undef bfd_elfNN_mkarchive
-#define bfd_elfNN_mkarchive _bfd_vms_lib_ia64_mkarchive
-
-#define bfd_elfNN_archive_slurp_armap \
-  _bfd_vms_lib_slurp_armap
-#define bfd_elfNN_archive_slurp_extended_name_table \
-  _bfd_vms_lib_slurp_extended_name_table
-#define bfd_elfNN_archive_construct_extended_name_table \
-  _bfd_vms_lib_construct_extended_name_table
-#define bfd_elfNN_archive_truncate_arname \
-  _bfd_vms_lib_truncate_arname
-#define bfd_elfNN_archive_write_armap \
-  _bfd_vms_lib_write_armap
-#define bfd_elfNN_archive_read_ar_hdr \
-  _bfd_vms_lib_read_ar_hdr
-#define bfd_elfNN_archive_write_ar_hdr \
-  _bfd_vms_lib_write_ar_hdr
-#define bfd_elfNN_archive_openr_next_archived_file \
-  _bfd_vms_lib_openr_next_archived_file
-#define bfd_elfNN_archive_get_elt_at_index \
-  _bfd_vms_lib_get_elt_at_index
-#define bfd_elfNN_archive_generic_stat_arch_elt \
-  _bfd_vms_lib_generic_stat_arch_elt
-#define bfd_elfNN_archive_update_armap_timestamp \
-  _bfd_vms_lib_update_armap_timestamp
-
-#include "elfNN-target.h"
-
-#endif /* INCLUDE_IA64_VMS */
index e3fef4fc775458098a67c05fd77b8067d0de20e3..2fee66b16e7cdc91edaf2c02e65b10727e0aa27d 100755 (executable)
--- a/configure
+++ b/configure
@@ -3093,7 +3093,9 @@ case "${target}" in
     noconfigdirs="$noconfigdirs target-libssp"
     ;;
   rl78-*-*)
-    # Dereferencing -1 is a compile-time error
+    # libssp uses a misaligned load to trigger a fault, but the RL78
+    # doesn't fault for those - instead, it gives a build-time error
+    # for explicit misaligned loads.
     noconfigdirs="$noconfigdirs target-libssp"
     ;;
 esac
@@ -3496,8 +3498,7 @@ case "${target}" in
     noconfigdirs="$noconfigdirs libgui itcl ld"
     ;;
   ia64*-*-*vms*)
-    # No ld support yet.
-    noconfigdirs="$noconfigdirs libgui itcl ld"
+    noconfigdirs="$noconfigdirs libgui itcl"
     ;;
   i[3456789]86-w64-mingw*)
     ;;
index 386ec78b668c548f880c74cfaae2fa180e405c12..c2795874c2b29d39e1285bcaa1c5b64787c2fb9c 100644 (file)
@@ -1,5 +1,5 @@
 #   Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
-#   2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011
+#   2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012
 #   Free Software Foundation, Inc.
 #
 # This file is free software; you can redistribute it and/or modify it
@@ -519,7 +519,9 @@ case "${target}" in
     noconfigdirs="$noconfigdirs target-libssp"
     ;;
   rl78-*-*)
-    # Dereferencing -1 is a compile-time error
+    # libssp uses a misaligned load to trigger a fault, but the RL78
+    # doesn't fault for those - instead, it gives a build-time error
+    # for explicit misaligned loads.
     noconfigdirs="$noconfigdirs target-libssp"
     ;;
 esac
@@ -922,8 +924,7 @@ case "${target}" in
     noconfigdirs="$noconfigdirs libgui itcl ld"
     ;;
   ia64*-*-*vms*)
-    # No ld support yet.
-    noconfigdirs="$noconfigdirs libgui itcl ld"
+    noconfigdirs="$noconfigdirs libgui itcl"
     ;;
   i[[3456789]]86-w64-mingw*)
     ;;
index ebad05a7a777b7455b66749468216842494a1273..b97aa868a3c6b6f018cf7cadf460c98041dad59d 100644 (file)
@@ -1,3 +1,14 @@
+2012-04-11  Tristan Gingold  <gingold@adacore.com>
+
+       * scripttempl/ia64vms.sc: New file.
+       * emultempl/vms.em (_before_parse): Support for ia64.
+       (elf64-ia64-vms): New fragment for ia64.
+       * emulparams/elf64_ia64_vms.sh: New file.
+       * configure.tgt (ia64-*-*vms*): Add.
+       * Makefile.am (ALL_64_EMULATION_SOURCES): Add eelf64_ia64_vms.c
+       (eelf64_ia64_vms.c): New rule.
+       * Makefile.in: Regenerate.
+
 2012-04-06  Roland McGrath  <mcgrathr@google.com>
 
        * configure.in (AC_CHECK_HEADERS): Add locale.h.
index 524350ea34c36eb395ad6af6d048c9cf6ce1d027..df7ab98d22ee105e99a67c83377126b8c9130831 100644 (file)
@@ -467,6 +467,7 @@ ALL_64_EMULATION_SOURCES = \
        eelf64_aix.c \
        eelf64_ia64.c \
        eelf64_ia64_fbsd.c \
+       eelf64_ia64_vms.c \
        eelf64_s390.c \
        eelf64_sparc.c \
        eelf64_sparc_fbsd.c \
@@ -1928,6 +1929,10 @@ eelf64_ia64_fbsd.c: $(srcdir)/emulparams/elf64_ia64_fbsd.sh \
   $(srcdir)/emultempl/needrelax.em \
   $(srcdir)/scripttempl/elf.sc ${GEN_DEPENDS}
        ${GENSCRIPTS} elf64_ia64_fbsd "$(tdir_elf64_ia64_fbsd)"
+eelf64_ia64_vms.c: $(srcdir)/emulparams/elf64_ia64_vms.sh \
+  $(srcdir)/emultempl/vms.em \
+  $(srcdir)/scripttempl/ia64vms.sc ${GEN_DEPENDS}
+       ${GENSCRIPTS} elf64_ia64_vms "$(tdir_elf64_ia64_vms)"
 eelf64_s390.c: $(srcdir)/emulparams/elf64_s390.sh \
   $(ELF_DEPS) $(srcdir)/scripttempl/elf.sc ${GEN_DEPENDS}
        ${GENSCRIPTS} elf64_s390 "$(tdir_elf64_s390)"
index 776a4e0380b2388128a2bce1becbc73fe1c00001..af27d3f8f3368e3f2d9c80ff6c0cf9ad79ab00c6 100644 (file)
@@ -55,6 +55,7 @@ am__aclocal_m4_deps = $(top_srcdir)/../bfd/acinclude.m4 \
        $(top_srcdir)/../config/depstand.m4 \
        $(top_srcdir)/../config/gettext-sister.m4 \
        $(top_srcdir)/../config/largefile.m4 \
+       $(top_srcdir)/../config/lcmessage.m4 \
        $(top_srcdir)/../config/lead-dot.m4 \
        $(top_srcdir)/../config/nls.m4 \
        $(top_srcdir)/../config/override.m4 \
@@ -772,6 +773,7 @@ ALL_64_EMULATION_SOURCES = \
        eelf64_aix.c \
        eelf64_ia64.c \
        eelf64_ia64_fbsd.c \
+       eelf64_ia64_vms.c \
        eelf64_s390.c \
        eelf64_sparc.c \
        eelf64_sparc_fbsd.c \
@@ -1176,6 +1178,7 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/eelf64_aix.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/eelf64_ia64.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/eelf64_ia64_fbsd.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/eelf64_ia64_vms.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/eelf64_s390.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/eelf64_sparc.Po@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/eelf64_sparc_fbsd.Po@am__quote@
@@ -3389,6 +3392,10 @@ eelf64_ia64_fbsd.c: $(srcdir)/emulparams/elf64_ia64_fbsd.sh \
   $(srcdir)/emultempl/needrelax.em \
   $(srcdir)/scripttempl/elf.sc ${GEN_DEPENDS}
        ${GENSCRIPTS} elf64_ia64_fbsd "$(tdir_elf64_ia64_fbsd)"
+eelf64_ia64_vms.c: $(srcdir)/emulparams/elf64_ia64_vms.sh \
+  $(srcdir)/emultempl/vms.em \
+  $(srcdir)/scripttempl/ia64vms.sc ${GEN_DEPENDS}
+       ${GENSCRIPTS} elf64_ia64_vms "$(tdir_elf64_ia64_vms)"
 eelf64_s390.c: $(srcdir)/emulparams/elf64_s390.sh \
   $(ELF_DEPS) $(srcdir)/scripttempl/elf.sc ${GEN_DEPENDS}
        ${GENSCRIPTS} elf64_s390 "$(tdir_elf64_s390)"
index 6418768c9591812eb7e994c1c2ca631ce1514942..abb47065fe50267a3517ee9860a7dba8004fefbb 100644 (file)
@@ -308,6 +308,7 @@ ia64-*-freebsd* | ia64-*-kfreebsd*-gnu)
                        targ_extra_emuls="elf64_ia64" ;;
 ia64-*-netbsd*)                targ_emul=elf64_ia64 ;;
 ia64-*-linux*)         targ_emul=elf64_ia64 ;;
+ia64-*-*vms*)          targ_emul=elf64_ia64_vms ;;
 ia64-*-aix*)           targ_emul=elf64_aix
                        ;;
 ip2k-*-elf)            targ_emul=elf32ip2k
diff --git a/ld/emulparams/elf64_ia64_vms.sh b/ld/emulparams/elf64_ia64_vms.sh
new file mode 100644 (file)
index 0000000..0688402
--- /dev/null
@@ -0,0 +1,7 @@
+SCRIPT_NAME=ia64vms
+
+OUTPUT_FORMAT="elf64-ia64-vms"
+ARCH=ia64
+
+COMPILE_IN=yes
+EXTRA_EM_FILE=vms
index 164f2e19038506f447ec3bd6733faa9f7fa558f3..98837e34a649a002ea493b63ccb567f7681d66a9 100644 (file)
@@ -31,6 +31,10 @@ gld${EMULATION_NAME}_before_parse (void)
   ldfile_set_output_arch ("${ARCH}", bfd_arch_`echo ${ARCH} | sed -e 's/:.*//'`);
   input_flags.dynamic = TRUE;
   config.has_shared = FALSE; /* Not yet.  */
+
+  /* For ia64, harmless for alpha.  */
+  link_info.emit_hash = FALSE;
+  link_info.spare_dynamic_tags = 0;
 }
 
 /* This is called before the input files are opened.  We add the
@@ -132,7 +136,8 @@ gld${EMULATION_NAME}_add_options
    int nrl ATTRIBUTE_UNUSED,
    struct option **really_longopts ATTRIBUTE_UNUSED)
 {
-  static const struct option xtra_long[] = {
+  static const struct option xtra_long[] =
+  {
     {"identification", required_argument, NULL, OPTION_IDENTIFICATION},
     {NULL, no_argument, NULL, 0}
   };
@@ -166,6 +171,63 @@ gld${EMULATION_NAME}_handle_option (int optc)
 
 EOF
 
+if test "$OUTPUT_FORMAT" = "elf64-ia64-vms"; then
+
+fragment <<EOF
+#include "elf-bfd.h"
+EOF
+
+source_em ${srcdir}/emultempl/elf-generic.em
+
+fragment <<EOF
+
+/* This is called after the sections have been attached to output
+   sections, but before any sizes or addresses have been set.  */
+
+static void
+gld${EMULATION_NAME}_before_allocation (void)
+{
+  const struct elf_backend_data *bed;
+
+  if (!is_elf_hash_table (link_info.hash))
+    return;
+
+  bed = get_elf_backend_data (link_info.output_bfd);
+
+  /* The backend must work out the sizes of all the other dynamic
+     sections.  */
+  if (elf_hash_table (&link_info)->dynamic_sections_created
+      && bed->elf_backend_size_dynamic_sections
+      && ! (*bed->elf_backend_size_dynamic_sections) (link_info.output_bfd,
+                                                      &link_info))
+    einfo ("%P%F: failed to set dynamic section sizes: %E\n");
+
+  before_allocation_default ();
+}
+
+static void
+gld${EMULATION_NAME}_after_allocation (void)
+{
+  bfd_boolean need_layout = bfd_elf_discard_info (link_info.output_bfd,
+                                                 &link_info);
+  gld${EMULATION_NAME}_map_segments (need_layout);
+}
+
+static void
+gld${EMULATION_NAME}_after_parse (void)
+{
+  link_info.relax_pass = 2;
+  after_parse_default ();
+}
+EOF
+
+LDEMUL_BEFORE_ALLOCATION=gld"$EMULATION_NAME"_before_allocation
+LDEMUL_AFTER_ALLOCATION=gld"$EMULATION_NAME"_after_allocation
+
+LDEMUL_AFTER_PARSE=gld${EMULATION_NAME}_after_parse
+source_em ${srcdir}/emultempl/needrelax.em
+fi
+
 LDEMUL_PLACE_ORPHAN=vms_place_orphan
 LDEMUL_BEFORE_PARSE=gld"$EMULATION_NAME"_before_parse
 LDEMUL_CREATE_OUTPUT_SECTION_STATEMENTS=gld"$EMULATION_NAME"_create_output_section_statements
diff --git a/ld/scripttempl/ia64vms.sc b/ld/scripttempl/ia64vms.sc
new file mode 100644 (file)
index 0000000..b0b1c42
--- /dev/null
@@ -0,0 +1,129 @@
+# Linker script for Itanium VMS systems.
+# Tristan Gingold <gingold@adacore.com>.
+
+PAGESIZE=0x10000
+BLOCKSIZE=0x200
+
+cat <<EOF
+OUTPUT_FORMAT("${OUTPUT_FORMAT}")
+${LIB_SEARCH_DIRS}
+ENTRY(__entry)
+
+SECTIONS
+{
+  /* RW segment.  */
+  ${RELOCATING+. = ${PAGESIZE};}
+
+  \$DATA\$ ALIGN (${BLOCKSIZE}) : {
+    *(\$DATA\$ .data .data.*)
+    *(\$BSS\$ .bss)
+  }
+
+  /* Code segment.  Note: name must be \$CODE\$ */
+  ${RELOCATING+. = ALIGN (${PAGESIZE});}
+
+  \$CODE\$ ALIGN (${BLOCKSIZE}) : {
+    *(\$CODE\$ .text)
+  }
+  .plt ALIGN (8) : {
+    *(.plt)
+  }
+
+  /* RO segment.  */
+  ${RELOCATING+. = ALIGN (${PAGESIZE});}
+
+  /* RO initialized data.  */
+  \$LITERAL\$ ALIGN (${BLOCKSIZE}) : {
+    *(\$LITERAL\$)
+    *(\$READONLY\$ .rodata)
+    *(.jcr)
+    *(.ctors)
+    *(.dtors)
+    *(.opd)
+    *(.gcc_except_table)
+
+    /* LIB$INITIALIZE stuff.  */
+    *(LIB\$INITIALIZDZ)        /* Start marker.  */
+    *(LIB\$INITIALIZD_)        /* Hi priority.  */
+    *(LIB\$INITIALIZE) /* User.  */
+    *(LIB\$INITIALIZE$)        /* End marker.  */
+  }
+
+  /* Short segment.  */
+  ${RELOCATING+. = ALIGN (${PAGESIZE});}
+
+  .srodata : {
+    *(.srodata)
+  }
+  .got ALIGN (8) : {
+    *(.got)
+  }
+  .IA_64.pltoff ALIGN (16) : {
+    *(.IA_64.pltoff)
+  }
+  \$TFR\$ ALIGN (16) : {
+    /* Tranfer vector.  */
+    __entry = .;
+    *(.transfer)
+  }
+
+  ${RELOCATING+. = ALIGN (${PAGESIZE});}
+
+  \$RW_SHORT\$ ALIGN (${BLOCKSIZE}) : {
+    *(.sdata .sdata.*)
+    *(.sbss)
+  }
+
+  ${RELOCATING+. = ALIGN (${PAGESIZE});}
+
+  .IA_64.unwind ALIGN (${BLOCKSIZE}) : {
+    *(.IA_64.unwind .IA_64.unwind.*)
+  }
+
+  .IA_64.unwind_info ALIGN (8) : {
+    *(.IA_64.unwind_info .IA_64.unwind_info.*)
+  }
+
+  ${RELOCATING+. = ALIGN (${PAGESIZE});}
+
+  .dynamic /* \$DYNAMIC\$ */ ALIGN (${BLOCKSIZE}) : {
+    *(.dynamic)
+    *(.vmsdynstr)
+    *(.fixups)
+  }
+
+  ${RELOCATING+. = ALIGN (${PAGESIZE});}
+
+  .dynstr : { *(.dynstr) }
+
+  .dynsym       ${RELOCATING-0} : { *(.dynsym) }
+  .rela.got : { *(.rela.got) }
+  .got.plt : { *(.got.plt) }
+  .gnu.version_d : { *(.gnu.version_d) }
+  .gnu.version : { *(.gnu.version) }
+  .gnu.version_r : { *(.gnu.version_r) }
+  .rela.IA_64.pltoff : { *(.rela.IA_64.pltoff) }
+
+  /* DWARF 2 */
+  .debug_info     0 : { *(.debug_info${RELOCATING+ .gnu.linkonce.wi.*}) }
+  .debug_abbrev   0 : { *(.debug_abbrev) }
+  .debug_line     0 : { *(.debug_line) }
+  .debug_frame    0 : { *(.debug_frame) }
+  .debug_str      0 : { *(.debug_str) }
+  .debug_loc      0 : { *(.debug_loc) }
+  .debug_macinfo  0 : { *(.debug_macinfo) }
+  .debug_aranges  0 : { *(.debug_aranges) }
+  .debug_pubnames 0 : { *(.debug_pubnames) }
+  .trace_info     0 : { *(.trace_info) }
+  .trace_abbrev   0 : { *(.trace_abbrev) }
+  .trace_aranges  0 : { *(.trace_aranges) }
+
+  /* DWARF 3 */
+  .debug_pubtypes 0 : { *(.debug_pubtypes) }
+  .debug_ranges   0 : { *(.debug_ranges) }
+
+  .note : { *(.vms.note) }
+
+  /DISCARD/ : { *(.note) }
+}
+EOF
This page took 0.10591 seconds and 4 git commands to generate.