Update.

2001-06-07  David Mosberger  <davidm@hpl.hp.com>

	* sysdeps/ia64/dl-symaddr.c (_dl_symbol_address): Update call to
	__ia64_make_fptr() for new interface.
	(_dl_function_address): Remove.

	* sysdeps/ia64/dl-machine.h (IA64_BOOT_FPTR_TABLE_LEN): Rename
	from IA64_BOOT_FPTR_SIZE and reduce its value from 256 to 200.
	(struct ia64_fdesc): Rename from struct ia64_fptr and drop "next"
	pointer.  Rename member "func" to "ip".
	(struct ia64_fdesc_table): New type.
	(__ia64_make_fptr): Change prototype.
	(__ia64_init_bootstrap_fdesc_table): New function.
	(ELF_MACHINE_BEFORE_RTLD_RELOC): New macro.
	(__boot_ldso_fptr): Removed.
	(__fptr_next): Removed.
	(__fptr_root): Removed.
	(__fptr_count): Removed.
	(TRAMPOLINE_TEMPLATE): Fix typo in comment.  Don't use multi-line
	strings to avoid warning from gcc3.
	(RTLD_START): Don't use multi-line strings.
	(ELF_MACHINE_START_ADDRESS): Use DL_STATIC_FUNCTION_ADDRESS.
	(elf_machine_fixup_plt): Update the code entry point in the
	function descriptor through a "volatile" pointer to ensure proper
	write ordering.
	(elf_machine_rela): Simplify handling of FPTR relocs: always call
	__ia64_make_fptr().

	* sysdeps/ia64/dl-lookupcfg.h (DL_AUTO_FUNCTION_ADDRESS): New macro.
	(DL_STATIC_FUNCTION_ADDRESS): Likewise.
	(DL_DT_INIT_ADDRESS): Use DL_AUTO_FUNCTION_ADDRESS.
	(DL_DT_FINI_ADDRESS): Likewise.

	* sysdeps/ia64/dl-fptr.c: Rewrite for better scalability and to
	avoid deadlocks.

	* include/link.h: Add member machine specific "l_mach" member to
	link_map.

	* sysdeps/generic/bits/link.h: New file.
	* sysdeps/ia64/bits/link.h: Likewise.

	* elf/rtld.c (_dl_start): Fix typo in comment.
	(_dl_start_final): Copy new "l_mach" member from temporary to
	final bootstrap map.

10 files changed, 594 insertions, 458 deletions

diff --git a/ChangeLog b/ChangeLog
index 65ebdf4460..fd0d20599e 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,49 @@
+2001-06-07  David Mosberger  <davidm@hpl.hp.com>
+
+	* sysdeps/ia64/dl-symaddr.c (_dl_symbol_address): Update call to
+	__ia64_make_fptr() for new interface.
+	(_dl_function_address): Remove.
+
+	* sysdeps/ia64/dl-machine.h (IA64_BOOT_FPTR_TABLE_LEN): Rename
+	from IA64_BOOT_FPTR_SIZE and reduce its value from 256 to 200.
+	(struct ia64_fdesc): Rename from struct ia64_fptr and drop "next"
+	pointer.  Rename member "func" to "ip".
+	(struct ia64_fdesc_table): New type.
+	(__ia64_make_fptr): Change prototype.
+	(__ia64_init_bootstrap_fdesc_table): New function.
+	(ELF_MACHINE_BEFORE_RTLD_RELOC): New macro.
+	(__boot_ldso_fptr): Removed.
+	(__fptr_next): Removed.
+	(__fptr_root): Removed.
+	(__fptr_count): Removed.
+	(TRAMPOLINE_TEMPLATE): Fix typo in comment.  Don't use multi-line
+	strings to avoid warning from gcc3.
+	(RTLD_START): Don't use multi-line strings.
+	(ELF_MACHINE_START_ADDRESS): Use DL_STATIC_FUNCTION_ADDRESS.
+	(elf_machine_fixup_plt): Update the code entry point in the
+	function descriptor through a "volatile" pointer to ensure proper
+	write ordering.
+	(elf_machine_rela): Simplify handling of FPTR relocs: always call
+	__ia64_make_fptr().
+
+	* sysdeps/ia64/dl-lookupcfg.h (DL_AUTO_FUNCTION_ADDRESS): New macro.
+	(DL_STATIC_FUNCTION_ADDRESS): Likewise.
+	(DL_DT_INIT_ADDRESS): Use DL_AUTO_FUNCTION_ADDRESS.
+	(DL_DT_FINI_ADDRESS): Likewise.
+
+	* sysdeps/ia64/dl-fptr.c: Rewrite for better scalability and to
+	avoid deadlocks.
+
+	* include/link.h: Add member machine specific "l_mach" member to
+	link_map.
+
+	* sysdeps/generic/bits/link.h: New file.
+	* sysdeps/ia64/bits/link.h: Likewise.
+
+	* elf/rtld.c (_dl_start): Fix typo in comment.
+	(_dl_start_final): Copy new "l_mach" member from temporary to
+	final bootstrap map.
+
 2001-07-25  Ulrich Drepper  <drepper@redhat.com>
 
 	* csu/gmon-start.c: Use ENTRY_POINT_DECL is defined instead of a
diff --git a/bits/link.h b/bits/link.h
new file mode 100644
index 0000000000..470b4d3e5f
--- /dev/null
+++ b/bits/link.h
@@ -0,0 +1,4 @@
+struct link_map_machine
+  {
+    /* empty by default */
+  };
diff --git a/elf/rtld.c b/elf/rtld.c
index 45b29c679d..36dbc6f94c 100644
--- a/elf/rtld.c
+++ b/elf/rtld.c
@@ -157,7 +157,7 @@ _dl_start (void *arg)
     HP_TIMING_NOW (start_time);
 
   /* Partly clean the `bootstrap_map' structure up.  Don't use `memset'
-     since it might nor be built in or inlined and we cannot make function
+     since it might not be built in or inlined and we cannot make function
      calls at this point.  */
   for (cnt = 0;
        cnt < sizeof (bootstrap_map.l_info) / sizeof (bootstrap_map.l_info[0]);
@@ -228,6 +228,7 @@ _dl_start_final (void *arg, struct link_map *bootstrap_map_p,
   memcpy (_dl_rtld_map.l_info, bootstrap_map_p->l_info,
 	  sizeof _dl_rtld_map.l_info);
   _dl_setup_hash (&_dl_rtld_map);
+  _dl_rtld_map.l_mach = bootstrap_map_p->l_mach;
 
 /* Don't bother trying to work out how ld.so is mapped in memory.  */
   _dl_rtld_map.l_map_start = ~0;
diff --git a/include/link.h b/include/link.h
index 1964b6bad3..9345c928d7 100644
--- a/include/link.h
+++ b/include/link.h
@@ -33,6 +33,7 @@
 #define _ElfW_1(e,w,t)	e##w##t
 
 #include <bits/elfclass.h>		/* Defines __ELF_NATIVE_CLASS.  */
+#include <bits/link.h>
 
 /* Rendezvous structure used by the run-time dynamic linker to communicate
    details of shared object loading to the debugger.  If the executable's
@@ -228,6 +229,8 @@ struct link_map
 
     /* Temporarily used in `dl_close'.  */
     unsigned int l_idx;
+
+    struct link_map_machine l_mach;
   };
 
 struct dl_phdr_info
diff --git a/sysdeps/generic/bits/link.h b/sysdeps/generic/bits/link.h
new file mode 100644
index 0000000000..470b4d3e5f
--- /dev/null
+++ b/sysdeps/generic/bits/link.h
@@ -0,0 +1,4 @@
+struct link_map_machine
+  {
+    /* empty by default */
+  };
diff --git a/sysdeps/ia64/bits/link.h b/sysdeps/ia64/bits/link.h
new file mode 100644
index 0000000000..7f8b0550d9
--- /dev/null
+++ b/sysdeps/ia64/bits/link.h
@@ -0,0 +1,5 @@
+struct link_map_machine
+  {
+    size_t fptr_table_len;
+    Elf64_Addr *fptr_table;
+  };
diff --git a/sysdeps/ia64/dl-fptr.c b/sysdeps/ia64/dl-fptr.c
index 588bb9d599..c31de86079 100644
--- a/sysdeps/ia64/dl-fptr.c
+++ b/sysdeps/ia64/dl-fptr.c
@@ -1,5 +1,5 @@
-/* Unmap a loaded object.  IA-64 version.
-   Copyright (C) 1999, 2000 Free Software Foundation, Inc.
+/* Manage function descriptors.  IA-64 version.
+   Copyright (C) 1999, 2000, 2001 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -17,6 +17,7 @@
    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
    02111-1307 USA.  */
 
+#include <ia64intrin.h>
 #include <unistd.h>
 #include <string.h>
 #include <sys/param.h>
@@ -27,186 +28,260 @@
 #include <dl-machine.h>
 #ifdef _LIBC_REENTRANT
 # include <pt-machine.h>
-
-static int __ia64_fptr_lock = 0;
+# include <signal.h>
+# include <time.h>
 #endif
 
-/* Because ld.so is now versioned, these functions can be in their own
-   file; no relocations need to be done to call them.  Of course, if
-   ld.so is not versioned...  */
-#if 0
-#ifndef DO_VERSIONING
-# error "This will not work with versioning turned off, sorry."
-#endif
+Elf64_Addr __ia64_boot_fptr_table[IA64_BOOT_FPTR_TABLE_LEN];
+
+static struct local
+  {
+    struct ia64_fdesc_table *root;
+    struct ia64_fdesc *free_list;
+    unsigned int npages;		/* # of pages to allocate */
+#ifdef _LIBC_REENTRANT
+    volatile int lock;
+    sigset_t full_sigset;
 #endif
+    /* the next to members MUST be consecutive! */
+    struct ia64_fdesc_table boot_table;
+    struct ia64_fdesc boot_fdescs[1024];
+  }
+local =
+  {
+    root: &local.boot_table,
+    npages: 2,
+    boot_table:
+      {
+	len: sizeof (local.boot_fdescs) / sizeof (local.boot_fdescs[0]),
+	first_unused: 0
+      }
+  };
 
-#ifdef MAP_ANON
-/* The fd is not examined when using MAP_ANON.  */
-#define ANONFD -1
+/* Locking is tricky: we may get a signal while holding the lock and
+   the signal handler may end up calling into the dynamic loader
+   again.  Also, if a real-time process spins on the lock, a
+   non-realtime process may never get the chance to release it's lock,
+   unless the realtime process relinquishes the CPU from time to time.
+   Hence we (a) block signals before acquiring the lock and (b) do a
+   nanosleep() when we detect prolongued contention.  */
+#ifdef _LIBC_REENTRANT
+# define lock(l)						\
+{								\
+  sigset_t _saved_set;						\
+  int i = 10000;						\
+  if (!__sigismember (&(l)->full_sigset, SIGINT))		\
+    __sigfillset (&(l)->full_sigset);				\
+								\
+  while (testandset ((int *) &(l)->lock))			\
+    {								\
+      struct timespec ts;					\
+      if (i > 0)						\
+	{							\
+	  --i;							\
+	  continue;						\
+	}							\
+      ts.tv_sec = 0;						\
+      ts.tv_nsec = 1*1000*1000;					\
+      __nanosleep (&ts, NULL);					\
+    }								\
+  __sigprocmask (SIG_BLOCK, &(l)->full_sigset, &_saved_set);
+# define unlock(l)						\
+  __sigprocmask (SIG_SETMASK, &_saved_set, NULL);		\
+  (l)->lock = 0;						\
+}
 #else
-extern int _dl_zerofd;
-#define ANONFD _dl_zerofd
+# define lock(l)
+# define unlock(l)
 #endif
 
+/* Create a new fdesc table and return a pointer to the first fdesc
+   entry.  The fdesc lock must have been acquired already.  */
 
-/* ld.so currently has 14 FPTR relocs, we take 256 and use them for
-   the relocs for the dynamic app itself.  */
-struct ia64_fptr __boot_ldso_fptr[IA64_BOOT_FPTR_SIZE];
-struct ia64_fptr *__fptr_root = NULL;
-struct ia64_fptr *__fptr_next = __boot_ldso_fptr;
-static struct ia64_fptr *__fptr_free = NULL;
-int __fptr_count = IA64_BOOT_FPTR_SIZE;
-
-Elf64_Addr
-__ia64_make_fptr (const struct link_map *sym_map, Elf64_Addr value,
-		  struct ia64_fptr **root, struct ia64_fptr *mem)
+static struct ia64_fdesc *
+new_fdesc_table (struct local *l)
 {
-  struct ia64_fptr **loc;
-  struct ia64_fptr *f;
+  size_t size = l->npages * _dl_pagesize;
+  struct ia64_fdesc_table *new_table;
+  struct ia64_fdesc *fdesc;
 
-#ifdef _LIBC_REENTRANT
-  /* Make sure we are alone. We don't need a lock during bootstrap. */
-  if (mem == NULL)
-    while (testandset (&__ia64_fptr_lock));
-#endif
+  l->npages += l->npages;
+  new_table = __mmap (0, size, PROT_READ | PROT_WRITE,
+		      MAP_ANON | MAP_PRIVATE, -1, 0);
+  if (new_table == MAP_FAILED)
+    _dl_signal_error (errno, NULL, "cannot map pages for fdesc table");
 
-  /* Search the sorted linked list for an existing entry for this
-     symbol.  */
-  loc = root;
-  f = *loc;
-  while (f != NULL && f->func <= value)
-    {
-      if (f->func == value)
-	goto found;
-      loc = &f->next;
-      f = *loc;
-    }
+  new_table->len = (size - sizeof (*new_table)) / sizeof (struct ia64_fdesc);
+  fdesc = &new_table->fdesc[0];
+  new_table->first_unused = 1;
+  new_table->next = l->root;
+  l->root = new_table;
+  return fdesc;
+}
 
-  /* Not found.  Create a new one.  */
-  if (mem != NULL)
-    f = mem;
-  else if (__fptr_free != NULL)
-    {
-      f = __fptr_free;
-      __fptr_free = f->next;
-    }
-  else
+static Elf64_Addr
+make_fdesc (Elf64_Addr ip, Elf64_Addr gp)
+{
+  struct ia64_fdesc *fdesc = NULL;
+  struct ia64_fdesc_table *t;
+  unsigned int old;
+  struct local *l;
+
+  asm ("addl %0 = @gprel (local), gp" : "=r" (l));
+
+  t = l->root;
+  while (1)
     {
-      if (__fptr_count == 0)
+      old = t->first_unused;
+      if (old >= t->len)
+	break;
+      else if (__sync_bool_compare_and_swap (&t->first_unused, old, old + 1))
 	{
-#ifndef MAP_ANON
-# define MAP_ANON 0
-	  if (_dl_zerofd == -1)
-	    {
-	      _dl_zerofd = _dl_sysdep_open_zero_fill ();
-	      if (_dl_zerofd == -1)
-		{
-		  __close (fd);
-		  _dl_signal_error (errno, NULL,
-				    "cannot open zero fill device");
-		}
-	    }
-#endif
-
-	  __fptr_next = __mmap (0, _dl_pagesize, PROT_READ | PROT_WRITE,
-				MAP_ANON | MAP_PRIVATE, ANONFD, 0);
-	  if (__fptr_next == MAP_FAILED)
-	    _dl_signal_error(errno, NULL, "cannot map page for fptr");
-	  __fptr_count = _dl_pagesize / sizeof (struct ia64_fptr);
+	  fdesc = &t->fdesc[old];
+	  goto install;
 	}
-      f = __fptr_next++;
-      __fptr_count--;
     }
 
-  f->func = value;
-  /* GOT has already been relocated in elf_get_dynamic_info - don't
-     try to relocate it again.  */
-  f->gp = sym_map->l_info[DT_PLTGOT]->d_un.d_ptr;
-  f->next = *loc;
-  *loc = f;
+  lock (l);
+  {
+    if (l->free_list)
+      {
+	fdesc = l->free_list;		/* get it from free-list */
+	l->free_list = (struct ia64_fdesc *) fdesc->ip;
+      }
+    else
+      fdesc = new_fdesc_table (l);	/* create new fdesc table */
+  }
+  unlock (l);
 
-found:
-#ifdef _LIBC_REENTRANT
-  /* Release the lock.  */
-  if (mem == NULL)
-    __ia64_fptr_lock = 0;
+ install:
+  fdesc->ip = ip;
+  fdesc->gp = gp;
+
+  return (Elf64_Addr) fdesc;
+}
+
+static inline Elf64_Addr *
+make_fptr_table (struct link_map *map)
+{
+  const Elf64_Sym *symtab = (const void *) D_PTR (map, l_info[DT_SYMTAB]);
+  const char *strtab = (const void *) D_PTR (map, l_info[DT_STRTAB]);
+  Elf64_Addr *fptr_table;
+  size_t size;
+  size_t len;
+
+  /* XXX Apparently the only way to find out the size of the dynamic
+     symbol section is to assume that the string table follows right
+     afterwards...  */
+  len = ((strtab - (char *) symtab) / map->l_info[DT_SYMENT]->d_un.d_val);
+  size = ((len * sizeof (fptr_table[0]) + _dl_pagesize - 1) & -_dl_pagesize);
+  /* XXX We don't support here in the moment systems without MAP_ANON.
+     There probably are none for IA-64.  In case this is proven wrong
+     we will have to open /dev/null here and use the file descriptor
+     instead of the hard-coded -1.  */
+  fptr_table = __mmap (NULL, size, PROT_READ | PROT_WRITE,
+		       MAP_ANON | MAP_PRIVATE, -1, 0);
+  if (fptr_table == MAP_FAILED)
+    _dl_signal_error (errno, NULL, "cannot map pages for fptr table");
+
+  map->l_mach.fptr_table_len = len;
+  map->l_mach.fptr_table = fptr_table;
+  return fptr_table;
+}
+
+Elf64_Addr
+__ia64_make_fptr (struct link_map *map, const Elf64_Sym *sym, Elf64_Addr ip)
+{
+  Elf64_Addr *ftab = map->l_mach.fptr_table;
+  const Elf64_Sym *symtab;
+  Elf_Symndx symidx;
+
+  if (__builtin_expect (!map->l_mach.fptr_table, 0))
+    ftab = make_fptr_table (map);
+
+  symtab = (const void *) D_PTR (map, l_info[DT_SYMTAB]);
+  symidx = sym - symtab;
+
+  if (symidx >= map->l_mach.fptr_table_len)
+    _dl_signal_error (0, NULL,
+		      "internal error: symidx out of range of fptr table");
+
+  if (!ftab[symidx])
+    {
+      /* GOT has already been relocated in elf_get_dynamic_info -
+	 don't try to relocate it again.  */
+      ftab[symidx] = make_fdesc (ip, map->l_info[DT_PLTGOT]->d_un.d_ptr);
+#if 0
+      {
+	const char *strtab = (const void *) D_PTR (map, l_info[DT_STRTAB]);
+	struct local *l;
+
+	asm ("addl %0 = @gprel (local), gp" : "=r" (l));
+	if (l->root != &l->boot_table || l->boot_table.first_unused > 20)
+	  _dl_debug_printf ("created fdesc symbol `%s' at %lx\n",
+			    strtab + sym->st_name, ftab[symidx]);
+      }
 #endif
+    }
 
-  return (Elf64_Addr) f;
+  return ftab[symidx];
 }
 
 void
 _dl_unmap (struct link_map *map)
 {
-  struct ia64_fptr **floc;
-  struct ia64_fptr *f;
-  struct ia64_fptr **lloc;
-  struct ia64_fptr *l;
+  Elf64_Addr *ftab = map->l_mach.fptr_table;
+  struct ia64_fdesc *head = NULL, *tail = NULL;
+  size_t i;
 
   __munmap ((void *) map->l_map_start, map->l_map_end - map->l_map_start);
 
-#ifdef _LIBC_REENTRANT
-  /* Make sure we are alone.  */
-  while (testandset (&__ia64_fptr_lock));
-#endif
+  if (!ftab)
+    return;
 
-  /* Search the sorted linked list for the first entry for this object.  */
-  floc = &__fptr_root;
-  f = *floc;
-  while (f != NULL && f->func < map->l_map_start)
+  /* String together the fdesc structures that are being freed.  */
+  for (i = 0; i < map->l_mach.fptr_table_len; ++i)
     {
-      floc = &f->next;
-      f = *floc;
-    }
-
-  /* We found one.  */
-  if (f != NULL && f->func < map->l_map_end)
-    {
-      /* Get the last entry.  */
-      lloc = floc;
-      l = f;
-      while (l && l->func < map->l_map_end)
+      if (ftab[i])
 	{
-	  lloc = &l->next;
-	  l = *lloc;
+	  *(struct ia64_fdesc **) ftab[i] = head;
+	  head = (struct ia64_fdesc *) ftab[i];
+	  if (!tail)
+	    tail = head;
 	}
+    }
 
-      /* Updated FPTR.  */
-      *floc = l;
-
-      /* Prepend them to the free list.  */
-      *lloc = __fptr_free;
-      __fptr_free = f;
+  /* Prepend the new list to the free_list: */
+  if (tail)
+    {
+      lock (&local);
+      {
+	*(struct ia64_fdesc **) tail = local.free_list;
+	local.free_list = head;
+      }
+      unlock (&local);
     }
 
-#ifdef _LIBC_REENTRANT
-  /* Release the lock.   */
-  __ia64_fptr_lock = 0;
-#endif
+  __munmap (ftab,
+	    map->l_mach.fptr_table_len * sizeof (map->l_mach.fptr_table[0]));
+  map->l_mach.fptr_table = NULL;
 }
 
 Elf64_Addr
 _dl_lookup_address (const void *address)
 {
   Elf64_Addr addr = (Elf64_Addr) address;
-  struct ia64_fptr *f;
-
-#ifdef _LIBC_REENTRANT
-  /* Make sure we are alone.  */
-  while (testandset (&__ia64_fptr_lock));
-#endif
-
-  for (f = __fptr_root; f != NULL; f = f->next)
-    if (f == address)
-      {
-	addr = f->func;
-	break;
-      }
-
-#ifdef _LIBC_REENTRANT
-  /* Release the lock.   */
-  __ia64_fptr_lock = 0;
-#endif
+  struct ia64_fdesc_table *t;
+  unsigned long int i;
 
+  for (t = local.root; t != NULL; t = t->next)
+    {
+      i = (struct ia64_fdesc *) addr - &t->fdesc[0];
+      if (i < t->first_unused && addr == (Elf64_Addr) &t->fdesc[i])
+	{
+	  addr = t->fdesc[i].ip;
+	  break;
+	}
+    }
   return addr;
 }
diff --git a/sysdeps/ia64/dl-lookupcfg.h b/sysdeps/ia64/dl-lookupcfg.h
index 884968e895..2f5ba8ed6c 100644
--- a/sysdeps/ia64/dl-lookupcfg.h
+++ b/sysdeps/ia64/dl-lookupcfg.h
@@ -36,9 +36,21 @@ extern void _dl_unmap (struct link_map *map);
 
 #define DL_UNMAP(map) _dl_unmap (map)
 
-extern Elf64_Addr _dl_function_address (const struct link_map *map,
-					Elf64_Addr start);
-
-#define DL_FUNCTION_ADDRESS(map, addr) _dl_function_address (map, addr)
-#define DL_DT_INIT_ADDRESS(map, addr) DL_FUNCTION_ADDRESS (map, addr)
-#define DL_DT_FINI_ADDRESS(map, addr) DL_FUNCTION_ADDRESS (map, addr)
+#define DL_AUTO_FUNCTION_ADDRESS(map, addr)		\
+({							\
+  unsigned long fptr[2];				\
+  fptr[0] = (addr);					\
+  fptr[1] = (map)->l_info[DT_PLTGOT]->d_un.d_ptr;	\
+  (Elf64_Addr) fptr;					\
+})
+
+#define DL_STATIC_FUNCTION_ADDRESS(map, addr)		\
+({							\
+  static unsigned long fptr[2];				\
+  fptr[0] = (addr);					\
+  fptr[1] = (map)->l_info[DT_PLTGOT]->d_un.d_ptr;	\
+  (Elf64_Addr) fptr;					\
+})
+
+#define DL_DT_INIT_ADDRESS(map, addr) DL_AUTO_FUNCTION_ADDRESS (map, addr)
+#define DL_DT_FINI_ADDRESS(map, addr) DL_AUTO_FUNCTION_ADDRESS (map, addr)
diff --git a/sysdeps/ia64/dl-machine.h b/sysdeps/ia64/dl-machine.h
index 9ad777e99f..a2192a1e2e 100644
--- a/sysdeps/ia64/dl-machine.h
+++ b/sysdeps/ia64/dl-machine.h
@@ -1,5 +1,5 @@
 /* Machine-dependent ELF dynamic relocation inline functions.  IA-64 version.
-   Copyright (C) 1995, 1996, 1997, 2000 Free Software Foundation, Inc.
+   Copyright (C) 1995, 1996, 1997, 2000, 2001 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -32,26 +32,43 @@
    in l_info array.  */
 #define DT_IA_64(x) (DT_IA_64_##x - DT_LOPROC + DT_NUM)
 
+/* There are currently 123 dynamic symbols in ld.so.
+   IA64_BOOT_FPTR_TABLE_LEN needs to be at least that big.  */
+#define IA64_BOOT_FPTR_TABLE_LEN	200
 
-/* An FPTR is a function descriptor.  Properly they consist of just
-   FUNC and GP.  But we want to traverse a binary tree too.  */
+/* An FDESC is a function descriptor.  */
 
-#define IA64_BOOT_FPTR_SIZE	256
+struct ia64_fdesc
+  {
+    Elf64_Addr ip;	/* code entry point */
+    Elf64_Addr gp;	/* global pointer */
+  };
 
-struct ia64_fptr
+struct ia64_fdesc_table
+  {
+    struct ia64_fdesc_table *next;
+    unsigned int len;			/* # of entries in fdesc table */
+    volatile unsigned int first_unused;	/* index of first available entry */
+    struct ia64_fdesc fdesc[0];
+  };
+
+extern Elf64_Addr __ia64_make_fptr (struct link_map *, const Elf64_Sym *,
+				    Elf64_Addr);
+
+static inline void
+__ia64_init_bootstrap_fdesc_table (struct link_map *map)
 {
-  Elf64_Addr func;
-  Elf64_Addr gp;
-  struct ia64_fptr *next;
-};
+  Elf64_Addr *boot_table;
 
-extern struct ia64_fptr __boot_ldso_fptr[];
-extern struct ia64_fptr *__fptr_next;
-extern struct ia64_fptr *__fptr_root;
-extern int __fptr_count;
+  /* careful: this will be called before got has been relocated... */
+  asm ("addl %0 = @gprel (__ia64_boot_fptr_table), gp" : "=r"(boot_table));
 
-extern Elf64_Addr __ia64_make_fptr (const struct link_map *, Elf64_Addr,
-				    struct ia64_fptr **, struct ia64_fptr *);
+  map->l_mach.fptr_table_len = IA64_BOOT_FPTR_TABLE_LEN;
+  map->l_mach.fptr_table = boot_table;
+}
+
+#define ELF_MACHINE_BEFORE_RTLD_RELOC(dynamic_info)		\
+	__ia64_init_bootstrap_fdesc_table (&bootstrap_map);
 
 /* Return nonzero iff ELF header is compatible with the running host.  */
 static inline int
@@ -67,7 +84,7 @@ elf_machine_dynamic (void)
 {
   Elf64_Addr *p;
 
-  __asm__(
+  __asm__ (
 	".section .sdata\n"
 	"	.type __dynamic_ltv#, @object\n"
 	"	.size __dynamic_ltv#, 8\n"
@@ -75,7 +92,7 @@ elf_machine_dynamic (void)
 	"	data8	@ltv(_DYNAMIC#)\n"
 	".previous\n"
 	"	addl	%0 = @gprel(__dynamic_ltv#), gp ;;"
-	: "=r"(p));
+	: "=r" (p));
 
   return *p;
 }
@@ -88,19 +105,18 @@ elf_machine_load_address (void)
   Elf64_Addr ip;
   int *p;
 
-  __asm__(
+  __asm__ (
 	"1:	mov %0 = ip\n"
 	".section .sdata\n"
 	"2:	data4	@ltv(1b)\n"
 	"       .align 8\n"
 	".previous\n"
 	"	addl	%1 = @gprel(2b), gp ;;"
-	: "=r"(ip), "=r"(p));
+	: "=r" (ip), "=r" (p));
 
-  return ip - (Elf64_Addr)*p;
+  return ip - (Elf64_Addr) *p;
 }
 
-
 /* Set up the loaded object described by L so its unrelocated PLT
    entries will jump to the on-demand fixup code in dl-runtime.c.  */
 
@@ -112,21 +128,21 @@ elf_machine_runtime_setup (struct link_map *l, int lazy, int profile)
 
   if (lazy)
     {
-      register Elf64_Addr gp __asm__("gp");
+      register Elf64_Addr gp __asm__ ("gp");
       Elf64_Addr *reserve, doit;
 
       /*
        * Careful with the typecast here or it will try to add l-l_addr
        * pointer elements
        */
-      reserve = (Elf64_Addr *)
-	      (l->l_info[DT_IA_64(PLT_RESERVE)]->d_un.d_ptr + l->l_addr);
+      reserve = ((Elf64_Addr *)
+		 (l->l_info[DT_IA_64 (PLT_RESERVE)]->d_un.d_ptr + l->l_addr));
       /* Identify this shared object.  */
       reserve[0] = (Elf64_Addr) l;
 
       /* This function will be called to perform the relocation.  */
       if (!profile)
-	doit = (Elf64_Addr) ((struct ia64_fptr *)&_dl_runtime_resolve)->func;
+	doit = (Elf64_Addr) ((struct ia64_fdesc *) &_dl_runtime_resolve)->ip;
       else
 	{
 	  if (_dl_name_match_p (_dl_profile, l))
@@ -135,7 +151,7 @@ elf_machine_runtime_setup (struct link_map *l, int lazy, int profile)
 		 want profiling and the timers are started.  */
 	      _dl_profile_map = l;
 	    }
-	  doit = (Elf64_Addr) ((struct ia64_fptr *)&_dl_runtime_profile)->func;
+	  doit = (Elf64_Addr) ((struct ia64_fdesc *) &_dl_runtime_profile)->ip;
 	}
 
       reserve[1] = doit;
@@ -149,111 +165,111 @@ elf_machine_runtime_setup (struct link_map *l, int lazy, int profile)
 /*
    This code is used in dl-runtime.c to call the `fixup' function
    and then redirect to the address it returns. `fixup()' takes two
-   arguments, however fixup_profile() takes three.
+   arguments, however profile_fixup() takes three.
 
    The ABI specifies that we will never see more than 8 input
    registers to a function call, thus it is safe to simply allocate
    those, and simpler than playing stack games.
 					                     - 12/09/99 Jes
  */
-#define TRAMPOLINE_TEMPLATE(tramp_name, fixup_name) \
-  extern void tramp_name (void); \
-  asm ( "\
-	.global " #tramp_name "#
-	.proc " #tramp_name "#
-" #tramp_name ":
-	{ .mmi
-	  alloc loc0 = ar.pfs, 8, 6, 3, 0
-	  adds r2 = -144, r12
-	  adds r3 = -128, r12
-	}
-	{ .mii
-	  adds r12 = -160, r12
-	  mov loc1 = b0
-	  mov out2 = b0		/* needed by fixup_profile */
-	  ;;
-	}
-	{ .mfb
-	  mov loc2 = r8		/* preserve struct value register */
-	  nop.f 0
-	  nop.b 0
-	}
-	{ .mii
-	  mov loc3 = r9		/* preserve language specific register */
-	  mov loc4 = r10	/* preserve language specific register */
-	  mov loc5 = r11	/* preserve language specific register */
-	}
-	{ .mmi
-	  stf.spill [r2] = f8, 32
-	  stf.spill [r3] = f9, 32
-	  mov out0 = r16
-	  ;;
-	}
-	{ .mmi
-	  stf.spill [r2] = f10, 32
-	  stf.spill [r3] = f11, 32
-	  shl out1 = r15, 4
-	  ;;
-	}
-	{ .mmi
-	  stf.spill [r2] = f12, 32
-	  stf.spill [r3] = f13, 32
-	  shladd out1 = r15, 3, out1
-	  ;;
-	}
-	{ .mmb
-	  stf.spill [r2] = f14
-	  stf.spill [r3] = f15
-	  br.call.sptk.many b0 = " #fixup_name "#
-	}
-	{ .mii
-	  ld8 r9 = [ret0], 8
-	  adds r2 = 16, r12
-	  adds r3 = 32, r12
-	  ;;
-	}
-	{ .mmi
-	  ldf.fill f8 = [r2], 32
-	  ldf.fill f9 = [r3], 32
-	  mov b0 = loc1
-	  ;;
-	}
-	{ .mmi
-	  ldf.fill f10 = [r2], 32
-	  ldf.fill f11 = [r3], 32
-	  mov b6 = r9
-	  ;;
-	}
-	{ .mmi
-	  ldf.fill f12 = [r2], 32
-	  ldf.fill f13 = [r3], 32
-	  mov ar.pfs = loc0
-	  ;;
-	}
-	{ .mmi
-	  ldf.fill f14 = [r2], 32
-	  ldf.fill f15 = [r3], 32
-	  adds r12 = 160, r12
-	  ;;
-	}
-	{ .mii
-	  mov r9 = loc3		/* restore language specific register */
-	  mov r10 = loc4	/* restore language specific register */
-	  mov r11 = loc5	/* restore language specific register */
-	}
-	{ .mii
-	  ld8 gp = [ret0]
-	  mov r8 = loc2		/* restore struct value register */
-	  ;;
-	}
-	/* An alloc is needed for the break system call to work.
-	   We don't care about the old value of the pfs register.  */
-	{ .mmb
-	  alloc r2 = ar.pfs, 0, 0, 8, 0
-	  br.sptk.many b6
-	  ;;
-	}
-	.endp " #tramp_name "#")
+#define TRAMPOLINE_TEMPLATE(tramp_name, fixup_name)			     \
+  extern void tramp_name (void);					     \
+  asm (									     \
+"	.global " #tramp_name "#\n"					     \
+"	.proc " #tramp_name "#\n"					     \
+#tramp_name ":\n"							     \
+"	{ .mmi\n"							     \
+"	  alloc loc0 = ar.pfs, 8, 6, 3, 0\n"				     \
+"	  adds r2 = -144, r12\n"					     \
+"	  adds r3 = -128, r12\n"					     \
+"	}\n"								     \
+"	{ .mii\n"							     \
+"	  adds r12 = -160, r12\n"					     \
+"	  mov loc1 = b0\n"						     \
+"	  mov out2 = b0		/* needed by fixup_profile */\n"	     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	{ .mfb\n"							     \
+"	  mov loc2 = r8		/* preserve struct value register */\n"	     \
+"	  nop.f 0\n"							     \
+"	  nop.b 0\n"							     \
+"	}\n"								     \
+"	{ .mii\n"							     \
+"	  mov loc3 = r9		/* preserve language specific register */\n" \
+"	  mov loc4 = r10	/* preserve language specific register */\n" \
+"	  mov loc5 = r11	/* preserve language specific register */\n" \
+"	}\n"								     \
+"	{ .mmi\n"							     \
+"	  stf.spill [r2] = f8, 32\n"					     \
+"	  stf.spill [r3] = f9, 32\n"					     \
+"	  mov out0 = r16\n"						     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	{ .mmi\n"							     \
+"	  stf.spill [r2] = f10, 32\n"					     \
+"	  stf.spill [r3] = f11, 32\n"					     \
+"	  shl out1 = r15, 4\n"						     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	{ .mmi\n"							     \
+"	  stf.spill [r2] = f12, 32\n"					     \
+"	  stf.spill [r3] = f13, 32\n"					     \
+"	  shladd out1 = r15, 3, out1\n"					     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	{ .mmb\n"							     \
+"	  stf.spill [r2] = f14\n"					     \
+"	  stf.spill [r3] = f15\n"					     \
+"	  br.call.sptk.many b0 = " #fixup_name "#\n"			     \
+"	}\n"								     \
+"	{ .mii\n"							     \
+"	  ld8 r9 = [ret0], 8\n"						     \
+"	  adds r2 = 16, r12\n"						     \
+"	  adds r3 = 32, r12\n"						     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	{ .mmi\n"							     \
+"	  ldf.fill f8 = [r2], 32\n"					     \
+"	  ldf.fill f9 = [r3], 32\n"					     \
+"	  mov b0 = loc1\n"						     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	{ .mmi\n"							     \
+"	  ldf.fill f10 = [r2], 32\n"					     \
+"	  ldf.fill f11 = [r3], 32\n"					     \
+"	  mov b6 = r9\n"						     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	{ .mmi\n"							     \
+"	  ldf.fill f12 = [r2], 32\n"					     \
+"	  ldf.fill f13 = [r3], 32\n"					     \
+"	  mov ar.pfs = loc0\n"						     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	{ .mmi\n"							     \
+"	  ldf.fill f14 = [r2], 32\n"					     \
+"	  ldf.fill f15 = [r3], 32\n"					     \
+"	  adds r12 = 160, r12\n"					     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	{ .mii\n"							     \
+"	  mov r9 = loc3		/* restore language specific register */\n"  \
+"	  mov r10 = loc4	/* restore language specific register */\n"  \
+"	  mov r11 = loc5	/* restore language specific register */\n"  \
+"	}\n"								     \
+"	{ .mii\n"							     \
+"	  ld8 gp = [ret0]\n"						     \
+"	  mov r8 = loc2		/* restore struct value register */\n"	     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	/* An alloc is needed for the break system call to work.\n"	     \
+"	   We don't care about the old value of the pfs register.  */\n"     \
+"	{ .mmb\n"							     \
+"	  alloc r2 = ar.pfs, 0, 0, 8, 0\n"				     \
+"	  br.sptk.many b6\n"						     \
+"	  ;;\n"								     \
+"	}\n"								     \
+"	.endp " #tramp_name "#\n");
 
 #ifndef PROF
 #define ELF_MACHINE_RUNTIME_TRAMPOLINE 				\
@@ -270,148 +286,148 @@ elf_machine_runtime_setup (struct link_map *l, int lazy, int profile)
    The C function `_dl_start' is the real entry point;
    its return value is the user program's entry point.  */
 
-#define RTLD_START asm ("\
-.text
-	.global _start#
-	.proc _start#
-_start:
-0:	{ .mii
-	  alloc loc0 = ar.pfs, 0, 3, 4, 0
-	  mov r2 = ip
-	  addl r3 = @gprel(0b), r0
-	  ;;
-	}
-	{ .mlx
-	  /* Calculate the GP, and save a copy in loc1.  */
-	  sub gp = r2, r3
-	  movl r8 = 0x9804c0270033f
-	  ;;
-	}
-	{ .mii
-	  mov ar.fpsr = r8
-	  sub loc1 = r2, r3
-	  /* _dl_start wants a pointer to the pointer to the arg block
-	     and the arg block starts with an integer, thus the magic 16.  */
-	  adds out0 = 16, sp
-	}
-	{ .bbb
-	  br.call.sptk.many b0 = _dl_start#
-	  ;;
-	}
-	.endp _start#
-	/* FALLTHRU */
-	.global _dl_start_user#
-	.proc _dl_start_user#
-_dl_start_user:
-	{ .mii
-	  /* Save the pointer to the user entry point fptr in loc2.  */
-	  mov loc2 = ret0
-	  /* Store the highest stack address.  */
-	  addl r2 = @ltoff(__libc_stack_end#), gp
-	  addl r3 = @gprel(_dl_skip_args), gp
-	  ;;
-	}
-	{ .mmi
-	  ld8 r2 = [r2]
-	  ld4 r3 = [r3]
-	  adds r11 = 24, sp	/* Load the address of argv. */
-	  ;;
-	}
-	{ .mii
-	  st8 [r2] = sp
-	  adds r10 = 16, sp	/* Load the address of argc. */
-	  mov out2 = r11
-	  ;;
-	  /* See if we were run as a command with the executable file
-	     name as an extra leading argument.  If so, adjust the argv
-	     pointer to skip _dl_skip_args words.
-	     Note that _dl_skip_args is an integer, not a long - Jes
-
-	     The stack pointer has to be 16 byte aligned. We cannot simply
-	     addjust the stack pointer. We have to move the whole argv and
-	     envp and adjust _dl_argv by _dl_skip_args.  H.J.  */
-	}
-	{ .mib
-	  ld8 out1 = [r10]	/* is argc actually stored as a long
-				   or as an int? */
-	  addl r2 = @ltoff(_dl_argv), gp
-	  ;;
-	}
-	{ .mmi
-	  ld8 r2 = [r2]		/* Get the address of _dl_argv. */
-	  sub out1 = out1, r3	/* Get the new argc. */
-	  shladd r3 = r3, 3, r0
-	  ;;
-	}
-	{
-	  .mib
-	  ld8 r17 = [r2]	/* Get _dl_argv. */
-	  add r15 = r11, r3	/* The address of the argv we move */
-	  ;;
-	}
-	/* ??? Could probably merge these two loops into 3 bundles.
-	   using predication to control which set of copies we're on.  */
-1:	/* Copy argv. */
-	{ .mfi
-	  ld8 r16 = [r15], 8	/* Load the value in the old argv. */
-	  ;;
-	}
-	{ .mib
-	  st8 [r11] = r16, 8	/* Store it in the new argv. */
-	  cmp.ne p6, p7 = 0, r16
-(p6)	  br.cond.dptk.few 1b
-	  ;;
-	}
-	{ .mmi
-	  mov out3 = r11
-	  sub r17 = r17, r3	/* Substract _dl_skip_args. */
-	  addl out0 = @ltoff(_dl_loaded), gp
-	}
-1:	/* Copy env. */
-	{ .mfi
-	  ld8 r16 = [r15], 8	/* Load the value in the old env. */
-	  ;;
-	}
-	{ .mib
-	  st8 [r11] = r16, 8	/* Store it in the new env. */
-	  cmp.ne p6, p7 = 0, r16
-(p6)	  br.cond.dptk.few 1b
-	  ;;
-	}
-	{ .mmb
-	  st8 [r10] = out1		/* Record the new argc. */
-	  ld8 out0 = [out0]
-	  ;;
-	}
-	{ .mmb
-	  ld8 out0 = [out0]		/* get the linkmap */
-	  st8 [r2] = r17		/* Load the new _dl_argv. */
-	  br.call.sptk.many b0 = _dl_init#
-	  ;;
-	}
-	/* Pass our finializer function to the user,
-	   and jump to the user's entry point.  */
-	{ .mmi
-	  ld8 r3 = [loc2], 8
-	  mov b0 = r0
-	}
-	{ .mmi
-	  addl ret0 = @ltoff(@fptr(_dl_fini#)), gp
-	  ;;
-	  mov b6 = r3
-	}
-	{ .mmi
-	  ld8 ret0 = [ret0]
-	  ld8 gp = [loc2]
-	  mov ar.pfs = loc0
-	  ;;
-	}
-	{ .mfb
-	  br.sptk.many b6
-	  ;;
-	}
-	.endp _dl_start_user#
-.previous");
+#define RTLD_START asm (						      \
+".text\n"								      \
+"	.global _start#\n"						      \
+"	.proc _start#\n"						      \
+"_start:\n"								      \
+"0:	{ .mii\n"							      \
+"	  alloc loc0 = ar.pfs, 0, 3, 4, 0\n"				      \
+"	  mov r2 = ip\n"						      \
+"	  addl r3 = @gprel(0b), r0\n"					      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mlx\n"							      \
+"	  /* Calculate the GP, and save a copy in loc1.  */\n"		      \
+"	  sub gp = r2, r3\n"						      \
+"	  movl r8 = 0x9804c0270033f\n"					      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mii\n"							      \
+"	  mov ar.fpsr = r8\n"						      \
+"	  sub loc1 = r2, r3\n"						      \
+"	  /* _dl_start wants a pointer to the pointer to the arg block and\n" \
+"	     the arg block starts with an integer, thus the magic 16. */\n"   \
+"	  adds out0 = 16, sp\n"						      \
+"	}\n"								      \
+"	{ .bbb\n"							      \
+"	  br.call.sptk.many b0 = _dl_start#\n"				      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	.endp _start#\n"						      \
+"	/* FALLTHRU */\n"						      \
+"	.global _dl_start_user#\n"					      \
+"	.proc _dl_start_user#\n"					      \
+"_dl_start_user:\n"							      \
+"	{ .mii\n"							      \
+"	  /* Save the pointer to the user entry point fptr in loc2.  */\n"    \
+"	  mov loc2 = ret0\n"						      \
+"	  /* Store the highest stack address.  */\n"			      \
+"	  addl r2 = @ltoff(__libc_stack_end#), gp\n"			      \
+"	  addl r3 = @gprel(_dl_skip_args), gp\n"			      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mmi\n"							      \
+"	  ld8 r2 = [r2]\n"						      \
+"	  ld4 r3 = [r3]\n"						      \
+"	  adds r11 = 24, sp	/* Load the address of argv. */\n"	      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mii\n"							      \
+"	  st8 [r2] = sp\n"						      \
+"	  adds r10 = 16, sp	/* Load the address of argc. */\n"	      \
+"	  mov out2 = r11\n"						      \
+"	  ;;\n"								      \
+"	  /* See if we were run as a command with the executable file\n"      \
+"	     name as an extra leading argument.  If so, adjust the argv\n"    \
+"	     pointer to skip _dl_skip_args words.\n"			      \
+"	     Note that _dl_skip_args is an integer, not a long - Jes\n"	      \
+"\n"									      \
+"	     The stack pointer has to be 16 byte aligned. We cannot simply\n" \
+"	     addjust the stack pointer. We have to move the whole argv and\n" \
+"	     envp and adjust _dl_argv by _dl_skip_args.  H.J.  */\n"	      \
+"	}\n"								      \
+"	{ .mib\n"							      \
+"	  ld8 out1 = [r10]	/* is argc actually stored as a long\n"	      \
+"				   or as an int? */\n"			      \
+"	  addl r2 = @ltoff(_dl_argv), gp\n"				      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mmi\n"							      \
+"	  ld8 r2 = [r2]		/* Get the address of _dl_argv. */\n"	      \
+"	  sub out1 = out1, r3	/* Get the new argc. */\n"		      \
+"	  shladd r3 = r3, 3, r0\n"					      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{\n"								      \
+"	  .mib\n"							      \
+"	  ld8 r17 = [r2]	/* Get _dl_argv. */\n"			      \
+"	  add r15 = r11, r3	/* The address of the argv we move */\n"      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	/* ??? Could probably merge these two loops into 3 bundles.\n"	      \
+"	   using predication to control which set of copies we're on.  */\n"  \
+"1:	/* Copy argv. */\n"						      \
+"	{ .mfi\n"							      \
+"	  ld8 r16 = [r15], 8	/* Load the value in the old argv. */\n"      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mib\n"							      \
+"	  st8 [r11] = r16, 8	/* Store it in the new argv. */\n"	      \
+"	  cmp.ne p6, p7 = 0, r16\n"					      \
+"(p6)	  br.cond.dptk.few 1b\n"					      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mmi\n"							      \
+"	  mov out3 = r11\n"						      \
+"	  sub r17 = r17, r3	/* Substract _dl_skip_args. */\n"	      \
+"	  addl out0 = @ltoff(_dl_loaded), gp\n"				      \
+"	}\n"								      \
+"1:	/* Copy env. */\n"						      \
+"	{ .mfi\n"							      \
+"	  ld8 r16 = [r15], 8	/* Load the value in the old env. */\n"	      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mib\n"							      \
+"	  st8 [r11] = r16, 8	/* Store it in the new env. */\n"	      \
+"	  cmp.ne p6, p7 = 0, r16\n"					      \
+"(p6)	  br.cond.dptk.few 1b\n"					      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mmb\n"							      \
+"	  st8 [r10] = out1		/* Record the new argc. */\n"	      \
+"	  ld8 out0 = [out0]\n"						      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mmb\n"							      \
+"	  ld8 out0 = [out0]		/* get the linkmap */\n"	      \
+"	  st8 [r2] = r17		/* Load the new _dl_argv. */\n"	      \
+"	  br.call.sptk.many b0 = _dl_init#\n"				      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	/* Pass our finializer function to the user,\n"			      \
+"	   and jump to the user's entry point.  */\n"			      \
+"	{ .mmi\n"							      \
+"	  ld8 r3 = [loc2], 8\n"						      \
+"	  mov b0 = r0\n"						      \
+"	}\n"								      \
+"	{ .mmi\n"							      \
+"	  addl ret0 = @ltoff(@fptr(_dl_fini#)), gp\n"			      \
+"	  ;;\n"								      \
+"	  mov b6 = r3\n"						      \
+"	}\n"								      \
+"	{ .mmi\n"							      \
+"	  ld8 ret0 = [ret0]\n"						      \
+"	  ld8 gp = [loc2]\n"						      \
+"	  mov ar.pfs = loc0\n"						      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	{ .mfb\n"							      \
+"	  br.sptk.many b6\n"						      \
+"	  ;;\n"								      \
+"	}\n"								      \
+"	.endp _dl_start_user#\n"					      \
+".previous\n");
 
 
 #ifndef RTLD_START_SPECIAL_INIT
@@ -434,8 +450,8 @@ _dl_start_user:
 #define ELF_MACHINE_NO_REL 1
 
 /* Return the address of the entry point. */
-#define ELF_MACHINE_START_ADDRESS(map, start) \
-  DL_FUNCTION_ADDRESS (map, start)
+#define ELF_MACHINE_START_ADDRESS(map, start)	\
+  DL_STATIC_FUNCTION_ADDRESS (map, start)
 
 #define elf_machine_profile_fixup_plt(l, reloc, rel_addr, value) \
   elf_machine_fixup_plt (l, reloc, rel_addr, value)
@@ -452,7 +468,9 @@ elf_machine_fixup_plt (struct link_map *l, lookup_t t,
    * being called */
   /* got has already been relocated in elf_get_dynamic_info() */
   reloc_addr[1] = t->l_info[DT_PLTGOT]->d_un.d_ptr;
-  reloc_addr[0] = value;
+  /* we need a "release" here to ensure that the gp is visible before
+     the code entry point is updated: */
+  ((volatile Elf64_Addr *) reloc_addr)[0] = value;
   return (Elf64_Addr) reloc_addr;
 }
 
@@ -487,7 +505,7 @@ elf_machine_rela (struct link_map *map,
 		  const struct r_found_version *version,
 		  Elf64_Addr *const reloc_addr)
 {
-  unsigned long const r_type = ELF64_R_TYPE (reloc->r_info);
+  const unsigned long int r_type = ELF64_R_TYPE (reloc->r_info);
   Elf64_Addr value;
 
 #ifndef RTLD_BOOTSTRAP
@@ -533,35 +551,9 @@ elf_machine_rela (struct link_map *map,
 	      return;
 	    }
 	  else if (R_IA64_TYPE (r_type) == R_IA64_TYPE (R_IA64_FPTR64LSB))
-#ifndef RTLD_BOOTSTRAP
-	    value = __ia64_make_fptr (sym_map, value, &__fptr_root, NULL);
-#else
-	  {
-	    struct ia64_fptr *p_boot_ldso_fptr;
-	    struct ia64_fptr **p_fptr_root;
-	    int *p_fptr_count;
-
-	    /* Special care must be taken to address these variables
-	       during bootstrap.  Further, since we don't know exactly
-	       when __fptr_next will be relocated, we index directly
-	       off __boot_ldso_fptr.  */
-	    asm ("addl %0 = @gprel(__boot_ldso_fptr#), gp\n\t"
-		 "addl %1 = @gprel(__fptr_root#), gp\n\t"
-		 "addl %2 = @gprel(__fptr_count#), gp"
-		 : "=r"(p_boot_ldso_fptr),
-	         "=r"(p_fptr_root),
-	         "=r"(p_fptr_count));
-
-	    /*
-	     * Go from the top - __ia64_make_fptr goes from the bottom,
-	     * this way we will never clash.
-	     */
-	    value = __ia64_make_fptr (sym_map, value, p_fptr_root,
-				      &p_boot_ldso_fptr[--*p_fptr_count]);
-	  }
-#endif
+	    value = __ia64_make_fptr (sym_map, sym, value);
 	  else if (R_IA64_TYPE (r_type) == R_IA64_TYPE (R_IA64_PCREL64LSB))
-	    value -= (Elf64_Addr)reloc_addr & -16;
+	    value -= (Elf64_Addr) reloc_addr & -16;
 	  else
 	    assert (! "unexpected dynamic reloc type");
 	}
@@ -573,7 +565,7 @@ elf_machine_rela (struct link_map *map,
   if (R_IA64_FORMAT (r_type) == R_IA64_FORMAT_64LSB)
     *reloc_addr = value;
   else if (R_IA64_FORMAT (r_type) == R_IA64_FORMAT_32LSB)
-    *(int *)reloc_addr = value;
+    *(int *) reloc_addr = value;
   else if (r_type == R_IA64_IPLTLSB)
     {
       reloc_addr[0] = 0;
@@ -589,8 +581,8 @@ static inline void
 elf_machine_lazy_rel (struct link_map *map,
 		      Elf64_Addr l_addr, const Elf64_Rela *reloc)
 {
-  Elf64_Addr * const reloc_addr = (void *)(l_addr + reloc->r_offset);
-  unsigned long const r_type = ELF64_R_TYPE (reloc->r_info);
+  Elf64_Addr *const reloc_addr = (void *) (l_addr + reloc->r_offset);
+  const unsigned long int r_type = ELF64_R_TYPE (reloc->r_info);
 
   if (r_type == R_IA64_IPLTLSB)
     {
diff --git a/sysdeps/ia64/dl-symaddr.c b/sysdeps/ia64/dl-symaddr.c
index 99b05baa60..52c5a12e14 100644
--- a/sysdeps/ia64/dl-symaddr.c
+++ b/sysdeps/ia64/dl-symaddr.c
@@ -1,5 +1,5 @@
 /* Get the symbol address.  IA-64 version.
-   Copyright (C) 1999, 2000 Free Software Foundation, Inc.
+   Copyright (C) 1999, 2000, 2001 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -27,13 +27,7 @@ _dl_symbol_address (const struct link_map *map, const Elf64_Sym *ref)
 
   /* On ia64, we have to return the pointer to function descriptor. */
   if (ELFW(ST_TYPE) (ref->st_info) == STT_FUNC)
-    return (void *) __ia64_make_fptr (map, value, &__fptr_root, NULL);
+    return (void *) __ia64_make_fptr (map, ref, value);
   else
     return (void *) value;
 }
-
-Elf64_Addr
-_dl_function_address (const struct link_map *map, Elf64_Addr start)
-{
-  return __ia64_make_fptr (map, start, &__fptr_root, NULL);
-}