* elf64-hppa.c, elf64-hppa.h: New files with PA64 support.

1 files changed, 2633 insertions, 0 deletions

diff --git a/bfd/elf64-hppa.c b/bfd/elf64-hppa.c
new file mode 100644
index 0000000..b5b9f7f
--- /dev/null
+++ b/bfd/elf64-hppa.c
@@ -0,0 +1,2633 @@
+/* Generic support for 64-bit ELF
+   Copyright 1999, 2000 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 2 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., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.  */
+
+#include "bfd.h"
+#include "sysdep.h"
+#include "libbfd.h"
+#include "elf-bfd.h"
+#include "elf/hppa.h"
+#include "libhppa.h"
+#include "elf64-hppa.h"
+#define ARCH_SIZE	       64
+
+#define PLT_ENTRY_SIZE 0x10
+#define DLT_ENTRY_SIZE 0x8
+#define OPD_ENTRY_SIZE 0x20
+ 
+#define ELF_DYNAMIC_INTERPRETER "/usr/lib/pa20_64/dld.sl"
+
+/* The stub is supposed to load the target address and target's DP
+   value out of the PLT, then do an external branch to the target
+   address.
+
+   LDD PLTOFF(%r27),%r1
+   BVE (%r1)
+   LDD PLTOFF+8(%r27),%r27
+
+   Note that we must use the LDD with a 14 bit displacement, not the one
+   with a 5 bit displacement.  */
+static char plt_stub[] = {0x53, 0x61, 0x00, 0x00, 0xe8, 0x20, 0xd0, 0x00,
+			  0x53, 0x7b, 0x00, 0x00 };
+
+struct elf64_hppa_dyn_hash_entry
+{
+  struct bfd_hash_entry root;
+
+  /* Offsets for this symbol in various linker sections.  */
+  bfd_vma dlt_offset;
+  bfd_vma plt_offset;
+  bfd_vma opd_offset;
+  bfd_vma stub_offset;
+
+  /* The symbol table entry, if any, that this was derrived from.  */
+  struct elf_link_hash_entry *h;
+
+  /* The index of the (possibly local) symbol in the input bfd and its
+     associated BFD.  Needed so that we can have relocs against local
+     symbols in shared libraries.  */
+  unsigned long sym_indx;
+  bfd *owner;
+
+  /* Dynamic symbols may need to have two different values.  One for
+     the dynamic symbol table, one for the normal symbol table.
+
+     In such cases we store the symbol's real value and section
+     index here so we can restore the real value before we write
+     the normal symbol table.  */
+  bfd_vma st_value;
+  int st_shndx;
+
+  /* Used to count non-got, non-plt relocations for delayed sizing
+     of relocation sections.  */
+  struct elf64_hppa_dyn_reloc_entry
+  {
+    /* Next relocation in the chain.  */
+    struct elf64_hppa_dyn_reloc_entry *next;
+
+    /* The type of the relocation.  */
+    int type;
+
+    /* The input section of the relocation.  */
+    asection *sec;
+
+    /* The index of the section symbol for the input section of
+       the relocation.  Only needed when building shared libraries.  */
+    int sec_symndx;
+
+    /* The offset within the input section of the relocation.  */
+    bfd_vma offset;
+
+    /* The addend for the relocation.  */
+    bfd_vma addend;
+
+  } *reloc_entries;
+
+  /* Nonzero if this symbol needs an entry in one of the linker
+     sections.  */
+  unsigned want_dlt;
+  unsigned want_plt;
+  unsigned want_opd;
+  unsigned want_stub;
+};
+
+struct elf64_hppa_dyn_hash_table
+{
+  struct bfd_hash_table root;
+};
+
+struct elf64_hppa_link_hash_table
+{
+  struct elf_link_hash_table root;
+
+  /* Shortcuts to get to the various linker defined sections.  */
+  asection *dlt_sec;
+  asection *dlt_rel_sec;
+  asection *plt_sec;
+  asection *plt_rel_sec;
+  asection *opd_sec;
+  asection *opd_rel_sec;
+  asection *other_rel_sec;
+
+  /* Offset of __gp within .plt section.  When the PLT gets large we want
+     to slide __gp into the PLT section so that we can continue to use
+     single DP relative instructions to load values out of the PLT.  */
+  bfd_vma gp_offset;
+
+  /* Note this is not strictly correct.  We should create a stub section for
+     each input section with calls.  The stub section should be placed before
+     the section with the call.  */
+  asection *stub_sec;
+
+  bfd_vma text_segment_base;
+  bfd_vma data_segment_base;
+
+  struct elf64_hppa_dyn_hash_table dyn_hash_table;
+
+  /* We build tables to map from an input section back to its
+     symbol index.  This is the BFD for which we currently have
+     a map.  */
+  bfd *section_syms_bfd;
+
+  /* Array of symbol numbers for each input section attached to the
+     current BFD.  */
+  int *section_syms;
+};
+
+#define elf64_hppa_hash_table(p) \
+  ((struct elf64_hppa_link_hash_table *) ((p)->hash))
+
+typedef struct bfd_hash_entry *(*new_hash_entry_func)
+  PARAMS ((struct bfd_hash_entry *, struct bfd_hash_table *, const char *));
+
+static boolean elf64_hppa_dyn_hash_table_init
+  PARAMS ((struct elf64_hppa_dyn_hash_table *ht, bfd *abfd,
+	   new_hash_entry_func new));
+static struct bfd_hash_entry *elf64_hppa_new_dyn_hash_entry
+  PARAMS ((struct bfd_hash_entry *entry, struct bfd_hash_table *table,
+	   const char *string));
+static struct bfd_link_hash_table *elf64_hppa_hash_table_create
+  PARAMS ((bfd *abfd));
+static struct elf64_hppa_dyn_hash_entry *elf64_hppa_dyn_hash_lookup
+  PARAMS ((struct elf64_hppa_dyn_hash_table *table, const char *string,
+	   boolean create, boolean copy));
+static void elf64_hppa_dyn_hash_traverse
+  PARAMS ((struct elf64_hppa_dyn_hash_table *table,
+	   boolean (*func)(struct elf64_hppa_dyn_hash_entry *, PTR),
+	   PTR info));
+
+static const char *get_dyn_name
+  PARAMS ((bfd *abfd, struct elf_link_hash_entry *h,
+	   const Elf_Internal_Rela *rel, char **pbuf, size_t *plen));
+
+
+/* This must follow the definitions of the various derived linker
+   hash tables and shared functions.  */
+#include "elf-hppa.h"
+
+
+static boolean elf64_hppa_object_p
+  PARAMS ((bfd *));
+
+static boolean elf64_hppa_section_from_shdr
+  PARAMS ((bfd *, Elf64_Internal_Shdr *, char *));
+
+static void elf64_hppa_post_process_headers
+  PARAMS ((bfd *, struct bfd_link_info *));
+
+static boolean elf64_hppa_create_dynamic_sections
+  PARAMS ((bfd *, struct bfd_link_info *));
+
+static boolean elf64_hppa_adjust_dynamic_symbol
+  PARAMS ((struct bfd_link_info *, struct elf_link_hash_entry *));
+
+static boolean elf64_hppa_size_dynamic_sections
+  PARAMS ((bfd *, struct bfd_link_info *));
+
+static boolean elf64_hppa_finish_dynamic_symbol
+  PARAMS ((bfd *, struct bfd_link_info *,
+	   struct elf_link_hash_entry *, Elf_Internal_Sym *));
+	
+static boolean elf64_hppa_finish_dynamic_sections
+  PARAMS ((bfd *, struct bfd_link_info *));
+
+static boolean elf64_hppa_check_relocs
+  PARAMS ((bfd *, struct bfd_link_info *,
+	   asection *, const Elf_Internal_Rela *));
+
+static boolean elf64_hppa_dynamic_symbol_p
+  PARAMS ((struct elf_link_hash_entry *, struct bfd_link_info *));
+
+static boolean elf64_hppa_mark_exported_functions
+  PARAMS ((struct elf_link_hash_entry *, PTR));
+
+static boolean elf64_hppa_finalize_opd
+  PARAMS ((struct elf64_hppa_dyn_hash_entry *, PTR));
+
+static boolean elf64_hppa_finalize_dlt
+  PARAMS ((struct elf64_hppa_dyn_hash_entry *, PTR));
+
+static boolean allocate_global_data_dlt
+  PARAMS ((struct elf64_hppa_dyn_hash_entry *, PTR));
+
+static boolean allocate_global_data_plt
+  PARAMS ((struct elf64_hppa_dyn_hash_entry *, PTR));
+
+static boolean allocate_global_data_stub
+  PARAMS ((struct elf64_hppa_dyn_hash_entry *, PTR));
+
+static boolean allocate_global_data_opd
+  PARAMS ((struct elf64_hppa_dyn_hash_entry *, PTR));
+
+static boolean get_reloc_section
+  PARAMS ((bfd *, struct elf64_hppa_link_hash_table *, asection *));
+
+static boolean count_dyn_reloc
+  PARAMS ((bfd *, struct elf64_hppa_dyn_hash_entry *,
+	   int, asection *, int, bfd_vma, bfd_vma));
+
+static boolean allocate_dynrel_entries
+  PARAMS ((struct elf64_hppa_dyn_hash_entry *, PTR));
+
+static boolean elf64_hppa_finalize_dynreloc
+  PARAMS ((struct elf64_hppa_dyn_hash_entry *, PTR));
+
+static boolean get_opd
+  PARAMS ((bfd *, struct bfd_link_info *, struct elf64_hppa_link_hash_table *));
+
+static boolean get_plt
+  PARAMS ((bfd *, struct bfd_link_info *, struct elf64_hppa_link_hash_table *));
+
+static boolean get_dlt
+  PARAMS ((bfd *, struct bfd_link_info *, struct elf64_hppa_link_hash_table *));
+
+static boolean get_stub
+  PARAMS ((bfd *, struct bfd_link_info *, struct elf64_hppa_link_hash_table *));
+
+static boolean
+elf64_hppa_dyn_hash_table_init (ht, abfd, new)
+     struct elf64_hppa_dyn_hash_table *ht;
+     bfd *abfd;
+     new_hash_entry_func new;
+{
+  memset (ht, 0, sizeof(*ht));
+  return bfd_hash_table_init (&ht->root, new);
+}
+
+static struct bfd_hash_entry*
+elf64_hppa_new_dyn_hash_entry (entry, table, string)
+     struct bfd_hash_entry *entry;
+     struct bfd_hash_table *table;
+     const char *string;
+{
+  struct elf64_hppa_dyn_hash_entry *ret;
+  ret = (struct elf64_hppa_dyn_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;
+
+  /* Initialize our local data.  All zeros, and definitely easier
+     than setting 8 bit fields.  */
+  memset (ret, 0, sizeof(*ret));
+
+  /* Call the allocation method of the superclass.  */
+  ret = ((struct elf64_hppa_dyn_hash_entry *)
+	 bfd_hash_newfunc ((struct bfd_hash_entry *) ret, table, string));
+
+  return &ret->root;
+}
+
+/* Create the derived linker hash table.  The PA64 ELF port uses this
+   derived hash table to keep information specific to the PA ElF
+   linker (without using static variables).  */
+
+static struct bfd_link_hash_table*
+elf64_hppa_hash_table_create (abfd)
+     bfd *abfd;
+{
+  struct elf64_hppa_link_hash_table *ret;
+
+  ret = bfd_zalloc (abfd, sizeof (*ret));
+  if (!ret)
+    return 0;
+  if (!_bfd_elf_link_hash_table_init (&ret->root, abfd,
+				      _bfd_elf_link_hash_newfunc))
+    {
+      bfd_release (abfd, ret);
+      return 0;
+    }
+
+  if (!elf64_hppa_dyn_hash_table_init (&ret->dyn_hash_table, abfd,
+				       elf64_hppa_new_dyn_hash_entry))
+    return 0;
+  return &ret->root.root;
+}
+
+/* Look up an entry in a PA64 ELF linker hash table.  */
+
+static struct elf64_hppa_dyn_hash_entry *
+elf64_hppa_dyn_hash_lookup(table, string, create, copy)
+     struct elf64_hppa_dyn_hash_table *table;
+     const char *string;
+     boolean create, copy;
+{
+  return ((struct elf64_hppa_dyn_hash_entry *)
+	  bfd_hash_lookup (&table->root, string, create, copy));
+}
+
+/* Traverse a PA64 ELF linker hash table.  */
+
+static void
+elf64_hppa_dyn_hash_traverse (table, func, info)
+     struct elf64_hppa_dyn_hash_table *table;
+     boolean (*func) PARAMS ((struct elf64_hppa_dyn_hash_entry *, PTR));
+     PTR info;
+{
+  (bfd_hash_traverse
+   (&table->root,
+    (boolean (*) PARAMS ((struct bfd_hash_entry *, PTR))) func,
+    info));
+}
+
+/* Return nonzero if ABFD represents a PA2.0 ELF64 file.
+
+   Additionally we set the default architecture and machine.  */
+static boolean
+elf64_hppa_object_p (abfd)
+     bfd *abfd;
+{
+  /* Set the right machine number for an HPPA ELF file.  */
+  return bfd_default_set_arch_mach (abfd, bfd_arch_hppa, 25);
+}
+
+/* Given section type (hdr->sh_type), return a boolean indicating
+   whether or not the section is an elf64-hppa specific section.  */
+static boolean
+elf64_hppa_section_from_shdr (abfd, hdr, name)
+     bfd *abfd;
+     Elf64_Internal_Shdr *hdr;
+     char *name;
+{
+  asection *newsect;
+
+  switch (hdr->sh_type)
+    {
+    case SHT_PARISC_EXT:
+      if (strcmp (name, ".PARISC.archext") != 0)
+	return false;
+      break;
+    case SHT_PARISC_UNWIND:
+      if (strcmp (name, ".PARISC.unwind") != 0)
+	return false;
+      break;
+    case SHT_PARISC_DOC:
+    case SHT_PARISC_ANNOT:
+    default:
+      return false;
+    }
+
+  if (! _bfd_elf_make_section_from_shdr (abfd, hdr, name))
+    return false;
+  newsect = hdr->bfd_section;
+
+  return true;
+}
+
+
+/* Construct a string for use in the elf64_hppa_dyn_hash_table.  The
+   name describes what was once potentially anonymous memory.  We 
+   allocate memory as necessary, possibly reusing PBUF/PLEN.  */
+
+static const char *
+get_dyn_name (abfd, h, rel, pbuf, plen)
+     bfd *abfd;
+     struct elf_link_hash_entry *h;
+     const Elf_Internal_Rela *rel;
+     char **pbuf;
+     size_t *plen;
+{
+  size_t nlen, tlen;
+  char *buf;
+  size_t len;
+
+  if (h && rel->r_addend == 0)
+    return h->root.root.string;
+
+  if (h)
+    nlen = strlen (h->root.root.string);
+  else
+    {
+      nlen = sizeof(void*)*2 + 1 + sizeof(bfd_vma)*4 + 1 + 1;
+      nlen += 10;	/* %p slop */
+    }
+  tlen = nlen + 1 + 16 + 1;
+
+  len = *plen;
+  buf = *pbuf;
+  if (len < tlen)
+    {
+      if (buf)
+	free (buf);
+      *pbuf = buf = malloc (tlen);
+      *plen = len = tlen;
+      if (!buf)
+	return NULL;
+    }
+
+  if (h)
+    {
+      memcpy (buf, h->root.root.string, nlen);
+      sprintf_vma (buf + nlen, rel->r_addend);
+    }
+  else
+    {
+      nlen = sprintf (buf, "%p:%lx", abfd, ELF64_R_SYM (rel->r_info));
+      if (rel->r_addend)
+	{
+	  buf[nlen++] = '+';
+	  sprintf_vma (buf + nlen, rel->r_addend);
+	}
+    }
+
+  return buf;
+}
+
+/* SEC is a section containing relocs for an input BFD when linking; return
+   a suitable section for holding relocs in the output BFD for a link.  */
+
+static boolean
+get_reloc_section (abfd, hppa_info, sec)
+     bfd *abfd;
+     struct elf64_hppa_link_hash_table *hppa_info;
+     asection *sec;
+{
+  const char *srel_name;
+  asection *srel;
+  bfd *dynobj;
+
+  srel_name = (bfd_elf_string_from_elf_section
+	       (abfd, elf_elfheader(abfd)->e_shstrndx,
+		elf_section_data(sec)->rel_hdr.sh_name));
+  if (srel_name == NULL)
+    return false;
+
+  BFD_ASSERT ((strncmp (srel_name, ".rela", 5) == 0
+	       && strcmp (bfd_get_section_name (abfd, sec),
+			  srel_name+5) == 0)
+	      || (strncmp (srel_name, ".rel", 4) == 0
+		  && strcmp (bfd_get_section_name (abfd, sec),
+			     srel_name+4) == 0));
+
+  dynobj = hppa_info->root.dynobj;
+  if (!dynobj)
+    hppa_info->root.dynobj = dynobj = abfd;
+
+  srel = bfd_get_section_by_name (dynobj, srel_name);
+  if (srel == NULL)
+    {
+      srel = bfd_make_section (dynobj, srel_name);
+      if (srel == NULL
+	  || !bfd_set_section_flags (dynobj, srel,
+				     (SEC_ALLOC
+				      | SEC_LOAD
+				      | SEC_HAS_CONTENTS
+				      | SEC_IN_MEMORY
+				      | SEC_LINKER_CREATED
+				      | SEC_READONLY))
+	  || !bfd_set_section_alignment (dynobj, srel, 3))
+	return false;
+    }
+
+  hppa_info->other_rel_sec = srel;
+  return true;
+}
+
+/* Add a new entry to the list of dynamic relocations against DYN_H. 
+
+   We use this to keep a record of all the FPTR relocations against a
+   particular symbol so that we can create FPTR relocations in the
+   output file.  */
+
+static boolean
+count_dyn_reloc (abfd, dyn_h, type, sec, sec_symndx, offset, addend)
+     bfd *abfd;
+     struct elf64_hppa_dyn_hash_entry *dyn_h;
+     int type;
+     asection *sec;
+     int sec_symndx;
+     bfd_vma offset;
+     bfd_vma addend;
+{
+  struct elf64_hppa_dyn_reloc_entry *rent;
+
+  rent = (struct elf64_hppa_dyn_reloc_entry *)
+  bfd_alloc (abfd, sizeof (*rent));
+  if (!rent)
+    return false;
+
+  rent->next = dyn_h->reloc_entries;
+  rent->type = type;
+  rent->sec = sec;
+  rent->sec_symndx = sec_symndx;
+  rent->offset = offset;
+  rent->addend = addend;
+  dyn_h->reloc_entries = rent;
+
+  return true;
+}
+
+/* Scan the RELOCS and record the type of dynamic entries that each
+   referenced symbol needs.  */
+
+static boolean
+elf64_hppa_check_relocs (abfd, info, sec, relocs)
+     bfd *abfd;
+     struct bfd_link_info *info;
+     asection *sec;
+     const Elf_Internal_Rela *relocs;
+{
+  struct elf64_hppa_link_hash_table *hppa_info;
+  const Elf_Internal_Rela *relend;
+  Elf_Internal_Shdr *symtab_hdr;
+  const Elf_Internal_Rela *rel;
+  asection *dlt, *plt, *stubs;
+  char *buf;
+  size_t buf_len;
+  int sec_symndx;
+
+  if (info->relocateable)
+    return true;
+
+  /* If this is the first dynamic object found in the link, create
+     the special sections required for dynamic linking.  */
+  if (! elf_hash_table (info)->dynamic_sections_created)
+    {
+      if (! bfd_elf64_link_create_dynamic_sections (abfd, info))
+	return false;
+    }
+
+  hppa_info = elf64_hppa_hash_table (info);
+  symtab_hdr = &elf_tdata (abfd)->symtab_hdr;
+
+  /* If necessary, build a new table holding section symbols indices
+     for this BFD.  This is disgusting.  */
+ 
+  if (info->shared && hppa_info->section_syms_bfd != abfd)
+    {
+      int i, highest_shndx;
+      asection *section;
+      Elf_Internal_Sym *local_syms, *isym;
+      Elf64_External_Sym *ext_syms, *esym;
+
+      /* We're done with the old cache of section index to section symbol
+	 index information.  Free it.
+
+	 ?!? Note we leak the last section_syms array.  Presumably we
+	 could free it in one of the later routines in this file.  */
+      if (hppa_info->section_syms)
+	free (hppa_info->section_syms);
+
+      /* Allocate memory for the internal and external symbols.  */
+      local_syms
+        = (Elf_Internal_Sym *) bfd_malloc (symtab_hdr->sh_info
+                                           * sizeof (Elf_Internal_Sym));
+      if (local_syms == NULL)
+	return false;
+
+      ext_syms
+        = (Elf64_External_Sym *) bfd_malloc (symtab_hdr->sh_info
+                                             * sizeof (Elf64_External_Sym));
+      if (ext_syms == NULL)
+	{
+	  free (local_syms);
+	  return false;
+	}
+
+      /* Read in the local symbols.  */
+      if (bfd_seek (abfd, symtab_hdr->sh_offset, SEEK_SET) != 0
+          || bfd_read (ext_syms, 1,
+                       (symtab_hdr->sh_info
+                        * sizeof (Elf64_External_Sym)), abfd)
+          != (symtab_hdr->sh_info * sizeof (Elf64_External_Sym)))
+        {
+	  free (local_syms);
+	  free (ext_syms);
+	  return false;
+        }
+
+      /* Swap in the local symbols, also record the highest section index
+	 referenced by the local symbols.  */
+      isym = local_syms;
+      esym = ext_syms;
+      highest_shndx = 0;
+      for (i = 0; i < symtab_hdr->sh_info; i++, esym++, isym++)
+	{
+	  bfd_elf64_swap_symbol_in (abfd, esym, isym);
+	  if (isym->st_shndx > highest_shndx)
+	    highest_shndx = isym->st_shndx;
+	}
+
+      /* Now we can free the external symbols.  */
+      free (ext_syms);
+
+      /* Allocate an array to hold the section index to section symbol index
+	 mapping.  Bump by one since we start counting at zero.  */
+      highest_shndx++;
+      hppa_info->section_syms = (int *) bfd_malloc (highest_shndx
+						    * sizeof (int));
+
+      /* Now walk the local symbols again.  If we find a section symbol,
+	 record the index of the symbol into the section_syms array.  */
+      for (isym = local_syms, i = 0; i < symtab_hdr->sh_info; i++, isym++)
+	{
+	  if (ELF_ST_TYPE (isym->st_info) == STT_SECTION)
+	    hppa_info->section_syms[isym->st_shndx] = i;
+	}
+
+      /* We are finished with the local symbols.  Get rid of them.  */
+      free (local_syms);
+
+      /* Record which BFD we built the section_syms mapping for.  */
+      hppa_info->section_syms_bfd = abfd;
+    }
+
+  /* Record the symbol index for this input section.  We may need it for
+     relocations when building shared libraries.  When not building shared
+     libraries this value is never really used, but assign it to zero to
+     prevent out of bounds memory accesses in other routines.  */
+  if (info->shared)
+    {
+      sec_symndx = _bfd_elf_section_from_bfd_section (abfd, sec);
+
+      /* If we did not find a section symbol for this section, then
+	 something went terribly wrong above.  */
+      if (sec_symndx == -1)
+	return false;
+
+      sec_symndx = hppa_info->section_syms[sec_symndx];
+    }
+  else
+    sec_symndx = 0;
+ 
+  dlt = plt = stubs = NULL;
+  buf = NULL;
+  buf_len = 0;
+
+  relend = relocs + sec->reloc_count;
+  for (rel = relocs; rel < relend; ++rel)
+    {
+      enum {
+	NEED_DLT = 1,
+	NEED_PLT = 2,
+	NEED_STUB = 4,
+	NEED_OPD = 8,
+	NEED_DYNREL = 16,
+      };
+
+      struct elf_link_hash_entry *h = NULL;
+      unsigned long r_symndx = ELF64_R_SYM (rel->r_info);
+      struct elf64_hppa_dyn_hash_entry *dyn_h;
+      int need_entry;
+      const char *addr_name;
+      boolean maybe_dynamic;
+      int dynrel_type = R_PARISC_NONE;
+      static reloc_howto_type *howto;
+
+      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->elf_link_hash_flags |= ELF_LINK_HASH_REF_REGULAR;
+	}
+
+      /* 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 = false;
+      if (h && ((info->shared && ! info->symbolic)
+		|| ! (h->elf_link_hash_flags & ELF_LINK_HASH_DEF_REGULAR)
+		|| h->root.type == bfd_link_hash_defweak))
+	maybe_dynamic = true;
+
+      howto = elf_hppa_howto_table + ELF64_R_TYPE (rel->r_info);
+      need_entry = 0;
+      switch (howto->type)
+	{
+	/* These are simple indirect references to symbols through the
+	   DLT.  We need to create a DLT entry for any symbols which
+	   appears in a DLTIND relocation.  */
+	case R_PARISC_DLTIND21L:
+	case R_PARISC_DLTIND14R:
+	case R_PARISC_DLTIND14F:
+	case R_PARISC_DLTIND14WR:
+	case R_PARISC_DLTIND14DR:
+	  need_entry = NEED_DLT;
+	  break;
+
+	/* ?!?  These need a DLT entry.  But I have no idea what to do with
+	   the "link time TP value.  */
+	case R_PARISC_LTOFF_TP21L:
+	case R_PARISC_LTOFF_TP14R:
+	case R_PARISC_LTOFF_TP14F:
+	case R_PARISC_LTOFF_TP64:
+	case R_PARISC_LTOFF_TP14WR:
+	case R_PARISC_LTOFF_TP14DR:
+	case R_PARISC_LTOFF_TP16F:
+	case R_PARISC_LTOFF_TP16WF:
+	case R_PARISC_LTOFF_TP16DF:
+	  need_entry = NEED_DLT;
+	  break;
+
+	/* These are function calls.  Depending on their precise target we
+	   may need to make a stub for them.  The stub uses the PLT, so we
+	   need to create PLT entries for these symbols too.  */
+	case R_PARISC_PCREL17F:
+	case R_PARISC_PCREL22F:
+	case R_PARISC_PCREL32:
+	case R_PARISC_PCREL64:
+	case R_PARISC_PCREL21L:
+	case R_PARISC_PCREL17R:
+	case R_PARISC_PCREL17C:
+	case R_PARISC_PCREL14R:
+	case R_PARISC_PCREL14F:
+	case R_PARISC_PCREL22C:
+	case R_PARISC_PCREL14WR:
+	case R_PARISC_PCREL14DR:
+	case R_PARISC_PCREL16F:
+	case R_PARISC_PCREL16WF:
+	case R_PARISC_PCREL16DF:
+	  need_entry = (NEED_PLT | NEED_STUB);
+	  break;
+
+	case R_PARISC_PLTOFF21L:
+	case R_PARISC_PLTOFF14R:
+	case R_PARISC_PLTOFF14F:
+	case R_PARISC_PLTOFF14WR:
+	case R_PARISC_PLTOFF14DR:
+	case R_PARISC_PLTOFF16F:
+	case R_PARISC_PLTOFF16WF:
+	case R_PARISC_PLTOFF16DF:
+	  need_entry = (NEED_PLT);
+	  break;
+
+	case R_PARISC_DIR64:
+	  if (info->shared || maybe_dynamic)
+	    need_entry = (NEED_DYNREL);
+	  dynrel_type = R_PARISC_DIR64;
+	  break;
+
+	/* This is an indirect reference through the DLT to get the address
+	   of a OPD descriptor.  Thus we need to make a DLT entry that points
+	   to an OPD entry.  */
+	case R_PARISC_LTOFF_FPTR21L:
+	case R_PARISC_LTOFF_FPTR14R:
+	case R_PARISC_LTOFF_FPTR14WR:
+	case R_PARISC_LTOFF_FPTR14DR:
+	case R_PARISC_LTOFF_FPTR32:
+	case R_PARISC_LTOFF_FPTR64:
+	case R_PARISC_LTOFF_FPTR16F:
+	case R_PARISC_LTOFF_FPTR16WF:
+	case R_PARISC_LTOFF_FPTR16DF:
+	  if (info->shared || maybe_dynamic)
+	    need_entry = (NEED_DLT | NEED_OPD);
+	  else
+	    need_entry = (NEED_DLT | NEED_OPD);
+	  dynrel_type = R_PARISC_FPTR64;
+	  break;
+
+	/* This is a simple OPD entry.  */
+	case R_PARISC_FPTR64:
+	  if (info->shared || maybe_dynamic)
+	    need_entry = (NEED_OPD | NEED_DYNREL);
+	  else
+	    need_entry = (NEED_OPD);
+	  dynrel_type = R_PARISC_FPTR64;
+	  break;
+
+	/* Add more cases as needed.  */
+	}
+
+      if (!need_entry)
+	continue;
+
+      /* Collect a canonical name for this address.  */
+      addr_name = get_dyn_name (abfd, h, rel, &buf, &buf_len);
+
+      /* Collect the canonical entry data for this address.  */
+      dyn_h = elf64_hppa_dyn_hash_lookup (&hppa_info->dyn_hash_table,
+					  addr_name, true, true);
+      BFD_ASSERT (dyn_h);
+
+      /* Stash away enough information to be able to find this symbol
+	 regardless of whether or not it is local or global.  */
+      dyn_h->h = h;
+      dyn_h->owner = abfd;
+      dyn_h->sym_indx = r_symndx;
+
+      /* ?!? We may need to do some error checking in here.  */
+      /* Create what's needed.  */
+      if (need_entry & NEED_DLT)
+	{
+	  if (! hppa_info->dlt_sec
+	      && ! get_dlt (abfd, info, hppa_info))
+	    goto err_out;
+	  dyn_h->want_dlt = 1;
+	}
+
+      if (need_entry & NEED_PLT)
+	{
+	  if (! hppa_info->plt_sec
+	      && ! get_plt (abfd, info, hppa_info))
+	    goto err_out;
+	  dyn_h->want_plt = 1;
+	}
+
+      if (need_entry & NEED_STUB)
+	{
+	  if (! hppa_info->stub_sec
+	      && ! get_stub (abfd, info, hppa_info))
+	    goto err_out;
+	  dyn_h->want_stub = 1;
+	}
+
+      if (need_entry & NEED_OPD)
+	{
+	  if (! hppa_info->opd_sec
+	      && ! get_opd (abfd, info, hppa_info))
+	    goto err_out;
+
+	  dyn_h->want_opd = 1;
+
+	  /* FPTRs are not allocated by the dynamic linker for PA64, though
+	     it is possible that will change in the future.  */
+	    
+	  /* This could be a local function that had its address taken, in
+	     which case H will be NULL.  */
+	  if (h)
+	    h->elf_link_hash_flags |= ELF_LINK_HASH_NEEDS_PLT;
+	}
+
+      /* Add a new dynamic relocation to the chain of dynamic
+	 relocations for this symbol.  */
+      if ((need_entry & NEED_DYNREL) && (sec->flags & SEC_ALLOC))
+	{
+	  if (! hppa_info->other_rel_sec
+	      && ! get_reloc_section (abfd, hppa_info, sec))
+	    goto err_out;
+
+	  if (!count_dyn_reloc (abfd, dyn_h, dynrel_type, sec,
+				sec_symndx, rel->r_offset, rel->r_addend))
+	    goto err_out;
+
+	  /* If we are building a shared library and we just recorded
+	     a dynamic R_PARISC_FPTR64 relocation, then make sure the
+	     section symbol for this section ends up in the dynamic
+	     symbol table.  */
+	  if (info->shared && dynrel_type == R_PARISC_FPTR64
+	      && ! (_bfd_elf64_link_record_local_dynamic_symbol
+		    (info, abfd, sec_symndx)))
+	    return false;
+	}
+    }
+
+  if (buf)
+    free (buf);
+  return true;
+
+ err_out:
+  if (buf)
+    free (buf);
+  return false;
+}
+
+struct elf64_hppa_allocate_data
+{
+  struct bfd_link_info *info;
+  bfd_size_type ofs;
+};
+
+/* Should we do dynamic things to this symbol?  */
+
+static boolean
+elf64_hppa_dynamic_symbol_p (h, info)
+     struct elf_link_hash_entry *h;
+     struct bfd_link_info *info;
+{
+  if (h == NULL)
+    return false;
+
+  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->dynindx == -1)
+    return false;
+
+  if (h->root.type == bfd_link_hash_undefweak
+      || h->root.type == bfd_link_hash_defweak)
+    return true;
+
+  if (h->root.root.string[0] == '$' && h->root.root.string[1] == '$')
+    return false;
+
+  if ((info->shared && !info->symbolic)
+      || ((h->elf_link_hash_flags
+	   & (ELF_LINK_HASH_DEF_DYNAMIC | ELF_LINK_HASH_REF_REGULAR))
+	  == (ELF_LINK_HASH_DEF_DYNAMIC | ELF_LINK_HASH_REF_REGULAR)))
+    return true;
+
+  return false;
+}
+
+/* Mark all funtions exported by this file so that we can later allocate
+   entries in .opd for them.  */
+
+static boolean
+elf64_hppa_mark_exported_functions (h, data)
+     struct elf_link_hash_entry *h;
+     PTR data;
+{
+  struct bfd_link_info *info = (struct bfd_link_info *)data;
+  struct elf64_hppa_link_hash_table *hppa_info;
+
+  hppa_info = elf64_hppa_hash_table (info);
+
+  if (h
+      && (h->root.type == bfd_link_hash_defined
+	  || h->root.type == bfd_link_hash_defweak)
+      && h->root.u.def.section->output_section != NULL
+      && h->type == STT_FUNC)
+    {
+       struct elf64_hppa_dyn_hash_entry *dyn_h;
+
+      /* Add this symbol to the PA64 linker hash table.  */
+      dyn_h = elf64_hppa_dyn_hash_lookup (&hppa_info->dyn_hash_table,
+					  h->root.root.string, true, true);
+      BFD_ASSERT (dyn_h);
+      dyn_h->h = h;
+
+      if (! hppa_info->opd_sec
+	  && ! get_opd (hppa_info->root.dynobj, info, hppa_info))
+	return false;
+
+      dyn_h->want_opd = 1;
+      h->elf_link_hash_flags |= ELF_LINK_HASH_NEEDS_PLT;
+    }
+
+  return true;
+}
+
+/* Allocate space for a DLT entry.  */
+
+static boolean
+allocate_global_data_dlt (dyn_h, data)
+     struct elf64_hppa_dyn_hash_entry *dyn_h;
+     PTR data;
+{
+  struct elf64_hppa_allocate_data *x = (struct elf64_hppa_allocate_data *)data;
+
+  if (dyn_h->want_dlt)
+    {
+      struct elf_link_hash_entry *h = dyn_h->h;
+
+      if (x->info->shared)
+	{
+	  /* Possibly add the symbol to the local dynamic symbol
+	     table since we might need to create a dynamic relocation
+	     against it.  */
+	  if (! h
+	      || (h && h->dynindx == -1))
+	    {
+	      bfd *owner;
+	      owner = (h ? h->root.u.def.section->owner : dyn_h->owner);
+
+	      if (!_bfd_elf64_link_record_local_dynamic_symbol
+		    (x->info, owner, dyn_h->sym_indx))
+		return false;
+	    }
+	}
+
+      dyn_h->dlt_offset = x->ofs;
+      x->ofs += DLT_ENTRY_SIZE;
+    }
+  return true;
+}
+
+/* Allocate space for a DLT.PLT entry.  */
+
+static boolean
+allocate_global_data_plt (dyn_h, data)
+     struct elf64_hppa_dyn_hash_entry *dyn_h;
+     PTR data;
+{
+  struct elf64_hppa_allocate_data *x = (struct elf64_hppa_allocate_data *)data;
+
+  if (dyn_h->want_plt
+      && elf64_hppa_dynamic_symbol_p (dyn_h->h, x->info)
+      && !((dyn_h->h->root.type == bfd_link_hash_defined
+	    || dyn_h->h->root.type == bfd_link_hash_defweak)
+	   && dyn_h->h->root.u.def.section->output_section != NULL))
+    {
+      dyn_h->plt_offset = x->ofs;
+      x->ofs += PLT_ENTRY_SIZE;
+      if (dyn_h->plt_offset < 0x2000)
+	elf64_hppa_hash_table (x->info)->gp_offset = dyn_h->plt_offset;
+    }
+  else
+    dyn_h->want_plt = 0;
+
+  return true;
+}
+
+/* Allocate space for a STUB entry.  */
+
+static boolean
+allocate_global_data_stub (dyn_h, data)
+     struct elf64_hppa_dyn_hash_entry *dyn_h;
+     PTR data;
+{
+  struct elf64_hppa_allocate_data *x = (struct elf64_hppa_allocate_data *)data;
+
+  if (dyn_h->want_stub
+      && elf64_hppa_dynamic_symbol_p (dyn_h->h, x->info)
+      && !((dyn_h->h->root.type == bfd_link_hash_defined
+	    || dyn_h->h->root.type == bfd_link_hash_defweak)
+	   && dyn_h->h->root.u.def.section->output_section != NULL))
+    {
+      dyn_h->stub_offset = x->ofs;
+      x->ofs += sizeof (plt_stub);
+    }
+  else
+    dyn_h->want_stub = 0;
+  return true;
+}
+
+/* Allocate space for a FPTR entry.  */
+
+static boolean
+allocate_global_data_opd (dyn_h, data)
+     struct elf64_hppa_dyn_hash_entry *dyn_h;
+     PTR data;
+{
+  struct elf64_hppa_allocate_data *x = (struct elf64_hppa_allocate_data *)data;
+
+  if (dyn_h->want_opd)
+    {
+      struct elf_link_hash_entry *h = dyn_h->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;
+
+      /* We never need an opd entry for a symbol which is not
+	 defined by this output file.  */
+      if (h && h->root.type == bfd_link_hash_undefined)
+	dyn_h->want_opd = 0;
+
+      /* If we are creating a shared library, took the address of a local
+	 function or might export this function from this object file, then
+	 we have to create an opd descriptor.  */
+      else if (x->info->shared
+	       || h == NULL
+	       || h->dynindx == -1
+	       || ((h->root.type == bfd_link_hash_defined
+		    || h->root.type == bfd_link_hash_defweak)
+		   && h->root.u.def.section->output_section != NULL))
+	{
+	  /* If we are creating a shared library, then we will have to
+	     create a runtime relocation for the symbol to properly
+	     initialize the .opd entry.  Make sure the symbol gets
+	     added to the dynamic symbol table.  */
+	  if (x->info->shared
+	      && (h == NULL || (h->dynindx == -1)))
+	    {
+	      bfd *owner;
+	      owner = (h ? h->root.u.def.section->owner : dyn_h->owner);
+
+	      if (!_bfd_elf64_link_record_local_dynamic_symbol
+		    (x->info, owner, dyn_h->sym_indx))
+		return false;
+	    }
+
+	  /* This may not be necessary or desirable anymore now that
+	     we have some support for dealing with section symbols
+	     in dynamic relocs.  But name munging does make the result
+	     much easier to debug.  ie, the EPLT reloc will reference
+	     a symbol like .foobar, instead of .text + offset.  */
+	  if (x->info->shared && h)
+	    {
+	      char *new_name;
+	      struct elf_link_hash_entry *nh;
+
+	      new_name = alloca (strlen (h->root.root.string) + 2);
+	      new_name[0] = '.';
+	      strcpy (new_name + 1, h->root.root.string);
+
+	      nh = elf_link_hash_lookup (elf_hash_table (x->info),
+					 new_name, true, true, true);
+
+	      nh->root.type = h->root.type;
+	      nh->root.u.def.value = h->root.u.def.value;
+	      nh->root.u.def.section = h->root.u.def.section;
+
+	      if (! bfd_elf64_link_record_dynamic_symbol (x->info, nh))
+		return false;
+
+	     }
+	  dyn_h->opd_offset = x->ofs;
+	  x->ofs += OPD_ENTRY_SIZE;
+	}
+
+      /* Otherwise we do not need an opd entry.  */
+      else
+	dyn_h->want_opd = 0;
+    }
+  return true;
+}
+
+/* HP requires the EI_OSABI field to be filled in.  The assignment to
+   EI_ABIVERSION may not be strictly necessary.  */
+
+static void
+elf64_hppa_post_process_headers (abfd, link_info)
+     bfd * abfd;
+     struct bfd_link_info * link_info ATTRIBUTE_UNUSED;
+{
+  Elf_Internal_Ehdr * i_ehdrp;
+
+  i_ehdrp = elf_elfheader (abfd);
+
+  i_ehdrp->e_ident[EI_OSABI] = ELFOSABI_HPUX;
+  i_ehdrp->e_ident[EI_ABIVERSION] = 1;
+}
+
+/* Create function descriptor section (.opd).  This section is called .opd
+   because it contains "official prodecure 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 boolean
+get_opd (abfd, info, hppa_info)
+     bfd *abfd;
+     struct bfd_link_info *info;
+     struct elf64_hppa_link_hash_table *hppa_info;
+{
+  asection *opd;
+  bfd *dynobj;
+
+  opd = hppa_info->opd_sec;
+  if (!opd)
+    {
+      dynobj = hppa_info->root.dynobj;
+      if (!dynobj)
+	hppa_info->root.dynobj = dynobj = abfd;
+
+      opd = bfd_make_section (dynobj, ".opd");
+      if (!opd
+	  || !bfd_set_section_flags (dynobj, opd,
+				     (SEC_ALLOC
+				      | SEC_LOAD
+				      | SEC_HAS_CONTENTS
+				      | SEC_IN_MEMORY
+				      | SEC_LINKER_CREATED))
+	  || !bfd_set_section_alignment (abfd, opd, 3))
+	{
+	  BFD_ASSERT (0);
+	  return false;
+	}
+
+      hppa_info->opd_sec = opd;
+    }
+
+  return true;
+}
+
+/* Create the PLT section.  */
+
+static boolean
+get_plt (abfd, info, hppa_info)
+     bfd *abfd;
+     struct bfd_link_info *info;
+     struct elf64_hppa_link_hash_table *hppa_info;
+{
+  asection *plt;
+  bfd *dynobj;
+
+  plt = hppa_info->plt_sec;
+  if (!plt)
+    {
+      dynobj = hppa_info->root.dynobj;
+      if (!dynobj)
+	hppa_info->root.dynobj = dynobj = abfd;
+
+      plt = bfd_make_section (dynobj, ".plt");
+      if (!plt
+	  || !bfd_set_section_flags (dynobj, plt,
+				     (SEC_ALLOC
+				      | SEC_LOAD
+				      | SEC_HAS_CONTENTS
+				      | SEC_IN_MEMORY
+				      | SEC_LINKER_CREATED))
+	  || !bfd_set_section_alignment (abfd, plt, 3))
+	{
+	  BFD_ASSERT (0);
+	  return false;
+	}
+
+      hppa_info->plt_sec = plt;
+    }
+
+  return true;
+}
+
+/* Create the DLT section.  */
+
+static boolean
+get_dlt (abfd, info, hppa_info)
+     bfd *abfd;
+     struct bfd_link_info *info;
+     struct elf64_hppa_link_hash_table *hppa_info;
+{
+  asection *dlt;
+  bfd *dynobj;
+
+  dlt = hppa_info->dlt_sec;
+  if (!dlt)
+    {
+      dynobj = hppa_info->root.dynobj;
+      if (!dynobj)
+	hppa_info->root.dynobj = dynobj = abfd;
+
+      dlt = bfd_make_section (dynobj, ".dlt");
+      if (!dlt
+	  || !bfd_set_section_flags (dynobj, dlt,
+				     (SEC_ALLOC
+				      | SEC_LOAD
+				      | SEC_HAS_CONTENTS
+				      | SEC_IN_MEMORY
+				      | SEC_LINKER_CREATED))
+	  || !bfd_set_section_alignment (abfd, dlt, 3))
+	{
+	  BFD_ASSERT (0);
+	  return false;
+	}
+
+      hppa_info->dlt_sec = dlt;
+    }
+
+  return true;
+}
+
+/* Create the stubs section.  */
+
+static boolean
+get_stub (abfd, info, hppa_info)
+     bfd *abfd;
+     struct bfd_link_info *info;
+     struct elf64_hppa_link_hash_table *hppa_info;
+{
+  asection *stub;
+  bfd *dynobj;
+
+  stub = hppa_info->stub_sec;
+  if (!stub)
+    {
+      dynobj = hppa_info->root.dynobj;
+      if (!dynobj)
+	hppa_info->root.dynobj = dynobj = abfd;
+
+      stub = bfd_make_section (dynobj, ".stub");
+      if (!stub
+	  || !bfd_set_section_flags (dynobj, stub,
+				     (SEC_ALLOC
+				      | SEC_LOAD
+				      | SEC_HAS_CONTENTS
+				      | SEC_IN_MEMORY
+				      | SEC_READONLY
+				      | SEC_LINKER_CREATED))
+	  || !bfd_set_section_alignment (abfd, stub, 3))
+	{
+	  BFD_ASSERT (0);
+	  return false;
+	}
+
+      hppa_info->stub_sec = stub;
+    }
+
+  return true;
+}
+
+/* Create sections necessary for dynamic linking.  This is only a rough
+   cut and will likely change as we learn more about the somewhat
+   unusual dynamic linking scheme HP uses.
+
+   .stub:
+	Contains code to implement cross-space calls.  The first time one
+	of the stubs is used it will call into the dynamic linker, later
+	calls will go straight to the target.
+
+	The only stub we support right now looks like
+
+	ldd OFFSET(%dp),%r1
+	bve %r0(%r1)
+	ldd OFFSET+8(%dp),%dp
+
+	Other stubs may be needed in the future.  We may want the remove
+	the break/nop instruction.  It is only used right now to keep the
+	offset of a .plt entry and a .stub entry in sync.
+
+   .dlt:
+	This is what most people call the .got.  HP used a different name.
+	Losers.
+
+   .rela.dlt:
+	Relocations for the DLT.
+
+   .plt:
+	Function pointers as address,gp pairs.
+
+   .rela.plt:
+	Should contain dynamic IPLT (and EPLT?) relocations.
+
+   .opd:
+	FPTRS 
+
+   .rela.opd:
+	EPLT relocations for symbols exported from shared libraries.  */
+
+static boolean
+elf64_hppa_create_dynamic_sections (abfd, info)
+     bfd *abfd;
+     struct bfd_link_info *info;
+{
+  asection *s;
+
+  if (! get_stub (abfd, info, elf64_hppa_hash_table (info)))
+    return false;
+
+  if (! get_dlt (abfd, info, elf64_hppa_hash_table (info)))
+    return false;
+
+  if (! get_plt (abfd, info, elf64_hppa_hash_table (info)))
+    return false;
+
+  if (! get_opd (abfd, info, elf64_hppa_hash_table (info)))
+    return false;
+
+  s = bfd_make_section(abfd, ".rela.dlt");
+  if (s == NULL
+      || !bfd_set_section_flags (abfd, s, (SEC_ALLOC | SEC_LOAD
+					   | SEC_HAS_CONTENTS
+					   | SEC_IN_MEMORY
+					   | SEC_READONLY
+					   | SEC_LINKER_CREATED))
+      || !bfd_set_section_alignment (abfd, s, 3))
+    return false;
+  elf64_hppa_hash_table (info)->dlt_rel_sec = s;
+
+  s = bfd_make_section(abfd, ".rela.plt");
+  if (s == NULL
+      || !bfd_set_section_flags (abfd, s, (SEC_ALLOC | SEC_LOAD
+					   | SEC_HAS_CONTENTS
+					   | SEC_IN_MEMORY
+					   | SEC_READONLY
+					   | SEC_LINKER_CREATED))
+      || !bfd_set_section_alignment (abfd, s, 3))
+    return false;
+  elf64_hppa_hash_table (info)->plt_rel_sec = s;
+
+  s = bfd_make_section(abfd, ".rela.data");
+  if (s == NULL
+      || !bfd_set_section_flags (abfd, s, (SEC_ALLOC | SEC_LOAD
+					   | SEC_HAS_CONTENTS
+					   | SEC_IN_MEMORY
+					   | SEC_READONLY
+					   | SEC_LINKER_CREATED))
+      || !bfd_set_section_alignment (abfd, s, 3))
+    return false;
+  elf64_hppa_hash_table (info)->other_rel_sec = s;
+
+  s = bfd_make_section(abfd, ".rela.opd");
+  if (s == NULL
+      || !bfd_set_section_flags (abfd, s, (SEC_ALLOC | SEC_LOAD
+					   | SEC_HAS_CONTENTS
+					   | SEC_IN_MEMORY
+					   | SEC_READONLY
+					   | SEC_LINKER_CREATED))
+      || !bfd_set_section_alignment (abfd, s, 3))
+    return false;
+  elf64_hppa_hash_table (info)->opd_rel_sec = s;
+
+  return true;
+}
+
+/* Allocate dynamic relocations for those symbols that turned out
+   to be dynamic.  */
+
+static boolean
+allocate_dynrel_entries (dyn_h, data)
+     struct elf64_hppa_dyn_hash_entry *dyn_h;
+     PTR data;
+{
+  struct elf64_hppa_allocate_data *x = (struct elf64_hppa_allocate_data *)data;
+  struct elf64_hppa_link_hash_table *hppa_info;
+  struct elf64_hppa_dyn_reloc_entry *rent;
+  boolean dynamic_symbol, shared;
+
+  hppa_info = elf64_hppa_hash_table (x->info);
+  dynamic_symbol = elf64_hppa_dynamic_symbol_p (dyn_h->h, x->info);
+  shared = x->info->shared;
+
+  /* We may need to allocate relocations for a non-dynamic symbol
+     when creating a shared library.  */
+  if (!dynamic_symbol && !shared)
+    return true;
+
+  /* Take care of the normal data relocations.  */
+
+  for (rent = dyn_h->reloc_entries; rent; rent = rent->next)
+    {
+      switch (rent->type)
+	{
+	case R_PARISC_FPTR64:
+	  /* Allocate one iff we are not building a shared library and
+	     !want_opd, which by this point will be true only if we're
+	     actually allocating one statically in the main executable.  */
+	  if (!x->info->shared && dyn_h->want_opd)
+	    continue;
+	  break;
+	}
+      hppa_info->other_rel_sec->_raw_size += sizeof (Elf64_External_Rela);
+
+      /* Make sure this symbol gets into the dynamic symbol table if it is
+	 not already recorded.  ?!? This should not be in the loop since
+	 the symbol need only be added once.  */
+      if (dyn_h->h == 0 || dyn_h->h->dynindx == -1)
+	if (!_bfd_elf64_link_record_local_dynamic_symbol
+	    (x->info, rent->sec->owner, dyn_h->sym_indx))
+	  return false;
+    }
+
+  /* Take care of the GOT and PLT relocations.  */
+
+  if ((dynamic_symbol || shared) && dyn_h->want_dlt)
+    hppa_info->dlt_rel_sec->_raw_size += sizeof (Elf64_External_Rela);
+
+  /* If we are building a shared library, then every symbol that has an
+     opd entry will need an EPLT relocation to relocate the symbol's address
+     and __gp value based on the runtime load address.  */
+  if (shared && dyn_h->want_opd)
+    hppa_info->opd_rel_sec->_raw_size += sizeof (Elf64_External_Rela);
+
+  if (dyn_h->want_plt && dynamic_symbol)
+    {
+      bfd_size_type t = 0;
+
+      /* Dynamic symbols get one IPLT relocation.  Local symbols in
+	 shared libraries get two REL relocations.  Local symbols in
+	 main applications get nothing.  */
+      if (dynamic_symbol)
+	t = sizeof (Elf64_External_Rela);
+      else if (shared)
+	t = 2 * sizeof (Elf64_External_Rela);
+
+      hppa_info->plt_rel_sec->_raw_size += t;
+    }
+
+  return true;
+}
+
+/* Adjust a symbol defined by a dynamic object and referenced by a
+   regular object.  */
+
+static boolean
+elf64_hppa_adjust_dynamic_symbol (info, h)
+     struct bfd_link_info *info;
+     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->weakdef != NULL)
+    {
+      BFD_ASSERT (h->weakdef->root.type == bfd_link_hash_defined
+		  || h->weakdef->root.type == bfd_link_hash_defweak);
+      h->root.u.def.section = h->weakdef->root.u.def.section;
+      h->root.u.def.value = h->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 PA64 code is canonically PIC, so as a rule we can avoid this sort
+     of hackery.  */
+
+  return true;
+}
+
+/* Set the final sizes of the dynamic sections and allocate memory for
+   the contents of our special sections.  */
+
+static boolean
+elf64_hppa_size_dynamic_sections (output_bfd, info)
+     bfd *output_bfd;
+     struct bfd_link_info *info;
+{
+  bfd *dynobj;
+  asection *s;
+  boolean plt;
+  boolean relocs;
+  boolean reltext;
+  boolean stubs;
+  struct elf64_hppa_allocate_data data;
+  struct elf64_hppa_link_hash_table *hppa_info;
+
+  hppa_info = elf64_hppa_hash_table (info);
+
+  dynobj = elf_hash_table (info)->dynobj;
+  BFD_ASSERT (dynobj != NULL);
+
+  if (elf_hash_table (info)->dynamic_sections_created)
+    {
+      /* Set the contents of the .interp section to the interpreter.  */
+      if (! info->shared)
+	{
+	  s = bfd_get_section_by_name (dynobj, ".interp");
+	  BFD_ASSERT (s != NULL);
+	  s->_raw_size = sizeof ELF_DYNAMIC_INTERPRETER;
+	  s->contents = (unsigned char *) ELF_DYNAMIC_INTERPRETER;
+	}
+    }
+  else
+    {
+      /* We may have created entries in the .rela.got section.
+	 However, if we are not creating the dynamic sections, we will
+	 not actually use these entries.  Reset the size of .rela.dlt,
+	 which will cause it to get stripped from the output file
+	 below.  */
+      s = bfd_get_section_by_name (dynobj, ".rela.dlt");
+      if (s != NULL)
+	s->_raw_size = 0;
+    }
+
+  /* Allocate the GOT entries.  */
+
+  data.info = info;
+  if (elf64_hppa_hash_table (info)->dlt_sec)
+    {
+      data.ofs = 0x0;
+      elf64_hppa_dyn_hash_traverse (&hppa_info->dyn_hash_table,
+				    allocate_global_data_dlt, &data);
+      hppa_info->dlt_sec->_raw_size = data.ofs;
+
+      data.ofs = 0x0;
+      elf64_hppa_dyn_hash_traverse (&hppa_info->dyn_hash_table,
+				    allocate_global_data_plt, &data);
+      hppa_info->plt_sec->_raw_size = data.ofs;
+
+      data.ofs = 0x0;
+      elf64_hppa_dyn_hash_traverse (&hppa_info->dyn_hash_table,
+				    allocate_global_data_stub, &data);
+      hppa_info->stub_sec->_raw_size = data.ofs;
+    }
+
+  /* Mark each function this program exports so that we will allocate
+     space in the .opd section for each function's FPTR.
+
+     We have to traverse the main linker hash table since we have to
+     find functions which may not have been mentioned in any relocs.  */
+  elf_link_hash_traverse (elf_hash_table (info),
+			  elf64_hppa_mark_exported_functions,
+			  info);
+
+  /* Allocate space for entries in the .opd section.  */
+  if (elf64_hppa_hash_table (info)->opd_sec)
+    {
+      data.ofs = 0;
+      elf64_hppa_dyn_hash_traverse (&hppa_info->dyn_hash_table,
+				    allocate_global_data_opd, &data);
+      hppa_info->opd_sec->_raw_size = data.ofs;
+    }
+
+  /* Now allocate space for dynamic relocations, if necessary.  */
+  if (hppa_info->root.dynamic_sections_created)
+    elf64_hppa_dyn_hash_traverse (&hppa_info->dyn_hash_table,
+				  allocate_dynrel_entries, &data);
+
+  /* The sizes of all the sections are set.  Allocate memory for them.  */
+  plt = false;
+  relocs = false;
+  reltext = false;
+  for (s = dynobj->sections; s != NULL; s = s->next)
+    {
+      const char *name;
+      boolean strip;
+
+      if ((s->flags & SEC_LINKER_CREATED) == 0)
+	continue;
+
+      /* 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, s);
+
+      strip = 0;
+
+      if (strcmp (name, ".plt") == 0)
+	{
+	  if (s->_raw_size == 0)
+	    {
+	      /* Strip this section if we don't need it; see the
+		 comment below.  */
+	      strip = true;
+	    }
+	  else
+	    {
+	      /* Remember whether there is a PLT.  */
+	      plt = true;
+	    }
+	}
+      else if (strcmp (name, ".dlt") == 0)
+	{
+	  if (s->_raw_size == 0)
+	    {
+	      /* Strip this section if we don't need it; see the
+		 comment below.  */
+	      strip = true;
+	    }
+	}
+      else if (strcmp (name, ".opd") == 0)
+	{
+	  if (s->_raw_size == 0)
+	    {
+	      /* Strip this section if we don't need it; see the
+		 comment below.  */
+	      strip = true;
+	    }
+	}
+      else if (strncmp (name, ".rela", 4) == 0)
+	{
+	  if (s->_raw_size == 0)
+	    {
+	      /* If we don't need this section, strip it from the
+		 output file.  This is mostly to handle .rela.bss and
+		 .rela.plt.  We must create both sections in
+		 create_dynamic_sections, because they must be created
+		 before the linker maps input sections to output
+		 sections.  The linker does that before
+		 adjust_dynamic_symbol is called, and it is that
+		 function which decides whether anything needs to go
+		 into these sections.  */
+	      strip = true;
+	    }
+	  else
+	    {
+	      asection *target;
+
+	      /* Remember whether there are any reloc sections other
+		 than .rela.plt.  */
+	      if (strcmp (name, ".rela.plt") != 0)
+		{
+		  const char *outname;
+
+		  relocs = true;
+
+		  /* If this relocation section applies to a read only
+		     section, then we probably need a DT_TEXTREL
+		     entry.  The entries in the .rela.plt section
+		     really apply to the .got section, which we
+		     created ourselves and so know is not readonly.  */
+		  outname = bfd_get_section_name (output_bfd,
+						  s->output_section);
+		  target = bfd_get_section_by_name (output_bfd, outname + 4);
+		  if (target != NULL
+		      && (target->flags & SEC_READONLY) != 0
+		      && (target->flags & SEC_ALLOC) != 0)
+		    reltext = true;
+		}
+
+	      /* We use the reloc_count field as a counter if we need
+		 to copy relocs into the output file.  */
+	      s->reloc_count = 0;
+	    }
+	}
+      else if (strncmp (name, ".dlt", 4) != 0
+	       && strcmp (name, ".stub") != 0
+	       && strcmp (name, ".got") != 0)
+	{
+	  /* It's not one of our sections, so don't allocate space.  */
+	  continue;
+	}
+
+      if (strip)
+	{
+	  _bfd_strip_section_from_output (info, s);
+	  continue;
+	}
+
+      /* Allocate memory for the section contents if it has not
+	 been allocated already.  */
+      if (s->contents == NULL)
+	{
+	  s->contents = (bfd_byte *) bfd_alloc (dynobj, s->_raw_size);
+	  if (s->contents == NULL && s->_raw_size != 0)
+	    return false;
+	}
+    }
+
+  if (elf_hash_table (info)->dynamic_sections_created)
+    {
+      /* Always create a DT_PLTGOT.  It actually has nothing to do with
+	 the PLT, it is how we communicate the __gp value of a load
+	 module to the dynamic linker.  */
+      if (! bfd_elf64_add_dynamic_entry (info, DT_HP_DLD_FLAGS, 0)
+	  || ! bfd_elf64_add_dynamic_entry (info, DT_PLTGOT, 0))
+	return false;
+
+      /* Add some entries to the .dynamic section.  We fill in the
+	 values later, in elf64_hppa_finish_dynamic_sections, but we
+	 must add the entries now so that we get the correct size for
+	 the .dynamic section.  The DT_DEBUG entry is filled in by the
+	 dynamic linker and used by the debugger.  */
+      if (! info->shared)
+	{
+	  if (! bfd_elf64_add_dynamic_entry (info, DT_DEBUG, 0)
+	      || ! bfd_elf64_add_dynamic_entry (info, DT_HP_DLD_HOOK, 0)
+	      || ! bfd_elf64_add_dynamic_entry (info, DT_HP_LOAD_MAP, 0))
+	    return false;
+	}
+
+      if (plt)
+	{
+	  if (! bfd_elf64_add_dynamic_entry (info, DT_PLTRELSZ, 0)
+	      || ! bfd_elf64_add_dynamic_entry (info, DT_PLTREL, DT_RELA)
+	      || ! bfd_elf64_add_dynamic_entry (info, DT_JMPREL, 0))
+	    return false;
+	}
+
+      if (relocs)
+	{
+	  if (! bfd_elf64_add_dynamic_entry (info, DT_RELA, 0)
+	      || ! bfd_elf64_add_dynamic_entry (info, DT_RELASZ, 0)
+	      || ! bfd_elf64_add_dynamic_entry (info, DT_RELAENT,
+						sizeof (Elf64_External_Rela)))
+	    return false;
+	}
+
+      if (reltext)
+	{
+	  if (! bfd_elf64_add_dynamic_entry (info, DT_TEXTREL, 0))
+	    return false;
+	}
+    }
+
+  return true;
+}
+
+/* Called after we have output the symbol into the dynamic symbol
+   table, but before we output the symbol into the normal symbol
+   table.
+
+   For some symbols we had to change their address when outputting
+   the dynamic symbol table.  We undo that change here so that
+   the symbols have their expected value in the normal symbol
+   table.  Ick.  */
+
+static boolean
+elf64_hppa_link_output_symbol_hook (abfd, info, name, sym, input_sec)
+     bfd *abfd;
+     struct bfd_link_info *info;
+     const char *name;
+     Elf_Internal_Sym *sym;
+     asection *input_sec;
+{
+  struct elf64_hppa_link_hash_table *hppa_info;
+  struct elf64_hppa_dyn_hash_entry *dyn_h;
+
+  /* We may be called with the file symbol or section symbols.
+     They never need munging, so it is safe to ignore them.  */
+  if (!name)
+    return true;
+
+  /* Get the PA dyn_symbol (if any) associated with NAME.  */
+  hppa_info = elf64_hppa_hash_table (info);
+  dyn_h = elf64_hppa_dyn_hash_lookup (&hppa_info->dyn_hash_table,
+				      name, false, false);
+
+  /* Function symbols for which we created .opd entries were munged
+     by finish_dynamic_symbol and have to be un-munged here.  */
+  if (dyn_h && dyn_h->want_opd)
+    {
+      /* Restore the saved value and section index.  */
+      sym->st_value = dyn_h->st_value;
+      sym->st_shndx = dyn_h->st_shndx; 
+    }
+
+  return true;
+}
+
+/* Finish up dynamic symbol handling.  We set the contents of various
+   dynamic sections here.  */
+
+static boolean
+elf64_hppa_finish_dynamic_symbol (output_bfd, info, h, sym)
+     bfd *output_bfd;
+     struct bfd_link_info *info;
+     struct elf_link_hash_entry *h;
+     Elf_Internal_Sym *sym;
+{
+  asection *stub, *splt, *sdlt, *sopd, *spltrel, *sdltrel;
+  struct elf64_hppa_link_hash_table *hppa_info;
+  struct elf64_hppa_dyn_hash_entry *dyn_h;
+
+  hppa_info = elf64_hppa_hash_table (info);
+  dyn_h = elf64_hppa_dyn_hash_lookup (&hppa_info->dyn_hash_table,
+				      h->root.root.string, false, false);
+
+  stub = hppa_info->stub_sec;
+  splt = hppa_info->plt_sec;
+  sdlt = hppa_info->dlt_sec;
+  sopd = hppa_info->opd_sec;
+  spltrel = hppa_info->plt_rel_sec;
+  sdltrel = hppa_info->dlt_rel_sec;
+
+  BFD_ASSERT (stub != NULL && splt != NULL
+	      && sopd != NULL && sdlt != NULL)
+
+  /* Incredible.  It is actually necessary to NOT use the symbol's real
+     value when building the dynamic symbol table for a shared library.
+     At least for symbols that refer to functions.
+
+     We will store a new value and section index into the symbol long
+     enough to output it into the dynamic symbol table, then we restore
+     the original values (in elf64_hppa_link_output_symbol_hook).  */
+  if (dyn_h && dyn_h->want_opd)
+    {
+      /* Save away the original value and section index so that we
+	 can restore them later.  */
+      dyn_h->st_value = sym->st_value;
+      dyn_h->st_shndx = sym->st_shndx;
+
+      /* For the dynamic symbol table entry, we want the value to be
+	 address of this symbol's entry within the .opd section.  */
+      sym->st_value = (dyn_h->opd_offset
+		       + sopd->output_offset
+		       + sopd->output_section->vma);
+      sym->st_shndx = _bfd_elf_section_from_bfd_section (output_bfd,
+							 sopd->output_section);
+    }
+
+  /* Initialize a .plt entry if requested.  */
+  if (dyn_h && dyn_h->want_plt
+      && elf64_hppa_dynamic_symbol_p (dyn_h->h, info))
+    {
+      bfd_vma value;
+      Elf_Internal_Rela rel;
+
+      /* We do not actually care about the value in the PLT entry
+	 if we are creating a shared library and the symbol is
+	 still undefined, we create a dynamic relocation to fill
+	 in the correct value.  */
+      if (info->shared && h->root.type == bfd_link_hash_undefined)
+	value = 0;
+      else
+	value = (h->root.u.def.value + h->root.u.def.section->vma);
+
+      /* Fill in the entry in the procedure linkage table. 
+
+	 The format of a plt entry is
+	 <funcaddr> <__gp>. 
+
+	 plt_offset is the offset within the PLT section at which to
+	 install the PLT entry. 
+
+	 We are modifying the in-memory PLT contents here, so we do not add
+	 in the output_offset of the PLT section.  */
+
+      bfd_put_64 (splt->owner, value, splt->contents + dyn_h->plt_offset);
+      value = _bfd_get_gp_value (splt->output_section->owner);
+      bfd_put_64 (splt->owner, value, splt->contents + dyn_h->plt_offset + 0x8);
+
+      /* Create a dynamic IPLT relocation for this entry.
+
+	 We are creating a relocation in the output file's PLT section,
+	 which is included within the DLT secton.  So we do need to include
+	 the PLT's output_offset in the computation of the relocation's
+	 address.  */
+      rel.r_offset = (dyn_h->plt_offset + splt->output_offset
+		      + splt->output_section->vma);
+      rel.r_info = ELF64_R_INFO (h->dynindx, R_PARISC_IPLT);
+      rel.r_addend = 0;
+
+      bfd_elf64_swap_reloca_out (splt->output_section->owner, &rel,
+				 (((Elf64_External_Rela *)
+				   spltrel->contents)
+				  + spltrel->reloc_count));
+      spltrel->reloc_count++;
+    }
+
+  /* Initialize an external call stub entry if requested.  */
+  if (dyn_h && dyn_h->want_stub
+      && elf64_hppa_dynamic_symbol_p (dyn_h->h, info))
+    {
+      bfd_vma value;
+      int insn;
+
+      /* Install the generic stub template.
+
+	 We are modifying the contents of the stub section, so we do not
+	 need to include the stub section's output_offset here.  */
+      memcpy (stub->contents + dyn_h->stub_offset, plt_stub, sizeof (plt_stub));
+
+      /* Fix up the first ldd instruction.
+
+	 We are modifying the contents of the STUB section in memory,
+	 so we do not need to include its output offset in this computation. 
+
+	 Note the plt_offset value is the value of the PLT entry relative to
+	 the start of the PLT section.  These instructions will reference
+	 data relative to the value of __gp, which may not necessarily have
+	 the same address as the start of the PLT section.
+
+	 gp_offset contains the offset of __gp within the PLT section.  */
+      value = dyn_h->plt_offset - hppa_info->gp_offset;
+      
+      insn = bfd_get_32 (stub->owner, stub->contents + dyn_h->stub_offset);
+      insn &= 0xffffc00e;
+      insn |= ((value & 0x2000) >> 13);
+      value &= 0x1ff8;
+      value <<= 1;
+      bfd_put_32 (stub->owner, (insn | value),
+		  stub->contents + dyn_h->stub_offset);
+
+      /* Fix up the second ldd instruction.  */
+      value = dyn_h->plt_offset - hppa_info->gp_offset + 8;
+      
+      insn = bfd_get_32 (stub->owner, stub->contents + dyn_h->stub_offset + 8);
+      insn &= 0xffffc00e;
+      insn |= ((value & 0x2000) >> 13);
+      value &= 0x1ff8;
+      value <<= 1;
+      bfd_put_32 (stub->owner, (insn | value),
+		  stub->contents + dyn_h->stub_offset + 8);
+    }
+
+  /* Millicode symbols should not be put in the dynamic
+     symbol table under any circumstances.  */
+  if (ELF_ST_TYPE (sym->st_info) == STT_PARISC_MILLI)
+    h->dynindx = -1;
+
+  return true;
+}
+
+/* The .opd section contains FPTRs for each function this file
+   exports.  Initialize the FPTR entries.  */
+
+static boolean
+elf64_hppa_finalize_opd (dyn_h, data)
+     struct elf64_hppa_dyn_hash_entry *dyn_h;
+     PTR data;
+{
+  struct bfd_link_info *info = (struct bfd_link_info *)data;
+  struct elf64_hppa_link_hash_table *hppa_info;
+  struct elf_link_hash_entry *h = dyn_h->h;
+  asection *sopd;
+  asection *sopdrel;
+
+  hppa_info = elf64_hppa_hash_table (info);
+  sopd = hppa_info->opd_sec;
+  sopdrel = hppa_info->opd_rel_sec;
+
+  if (h && dyn_h && dyn_h->want_opd)
+    {
+      bfd_vma value;
+
+      /* The first two words of an .opd entry are zero. 
+
+	 We are modifying the contents of the OPD section in memory, so we
+	 do not need to include its output offset in this computation.  */
+      memset (sopd->contents + dyn_h->opd_offset, 0, 16);
+
+      value = (h->root.u.def.value
+	       + h->root.u.def.section->output_section->vma
+	       + h->root.u.def.section->output_offset);
+
+      /* The next word is the address of the function.  */
+      bfd_put_64 (sopd->owner, value, sopd->contents + dyn_h->opd_offset + 16);
+
+      /* The last word is our local __gp value.  */
+      value = _bfd_get_gp_value (sopd->output_section->owner);
+      bfd_put_64 (sopd->owner, value, sopd->contents + dyn_h->opd_offset + 24);
+    }
+
+  /* If we are generating a shared library, we must generate EPLT relocations
+     for each entry in the .opd, even for static functions (they may have
+     had their address taken).  */
+  if (info->shared && dyn_h && dyn_h->want_opd)
+    {
+      Elf64_Internal_Rela rel;
+      bfd_vma value;
+      int dynindx;
+
+      /* We may need to do a relocation against a local symbol, in
+	 which case we have to look up it's dynamic symbol index off
+	 the local symbol hash table.  */
+      if (h && h->dynindx != -1)
+	dynindx = h->dynindx;
+      else
+	dynindx
+	  = _bfd_elf_link_lookup_local_dynindx (info, dyn_h->owner,
+						dyn_h->sym_indx);
+
+      /* The offset of this relocation is the absolute address of the
+	 .opd entry for this symbol.  */
+      rel.r_offset = (dyn_h->opd_offset + sopd->output_offset
+		      + sopd->output_section->vma);
+
+      /* If H is non-null, then we have an external symbol.
+
+	 It is imperative that we use a different dynamic symbol for the
+	 EPLT relocation if the symbol has global scope.
+
+	 In the dynamic symbol table, the function symbol will have a value
+	 which is address of the function's .opd entry.
+
+	 Thus, we can not use that dynamic symbol for the EPLT relocation
+	 (if we did, the data in the .opd would reference itself rather
+	 than the actual address of the function).  Instead we have to use
+	 a new dynamic symbol which has the same value as the original global
+	 function symbol. 
+
+	 We prefix the original symbol with a "." and use the new symbol in
+	 the EPLT relocation.  This new symbol has already been recorded in
+	 the symbol table, we just have to look it up and use it.
+
+	 We do not have such problems with static functions because we do
+	 not make their addresses in the dynamic symbol table point to
+	 the .opd entry.  Ultimately this should be safe since a static
+	 function can not be directly referenced outside of its shared
+	 library.
+
+	 We do have to play similar games for FPTR relocations in shared
+	 libraries, including those for static symbols.  See the FPTR
+	 handling in elf64_hppa_finalize_dynreloc.  */
+      if (h)
+	{
+	  char *new_name;
+	  struct elf_link_hash_entry *nh;
+
+	  new_name = alloca (strlen (h->root.root.string) + 2);
+	  new_name[0] = '.';
+	  strcpy (new_name + 1, h->root.root.string);
+
+	  nh = elf_link_hash_lookup (elf_hash_table (info),
+				     new_name, false, false, false);
+
+	  /* All we really want from the new symbol is its dynamic
+	     symbol index.  */
+	  dynindx = nh->dynindx;
+	}
+
+      rel.r_addend = 0;
+      rel.r_info = ELF64_R_INFO (dynindx, R_PARISC_EPLT);
+
+      bfd_elf64_swap_reloca_out (sopd->output_section->owner, &rel,
+				 (((Elf64_External_Rela *)
+				   sopdrel->contents)
+				  + sopdrel->reloc_count));
+      sopdrel->reloc_count++;
+    }
+  return true;
+}
+
+/* The .dlt section contains addresses for items referenced through the
+   dlt.  Note that we can have a DLTIND relocation for a local symbol, thus
+   we can not depend on finish_dynamic_symbol to initialize the .dlt.  */
+
+static boolean
+elf64_hppa_finalize_dlt (dyn_h, data)
+     struct elf64_hppa_dyn_hash_entry *dyn_h;
+     PTR data;
+{
+  struct bfd_link_info *info = (struct bfd_link_info *)data;
+  struct elf64_hppa_link_hash_table *hppa_info;
+  asection *sdlt, *sdltrel;
+  struct elf_link_hash_entry *h = dyn_h->h;
+
+  hppa_info = elf64_hppa_hash_table (info);
+
+  sdlt = hppa_info->dlt_sec;
+  sdltrel = hppa_info->dlt_rel_sec;
+
+  /* H/DYN_H may refer to a local variable and we know it's
+     address, so there is no need to create a relocation.  Just install
+     the proper value into the DLT, note this shortcut can not be
+     skipped when building a shared library.  */
+  if (! info->shared && h && dyn_h && dyn_h->want_dlt)
+    {
+      bfd_vma value;
+
+      /* If we had an LTOFF_FPTR style relocation we want the DLT entry
+	 to point to the FPTR entry in the .opd section. 
+
+	 We include the OPD's output offset in this computation as
+	 we are referring to an absolute address in the resulting
+	 object file.  */
+      if (dyn_h->want_opd)
+	{
+	  value = (dyn_h->opd_offset
+		   + hppa_info->opd_sec->output_offset
+		   + hppa_info->opd_sec->output_section->vma);
+	}
+      else
+	{
+	  value = (h->root.u.def.value
+		   + h->root.u.def.section->output_offset);
+
+	  if (h->root.u.def.section->output_section)
+	    value += h->root.u.def.section->output_section->vma;
+	  else
+	    value += h->root.u.def.section->vma;
+	}
+
+      /* We do not need to include the output offset of the DLT section
+	 here because we are modifying the in-memory contents.  */
+      bfd_put_64 (sdlt->owner, value, sdlt->contents + dyn_h->dlt_offset);
+    }
+
+  /* Create a relocation for the DLT entry assocated with this symbol.
+     When building a shared library the symbol does not have to be dynamic.  */
+  if (dyn_h->want_dlt
+      && (elf64_hppa_dynamic_symbol_p (dyn_h->h, info) || info->shared))
+    {
+      Elf64_Internal_Rela rel;
+      int dynindx;
+
+      /* We may need to do a relocation against a local symbol, in
+	 which case we have to look up it's dynamic symbol index off
+	 the local symbol hash table.  */
+      if (h && h->dynindx != -1)
+	dynindx = h->dynindx;
+      else
+	dynindx
+	  = _bfd_elf_link_lookup_local_dynindx (info, dyn_h->owner,
+						dyn_h->sym_indx);
+
+
+      /* Create a dynamic relocation for this entry.  Do include the output
+	 offset of the DLT entry since we need an absolute address in the
+	 resulting object file.  */
+      rel.r_offset = (dyn_h->dlt_offset + sdlt->output_offset
+		      + sdlt->output_section->vma);
+      if (h && h->type == STT_FUNC)
+	  rel.r_info = ELF64_R_INFO (dynindx, R_PARISC_FPTR64);
+      else
+	  rel.r_info = ELF64_R_INFO (dynindx, R_PARISC_DIR64);
+      rel.r_addend = 0;
+
+      bfd_elf64_swap_reloca_out (sdlt->output_section->owner, &rel,
+				 (((Elf64_External_Rela *)
+				   sdltrel->contents)
+				  + sdltrel->reloc_count));
+      sdltrel->reloc_count++;
+    }
+  return true;
+}
+
+/* Finalize the dynamic relocations.  Specifically the FPTR relocations
+   for dynamic functions used to initialize static data.  */
+
+static boolean
+elf64_hppa_finalize_dynreloc (dyn_h, data)
+     struct elf64_hppa_dyn_hash_entry *dyn_h;
+     PTR data;
+{
+  struct bfd_link_info *info = (struct bfd_link_info *)data;
+  struct elf64_hppa_link_hash_table *hppa_info;
+  struct elf_link_hash_entry *h;
+  int dynamic_symbol;
+
+  dynamic_symbol = elf64_hppa_dynamic_symbol_p (dyn_h->h, info);
+
+  if (!dynamic_symbol && !info->shared)
+    return true;
+
+  if (dyn_h->reloc_entries)
+    {
+      struct elf64_hppa_dyn_reloc_entry *rent;
+      int dynindx;
+
+      hppa_info = elf64_hppa_hash_table (info);
+      h = dyn_h->h;
+
+      /* We may need to do a relocation against a local symbol, in
+	 which case we have to look up it's dynamic symbol index off
+	 the local symbol hash table.  */
+      if (h && h->dynindx != -1)
+	dynindx = h->dynindx;
+      else
+	dynindx
+	  = _bfd_elf_link_lookup_local_dynindx (info, dyn_h->owner,
+						dyn_h->sym_indx);
+
+      for (rent = dyn_h->reloc_entries; rent; rent = rent->next)
+	{
+	  Elf64_Internal_Rela rel;
+
+	  switch (rent->type)
+	    {
+	      case R_PARISC_FPTR64:
+	      /* Allocate one iff we are not building a shared library and
+		 !want_opd, which by this point will be true only if we're
+		 actually allocating one statically in the main executable.  */
+	      if (!info->shared && dyn_h->want_opd)
+		continue;
+	      break;
+	    }
+
+	  /* Create a dynamic relocation for this entry. 
+
+	     We need the output offset for the reloc's section because
+	     we are creating an absolute address in the resulting object
+	     file.  */
+	  rel.r_offset = (rent->offset + rent->sec->output_offset
+			  + rent->sec->output_section->vma);
+
+	  /* An FPTR64 relocation implies that we took the address of
+	     a function and that the function has an entry in the .opd
+	     section.  We want the FPTR64 relocation to reference the
+	     entry in .opd.
+
+	     We could munge the symbol value in the dynamic symbol table
+	     (in fact we already do for functions with global scope) to point
+	     to the .opd entry.  Then we could use that dynamic symbol in
+	     this relocation.
+
+	     Or we could do something sensible, not munge the symbol's
+	     address and instead just use a different symbol to reference
+	     the .opd entry.  At least that seems sensible until you
+	     realize there's no local dynamic symbols we can use for that
+	     purpose.  Thus the hair in the check_relocs routine.
+	
+	     We use a section symbol recorded by check_relocs as the
+	     base symbol for the relocation.  The addend is the difference
+	     between the section symbol and the address of the .opd entry.  */
+	  if (info->shared && rent->type == R_PARISC_FPTR64)
+	    {
+	      bfd_vma value, value2;
+	      asymbol *sym;
+	      int elf_index;
+
+	      /* First compute the address of the opd entry for this symbol.  */
+	      value = (dyn_h->opd_offset
+		       + hppa_info->opd_sec->output_section->vma
+		       + hppa_info->opd_sec->output_offset);
+
+	      /* Compute the value of the start of the section with
+		 the relocation.  */
+	      value2 = (rent->sec->output_section->vma
+			+ rent->sec->output_offset);
+
+	      /* Compute the difference between the start of the section
+		 with the relocation and the opd entry.  */
+	      value -= value2;
+		
+	      /* The result becomes the addend of the relocation.  */
+	      rel.r_addend = value;
+
+	      /* The section symbol becomes the symbol for the dynamic
+		 relocation.  */
+	      dynindx
+		= _bfd_elf_link_lookup_local_dynindx (info,
+						      rent->sec->owner,
+						      rent->sec_symndx);
+	    }
+	  else
+	    rel.r_addend = rent->addend;
+
+	  rel.r_info = ELF64_R_INFO (dynindx, rent->type);
+
+	  bfd_elf64_swap_reloca_out (hppa_info->other_rel_sec->output_section->owner,
+				     &rel,
+				     (((Elf64_External_Rela *)
+				      hppa_info->other_rel_sec->contents)
+				      + hppa_info->other_rel_sec->reloc_count));
+	  hppa_info->other_rel_sec->reloc_count++;
+	}
+    }
+
+  return true;
+}
+
+/* Finish up the dynamic sections.  */
+
+static boolean
+elf64_hppa_finish_dynamic_sections (output_bfd, info)
+     bfd *output_bfd;
+     struct bfd_link_info *info;
+{
+  bfd *dynobj;
+  asection *sdyn;
+  struct elf64_hppa_link_hash_table *hppa_info;
+
+  hppa_info = elf64_hppa_hash_table (info);
+
+  /* Finalize the contents of the .opd section.  */
+  elf64_hppa_dyn_hash_traverse (&hppa_info->dyn_hash_table,
+				elf64_hppa_finalize_opd,
+				info);
+
+  elf64_hppa_dyn_hash_traverse (&hppa_info->dyn_hash_table,
+				elf64_hppa_finalize_dynreloc,
+				info);
+
+  /* Finalize the contents of the .dlt section.  */
+  dynobj = elf_hash_table (info)->dynobj;
+  /* Finalize the contents of the .dlt section.  */
+  elf64_hppa_dyn_hash_traverse (&hppa_info->dyn_hash_table,
+				elf64_hppa_finalize_dlt,
+				info);
+
+
+  sdyn = bfd_get_section_by_name (dynobj, ".dynamic");
+
+  if (elf_hash_table (info)->dynamic_sections_created)
+    {
+      Elf64_External_Dyn *dyncon, *dynconend;
+      struct elf_link_hash_entry *h;
+
+      BFD_ASSERT (sdyn != NULL);
+
+      dyncon = (Elf64_External_Dyn *) sdyn->contents;
+      dynconend = (Elf64_External_Dyn *) (sdyn->contents + sdyn->_raw_size);
+      for (; dyncon < dynconend; dyncon++)
+	{
+	  Elf_Internal_Dyn dyn;
+	  asection *s;
+
+	  bfd_elf64_swap_dyn_in (dynobj, dyncon, &dyn);
+
+	  switch (dyn.d_tag)
+	    {
+	    default:
+	      break;
+
+	    case DT_HP_LOAD_MAP:
+	      /* Compute the absolute address of 16byte scratchpad area
+		 for the dynamic linker.
+
+		 By convention the linker script will allocate the scratchpad
+		 area at the start of the .data section.  So all we have to
+		 to is find the start of the .data section.  */
+	      s = bfd_get_section_by_name (output_bfd, ".data");
+	      dyn.d_un.d_ptr = s->vma;
+	      bfd_elf64_swap_dyn_out (output_bfd, &dyn, dyncon);
+	      break;
+
+	    case DT_PLTGOT:
+	      /* HP's use PLTGOT to set the GOT register.  */
+	      dyn.d_un.d_ptr = _bfd_get_gp_value (output_bfd);
+	      bfd_elf64_swap_dyn_out (output_bfd, &dyn, dyncon);
+	      break;
+
+	    case DT_JMPREL:
+	      s = hppa_info->plt_rel_sec;
+	      dyn.d_un.d_ptr = s->output_section->vma + s->output_offset;
+	      bfd_elf64_swap_dyn_out (output_bfd, &dyn, dyncon);
+	      break;
+
+	    case DT_PLTRELSZ:
+	      s = hppa_info->plt_rel_sec;
+	      dyn.d_un.d_val = s->_raw_size;
+	      bfd_elf64_swap_dyn_out (output_bfd, &dyn, dyncon);
+	      break;
+
+	    case DT_RELA:
+	      s = hppa_info->other_rel_sec;
+	      if (! s)
+		s = hppa_info->dlt_rel_sec;
+	      dyn.d_un.d_ptr = s->output_section->vma + s->output_offset;
+	      bfd_elf64_swap_dyn_out (output_bfd, &dyn, dyncon);
+	      break;
+
+	    case DT_RELASZ:
+	      s = hppa_info->other_rel_sec;
+	      dyn.d_un.d_val = s->_raw_size;
+	      s = hppa_info->dlt_rel_sec;
+	      dyn.d_un.d_val += s->_raw_size;
+	      s = hppa_info->opd_rel_sec;
+	      dyn.d_un.d_val += s->_raw_size;
+	      /* There is some question about whether or not the size of
+		 the PLT relocs should be included here.  HP's tools do
+		 it, so we'll emulate them.  */
+	      s = hppa_info->plt_rel_sec;
+	      dyn.d_un.d_val += s->_raw_size;
+	      bfd_elf64_swap_dyn_out (output_bfd, &dyn, dyncon);
+	      break;
+
+	    }
+	}
+    }
+
+  return true;
+}
+
+
+/* Return the number of additional phdrs we will need.
+
+   The generic ELF code only creates PT_PHDRs for executables.  The HP
+   dynamic linker requires PT_PHDRs for dynamic libraries too. 
+
+   This routine indicates that the backend needs one additional program
+   header for that case.
+
+   Note we do not have access to the link info structure here, so we have
+   to guess whether or not we are building a shared library based on the
+   existence of a .interp section.  */
+
+static int
+elf64_hppa_additional_program_headers (abfd)
+     bfd *abfd;
+{
+  asection *s;
+
+  /* If we are creating a shared library, then we have to create a
+     PT_PHDR segment.  HP's dynamic linker chokes without it.  */
+  s = bfd_get_section_by_name (abfd, ".interp");
+  if (! s)
+    return 1;
+  return 0;
+}
+
+/* Allocate and initialize any program headers required by this
+   specific backend.
+
+   The generic ELF code only creates PT_PHDRs for executables.  The HP
+   dynamic linker requires PT_PHDRs for dynamic libraries too. 
+
+   This allocates the PT_PHDR and initializes it in a manner suitable
+   for the HP linker. 
+
+   Note we do not have access to the link info structure here, so we have
+   to guess whether or not we are building a shared library based on the
+   existence of a .interp section.  */
+
+static boolean
+elf64_hppa_modify_segment_map (abfd)
+     bfd *abfd;
+{
+  struct elf_segment_map *m, **pm;
+  asection *s;
+
+  s = bfd_get_section_by_name (abfd, ".interp");
+  if (! s)
+    {
+      for (m = elf_tdata (abfd)->segment_map; m != NULL; m = m->next)
+	if (m->p_type == PT_PHDR)
+	  break;
+      if (m == NULL)
+	{
+	  m = (struct elf_segment_map *) bfd_zalloc (abfd, sizeof *m);
+	  if (m == NULL)
+	    return false;
+
+	  m->p_type = PT_PHDR;
+	  m->p_flags = PF_R | PF_X;
+	  m->p_flags_valid = 1;
+	  m->p_paddr_valid = 1;
+	  m->includes_phdrs = 1;
+
+	  m->next = elf_tdata (abfd)->segment_map;
+	  elf_tdata (abfd)->segment_map = m;
+	}
+    }
+
+  for (m = elf_tdata (abfd)->segment_map; m != NULL; m = m->next)
+    if (m->p_type == PT_LOAD)
+      {
+        int i;
+
+	for (i = 0; i < m->count; i++)
+	  {
+	    /* The code "hint" is not really a hint.  It is a requirement
+	       for certain versions of the HP dynamic linker.  Worse yet,
+	       it must be set even if the shared library does not have
+	       any code in its "text" segment (thus the check for .hash
+	       to catch this situation).  */
+	    if (m->sections[i]->flags & SEC_CODE
+		|| (strcmp (m->sections[i]->name, ".hash") == 0))
+	      m->p_flags |= (PF_X | PF_HP_CODE);
+	  }
+      }
+
+  return true;
+}
+
+/* The hash bucket size is the standard one, namely 4.  */
+
+const struct elf_size_info hppa64_elf_size_info =
+{
+  sizeof (Elf64_External_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, 8,
+  ELFCLASS64, EV_CURRENT,
+  bfd_elf64_write_out_phdrs,
+  bfd_elf64_write_shdrs_and_ehdr,
+  bfd_elf64_write_relocs,
+  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,
+  NULL,
+  NULL,
+  NULL,
+  NULL
+};
+
+#define TARGET_BIG_SYM			bfd_elf64_hppa_vec
+#define TARGET_BIG_NAME			"elf64-hppa"
+#define ELF_ARCH			bfd_arch_hppa
+#define ELF_MACHINE_CODE		EM_PARISC
+/* This is not strictly correct.  The maximum page size for PA2.0 is
+   64M.  But everything still uses 4k.  */
+#define ELF_MAXPAGESIZE			0x1000
+#define bfd_elf64_bfd_reloc_type_lookup elf_hppa_reloc_type_lookup
+#define bfd_elf64_bfd_is_local_label_name       elf_hppa_is_local_label_name
+#define elf_info_to_howto		elf_hppa_info_to_howto
+#define elf_info_to_howto_rel		elf_hppa_info_to_howto_rel
+
+#define elf_backend_section_from_shdr	elf64_hppa_section_from_shdr
+#define elf_backend_object_p		elf64_hppa_object_p
+#define elf_backend_final_write_processing \
+					elf_hppa_final_write_processing
+#define elf_backend_fake_sections 	elf_hppa_fake_sections
+#define elf_backend_add_symbol_hook	elf_hppa_add_symbol_hook
+
+#define elf_backend_relocate_section	    elf_hppa_relocate_section
+
+#define bfd_elf64_bfd_final_link	elf_hppa_final_link
+
+#define elf_backend_create_dynamic_sections \
+					elf64_hppa_create_dynamic_sections
+#define elf_backend_post_process_headers	elf64_hppa_post_process_headers
+
+#define elf_backend_adjust_dynamic_symbol \
+					elf64_hppa_adjust_dynamic_symbol
+
+#define elf_backend_size_dynamic_sections \
+					elf64_hppa_size_dynamic_sections
+
+#define elf_backend_finish_dynamic_symbol \
+					elf64_hppa_finish_dynamic_symbol
+#define elf_backend_finish_dynamic_sections \
+					elf64_hppa_finish_dynamic_sections
+
+/* Stuff for the BFD linker: */
+#define bfd_elf64_bfd_link_hash_table_create \
+	elf64_hppa_hash_table_create
+
+#define elf_backend_check_relocs \
+	elf64_hppa_check_relocs
+
+#define elf_backend_size_info \
+  hppa64_elf_size_info
+
+#define elf_backend_additional_program_headers \
+	elf64_hppa_additional_program_headers
+
+#define elf_backend_modify_segment_map \
+	elf64_hppa_modify_segment_map
+
+#define elf_backend_link_output_symbol_hook \
+	elf64_hppa_link_output_symbol_hook
+
+
+#define elf_backend_want_got_plt	0
+#define elf_backend_plt_readonly	0
+#define elf_backend_want_plt_sym	0
+#define elf_backend_got_header_size     0
+#define elf_backend_plt_header_size     0
+#define elf_backend_type_change_ok true
+
+#include "elf64-target.h"