2007-06-09 Ulrich Drepper <drepper@redhat.com>

	* elf/do-lookup.h (do_lookup_x): Read r_nlist before r_list and
	make sure gcc doesn't mess around with this.

2007-06-08  Ulrich Drepper  <drepper@redhat.com>

	* elf/dl-lookup.c (_dl_lookup_symbol_x): Remove use of r_nlist.

2007-06-08  Jakub Jelinek  <jakub@redhat.com>

	* elf/dl-close.c (_dl_close_worker): Remove all to be removed
	libraries from the global scope at once and call THREAD_GSCOPE_WAIT

2007-05-18  Ulrich Drepper  <drepper@redhat.com>

	* elf/dl-close.c (_dl_close_worker): When removing object from
	global scope, wait for all lookups to finish afterwards.
	* elf/dl-open.c (add_to_global): When global scope array must
	grow, allocate a new one and free old array only after all
	lookups finish.
	* elf/dl-runtime.c (_dl_fixup): Protect using global scope.
	(_dl_lookup_symbol_x): Likewise.
	* elf/dl-support.c: Define _dl_wait_lookup_done.
	* sysdeps/generic/ldsodefs.h (struct rtld_global): Add
	_dl_wait_lookup_done.

nptl/
2007-05-28  Jakub Jelinek  <jakub@redhat.com>

	* sysdeps/i386/tls.h (THREAD_GSCOPE_RESET_FLAG): Use explicit
	insn suffix.
	(THREAD_GSCOPE_GET_FLAG): Remove.
	* sysdeps/x86_64/tls.h (THREAD_GSCOPE_GET_FLAG): Remove.
	* allocatestack.c (__wait_lookup_done): Revert 2007-05-24
	changes.
	* sysdeps/powerpc/tls.h (tcbhead_t): Remove gscope_flag.
	(THREAD_GSCOPE_GET_FLAG): Remove.
	(THREAD_GSCOPE_RESET_FLAG): Use THREAD_SELF->header.gscope_flag
	instead of THREAD_GSCOPE_GET_FLAG.
	(THREAD_GSCOPE_SET_FLAG): Likewise.  Add atomic_write_barrier after
	it.
	* sysdeps/s390/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
	THREAD_GSCOPE_WAIT): Define.
	* sysdeps/sparc/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
	THREAD_GSCOPE_WAIT): Define.
	* sysdeps/sh/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
	THREAD_GSCOPE_WAIT): Define.
	* sysdeps/ia64/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
	THREAD_GSCOPE_WAIT): Define.

2007-05-24  Richard Henderson  <rth@redhat.com>

	* descr.h (struct pthread): Add header.gscope_flag.
	* sysdeps/alpha/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
	THREAD_GSCOPE_WAIT): Define.

2007-05-26  Ulrich Drepper  <drepper@redhat.com>

	* allocatestack.c: Revert last change.
	* init.c: Likewise.
	* sysdeps/i386/tls.h: Likewise.
	* sysdeps/x86_64/tls.h: Likewise.

2007-05-24  Jakub Jelinek  <jakub@redhat.com>

	* sysdeps/powerpc/tls.h (tcbhead_t): Add gscope_flag.
	(THREAD_GSCOPE_FLAG_UNUSED, THREAD_GSCOPE_FLAG_USED,
	THREAD_GSCOPE_FLAG_WAIT): Define.
	(THREAD_GSCOPE_GET_FLAG, THREAD_GSCOPE_SET_FLAG,
	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_WAIT): Define.
	* sysdeps/i386/tls.h (THREAD_GSCOPE_WAIT): Don't use
	PTR_DEMANGLE.
	(THREAD_GSCOPE_GET_FLAG): Define.
	* sysdeps/x86_64/tls.h (THREAD_GSCOPE_GET_FLAG): Define.
	* allocatestack.c (__wait_lookup_done): Use THREAD_GSCOPE_GET_FLAG
	instead of ->header.gscope_flag directly.

2007-05-21  Ulrich Drepper  <drepper@redhat.com>

	* sysdeps/pthread/pthread-functions.h (struct pthread_functions):
	Remove ptr_wait_lookup_done again.
	* init.c (pthread_functions): Don't add .ptr_wait_lookup_done here.
	(__pthread_initialize_minimal_internal): Initialize
	_dl_wait_lookup_done pointer in _rtld_global directly.
	* sysdeps/unix/sysv/linux/libc_pthread_init.c (__libc_pthread_init):
	Remove code to code _dl_wait_lookup_done.
	* sysdeps/x86_64/tls.h (THREAD_GSCOPE_WAIT): The pointer is not
	encrypted for now.

2007-05-19  Ulrich Drepper  <drepper@redhat.com>

	* allocatestack.c (__wait_lookup_done): New function.
	* sysdeps/pthread/pthread-functions.h (struct pthread_functions):
	Add ptr_wait_lookup_done.
	* init.c (pthread_functions): Initialize .ptr_wait_lookup_done.
	* pthreadP.h: Declare __wait_lookup_done.
	* sysdeps/i386/tls.h (tcbhead_t): Add gscope_flag.
	Define macros to implement reference handling of global scope.
	* sysdeps/x86_64/tls.h: Likewise.
	* sysdeps/unix/sysv/linux/libc_pthread_init.c (__libc_pthread_init):
	Initialize GL(dl_wait_lookup_done).

22 files changed, 445 insertions, 76 deletions

diff --git a/ChangeLog b/ChangeLog
index ad7cad416d..3b1946a483 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,30 @@
+2007-06-09  Ulrich Drepper  <drepper@redhat.com>
+
+	* elf/do-lookup.h (do_lookup_x): Read r_nlist before r_list and
+	make sure gcc doesn't mess around with this.
+
+2007-06-08  Ulrich Drepper  <drepper@redhat.com>
+
+	* elf/dl-lookup.c (_dl_lookup_symbol_x): Remove use of r_nlist.
+
+2007-06-08  Jakub Jelinek  <jakub@redhat.com>
+
+	* elf/dl-close.c (_dl_close_worker): Remove all to be removed
+	libraries from the global scope at once and call THREAD_GSCOPE_WAIT
+
+2007-05-18  Ulrich Drepper  <drepper@redhat.com>
+
+	* elf/dl-close.c (_dl_close_worker): When removing object from
+	global scope, wait for all lookups to finish afterwards.
+	* elf/dl-open.c (add_to_global): When global scope array must
+	grow, allocate a new one and free old array only after all
+	lookups finish.
+	* elf/dl-runtime.c (_dl_fixup): Protect using global scope.
+	(_dl_lookup_symbol_x): Likewise.
+	* elf/dl-support.c: Define _dl_wait_lookup_done.
+	* sysdeps/generic/ldsodefs.h (struct rtld_global): Add
+	_dl_wait_lookup_done.
+
 2007-06-05  Jakub Jelinek  <jakub@redhat.com>
 
 	* sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c
@@ -56,12 +83,6 @@
 	* sysdeps/x86_64/mempcpy.S: Adjust appropriately.
 	Patch by Evandro Menezes <evandro.menezes@amd.com>.
 
-2007-01-15  Jakub Jelinek  <jakub@redhat.com>
-
-	* elf/dl-open.c (add_to_global): If the main searchlist is 256
-	entries or more, on each reallocation at least double the size
-	of the search list rather than growing it linearly.
-
 2007-05-21  Ulrich Drepper  <drepper@redhat.com>
 
 	* sysdeps/unix/sysv/linux/i386/epoll_pwait.S: New file.
diff --git a/elf/dl-close.c b/elf/dl-close.c
index e0fe26ad02..2c2b3b6163 100644
--- a/elf/dl-close.c
+++ b/elf/dl-close.c
@@ -31,6 +31,7 @@
 #include <sys/types.h>
 #include <sys/mman.h>
 #include <sysdep-cancel.h>
+#include <tls.h>
 
 
 /* Type of the constructor functions.  */
@@ -228,6 +229,7 @@ _dl_close_worker (struct link_map *map)
   bool do_audit = GLRO(dl_naudit) > 0 && !ns->_ns_loaded->l_auditing;
 #endif
   bool unload_any = false;
+  unsigned int unload_global = 0;
   unsigned int first_loaded = ~0;
   for (unsigned int i = 0; i < nloaded; ++i)
     {
@@ -292,6 +294,9 @@ _dl_close_worker (struct link_map *map)
 	  /* We indeed have an object to remove.  */
 	  unload_any = true;
 
+	  if (imap->l_global)
+	    ++unload_global;
+
 	  /* Remember where the first dynamically loaded object is.  */
 	  if (i < first_loaded)
 	    first_loaded = i;
@@ -457,6 +462,34 @@ _dl_close_worker (struct link_map *map)
   r->r_state = RT_DELETE;
   _dl_debug_state ();
 
+  if (unload_global)
+    {
+      /* Some objects are in the global scope list.  Remove them.  */
+      struct r_scope_elem *ns_msl = ns->_ns_main_searchlist;
+      unsigned int i;
+      unsigned int j = 0;
+      unsigned int cnt = ns_msl->r_nlist;
+
+      while (cnt > 0 && ns_msl->r_list[cnt - 1]->l_removed)
+	--cnt;
+
+      if (cnt + unload_global == ns_msl->r_nlist)
+	/* Speed up removing most recently added objects.  */
+	j = cnt;
+      else
+ 	for (i = 0; i < cnt; i++)
+	  if (ns_msl->r_list[i]->l_removed == 0)
+	    {
+	      if (i != j)
+		ns_msl->r_list[j] = ns_msl->r_list[i];
+	      j++;
+	    }
+      ns_msl->r_nlist = j;
+
+      if (!RTLD_SINGLE_THREAD_P)
+	THREAD_GSCOPE_WAIT ();
+    }
+
   size_t tls_free_start;
   size_t tls_free_end;
   tls_free_start = tls_free_end = NO_TLS_OFFSET;
@@ -472,22 +505,6 @@ _dl_close_worker (struct link_map *map)
 
 	  /* That was the last reference, and this was a dlopen-loaded
 	     object.  We can unmap it.  */
-	  if (__builtin_expect (imap->l_global, 0))
-	    {
-	      /* This object is in the global scope list.  Remove it.  */
-	      struct r_scope_elem *ns_msl = ns->_ns_main_searchlist;
-	      unsigned int cnt = ns_msl->r_nlist;
-
-	      do
-		--cnt;
-	      while (ns_msl->r_list[cnt] != imap);
-
-	      /* The object was already correctly registered.  */
-	      while (++cnt < ns_msl->r_nlist)
-		ns_msl->r_list[cnt - 1] = ns_msl->r_list[cnt];
-
-	      --ns_msl->r_nlist;
-	    }
 
 	  /* Remove the object from the dtv slotinfo array if it uses TLS.  */
 	  if (__builtin_expect (imap->l_tls_blocksize > 0, 0))
diff --git a/elf/dl-lookup.c b/elf/dl-lookup.c
index a6a958419f..dc1b865bc7 100644
--- a/elf/dl-lookup.c
+++ b/elf/dl-lookup.c
@@ -244,13 +244,9 @@ _dl_lookup_symbol_x (const char *undef_name, struct link_map *undef_map,
 
   size_t i = 0;
   if (__builtin_expect (skip_map != NULL, 0))
-    {
-      /* Search the relevant loaded objects for a definition.  */
-      while ((*scope)->r_list[i] != skip_map)
-	++i;
-
-      assert (i < (*scope)->r_nlist);
-    }
+    /* Search the relevant loaded objects for a definition.  */
+    while ((*scope)->r_list[i] != skip_map)
+      ++i;
 
   /* Search the relevant loaded objects for a definition.  */
   for (size_t start = i; *scope != NULL; start = 0, ++scope)
diff --git a/elf/dl-open.c b/elf/dl-open.c
index 0afe94b5bf..a043cf61b6 100644
--- a/elf/dl-open.c
+++ b/elf/dl-open.c
@@ -32,6 +32,7 @@
 #include <bp-sym.h>
 #include <caller.h>
 #include <sysdep-cancel.h>
+#include <tls.h>
 
 #include <dl-dst.h>
 
@@ -125,19 +126,25 @@ add_to_global (struct link_map *new)
     {
       /* We have to extend the existing array of link maps in the
 	 main map.  */
-      size_t new_size = ns->_ns_global_scope_alloc;
-      if (new_size >= 256 && new_size > to_add + 8)
-	new_size *= 2;
-      else
-	new_size += to_add + 8;
+      struct link_map **old_global
+	= GL(dl_ns)[new->l_ns]._ns_main_searchlist->r_list;
+      size_t new_nalloc = ((ns->_ns_global_scope_alloc + to_add) * 2);
+
       new_global = (struct link_map **)
-	realloc (ns->_ns_main_searchlist->r_list,
-		 new_size * sizeof (struct link_map *));
+	malloc (new_nalloc * sizeof (struct link_map *));
       if (new_global == NULL)
 	goto nomem;
 
-      ns->_ns_global_scope_alloc = new_size;
+      memcpy (new_global, old_global,
+	      ns->_ns_global_scope_alloc * sizeof (struct link_map *));
+
+      ns->_ns_global_scope_alloc = new_nalloc;
       ns->_ns_main_searchlist->r_list = new_global;
+
+      if (!RTLD_SINGLE_THREAD_P)
+	THREAD_GSCOPE_WAIT ();
+
+      free (old_global);
     }
 
   /* Now add the new entries.  */
diff --git a/elf/dl-runtime.c b/elf/dl-runtime.c
index 9ecf62b436..6add5e4fff 100644
--- a/elf/dl-runtime.c
+++ b/elf/dl-runtime.c
@@ -26,6 +26,8 @@
 #include <ldsodefs.h>
 #include <sysdep-cancel.h>
 #include "dynamic-link.h"
+#include <tls.h>
+
 
 #if (!defined ELF_MACHINE_NO_RELA && !defined ELF_MACHINE_PLT_REL) \
     || ELF_MACHINE_NO_REL
@@ -97,10 +99,15 @@ _dl_fixup (
 	 not necessary for objects which cannot be unloaded or when
 	 we are not using any threads (yet).  */
       int flags = DL_LOOKUP_ADD_DEPENDENCY;
-      if (l->l_type == lt_loaded && !RTLD_SINGLE_THREAD_P)
+      if (!RTLD_SINGLE_THREAD_P)
 	{
-	  __rtld_mrlock_lock (l->l_scope_lock);
-	  flags |= DL_LOOKUP_SCOPE_LOCK;
+	  THREAD_GSCOPE_SET_FLAG ();
+
+	  if (l->l_type == lt_loaded)
+	    {
+	      __rtld_mrlock_lock (l->l_scope_lock);
+	      flags |= DL_LOOKUP_SCOPE_LOCK;
+	    }
 	}
 
       result = _dl_lookup_symbol_x (strtab + sym->st_name, l, &sym, l->l_scope,
@@ -109,6 +116,10 @@ _dl_fixup (
       if ((flags & DL_LOOKUP_SCOPE_LOCK) != 0)
 	__rtld_mrlock_unlock (l->l_scope_lock);
 
+      /* We are done with the global scope.  */
+      if (!RTLD_SINGLE_THREAD_P)
+	THREAD_GSCOPE_RESET_FLAG ();
+
       /* Currently result contains the base load address (or link map)
 	 of the object that defines sym.  Now add in the symbol
 	 offset.  */
@@ -191,10 +202,15 @@ _dl_profile_fixup (
 	     not necessary for objects which cannot be unloaded or when
 	     we are not using any threads (yet).  */
 	  int flags = DL_LOOKUP_ADD_DEPENDENCY;
-	  if (l->l_type == lt_loaded && !RTLD_SINGLE_THREAD_P)
+	  if (!RTLD_SINGLE_THREAD_P)
 	    {
-	      __rtld_mrlock_lock (l->l_scope_lock);
-	      flags |= DL_LOOKUP_SCOPE_LOCK;
+	      THREAD_GSCOPE_SET_FLAG ();
+
+	      if (l->l_type == lt_loaded)
+		{
+		  __rtld_mrlock_lock (l->l_scope_lock);
+		  flags |= DL_LOOKUP_SCOPE_LOCK;
+		}
 	    }
 
 	  result = _dl_lookup_symbol_x (strtab + refsym->st_name, l,
@@ -204,6 +220,10 @@ _dl_profile_fixup (
 	  if ((flags & DL_LOOKUP_SCOPE_LOCK) != 0)
 	    __rtld_mrlock_unlock (l->l_scope_lock);
 
+	  /* We are done with the global scope.  */
+	  if (!RTLD_SINGLE_THREAD_P)
+	    THREAD_GSCOPE_RESET_FLAG ();
+
 	  /* Currently result contains the base load address (or link map)
 	     of the object that defines sym.  Now add in the symbol
 	     offset.  */
diff --git a/elf/dl-support.c b/elf/dl-support.c
index c7479dc639..cecb603ae6 100644
--- a/elf/dl-support.c
+++ b/elf/dl-support.c
@@ -1,5 +1,5 @@
 /* Support for dynamic linking code in static libc.
-   Copyright (C) 1996-2005, 2006 Free Software Foundation, Inc.
+   Copyright (C) 1996-2005, 2006, 2007 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
@@ -132,6 +132,9 @@ int (*_dl_make_stack_executable_hook) (void **) internal_function
   = _dl_make_stack_executable;
 
 
+/* Function in libpthread to wait for termination of lookups.  */
+void (*_dl_wait_lookup_done) (void);
+
 #ifdef NEED_DL_SYSINFO
 /* Needed for improved syscall handling on at least x86/Linux.  */
 uintptr_t _dl_sysinfo = DL_SYSINFO_DEFAULT;
diff --git a/elf/do-lookup.h b/elf/do-lookup.h
index 2585d83005..ab9a510ba1 100644
--- a/elf/do-lookup.h
+++ b/elf/do-lookup.h
@@ -1,5 +1,5 @@
 /* Look up a symbol in the loaded objects.
-   Copyright (C) 1995-2004, 2005, 2006 Free Software Foundation, Inc.
+   Copyright (C) 1995-2004, 2005, 2006, 2007 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
@@ -29,8 +29,13 @@ do_lookup_x (const char *undef_name, uint_fast32_t new_hash,
 	     const struct r_found_version *const version, int flags,
 	     struct link_map *skip, int type_class)
 {
-  struct link_map **list = scope->r_list;
   size_t n = scope->r_nlist;
+  /* Make sure we read the value before proceeding.  Otherwise we
+     might use r_list pointing to the initial scope and r_nlist being
+     the value after a resize.  That is the only path in dl-open.c not
+     protected by GSCOPE.  A read barrier here might be to expensive.  */
+  __asm volatile ("" : "+r" (n), "+m" (scope->r_list));
+  struct link_map **list = scope->r_list;
 
   do
     {
diff --git a/nptl/ChangeLog b/nptl/ChangeLog
index ea877d0374..bb7dd528ab 100644
--- a/nptl/ChangeLog
+++ b/nptl/ChangeLog
@@ -1,3 +1,88 @@
+2007-05-28  Jakub Jelinek  <jakub@redhat.com>
+
+	* sysdeps/i386/tls.h (THREAD_GSCOPE_RESET_FLAG): Use explicit
+	insn suffix.
+	(THREAD_GSCOPE_GET_FLAG): Remove.
+	* sysdeps/x86_64/tls.h (THREAD_GSCOPE_GET_FLAG): Remove.
+	* allocatestack.c (__wait_lookup_done): Revert 2007-05-24
+	changes.
+	* sysdeps/powerpc/tls.h (tcbhead_t): Remove gscope_flag.
+	(THREAD_GSCOPE_GET_FLAG): Remove.
+	(THREAD_GSCOPE_RESET_FLAG): Use THREAD_SELF->header.gscope_flag
+	instead of THREAD_GSCOPE_GET_FLAG.
+	(THREAD_GSCOPE_SET_FLAG): Likewise.  Add atomic_write_barrier after
+	it.
+	* sysdeps/s390/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
+	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
+	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
+	THREAD_GSCOPE_WAIT): Define.
+	* sysdeps/sparc/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
+	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
+	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
+	THREAD_GSCOPE_WAIT): Define.
+	* sysdeps/sh/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
+	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
+	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
+	THREAD_GSCOPE_WAIT): Define.
+	* sysdeps/ia64/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
+	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
+	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
+	THREAD_GSCOPE_WAIT): Define.
+
+2007-05-24  Richard Henderson  <rth@redhat.com>
+
+	* descr.h (struct pthread): Add header.gscope_flag.
+	* sysdeps/alpha/tls.h (THREAD_GSCOPE_FLAG_UNUSED,
+	THREAD_GSCOPE_FLAG_USED, THREAD_GSCOPE_FLAG_WAIT,
+	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_SET_FLAG,
+	THREAD_GSCOPE_WAIT): Define.
+
+2007-05-26  Ulrich Drepper  <drepper@redhat.com>
+
+	* allocatestack.c: Revert last change.
+	* init.c: Likewise.
+	* sysdeps/i386/tls.h: Likewise.
+	* sysdeps/x86_64/tls.h: Likewise.
+
+2007-05-24  Jakub Jelinek  <jakub@redhat.com>
+
+	* sysdeps/powerpc/tls.h (tcbhead_t): Add gscope_flag.
+	(THREAD_GSCOPE_FLAG_UNUSED, THREAD_GSCOPE_FLAG_USED,
+	THREAD_GSCOPE_FLAG_WAIT): Define.
+	(THREAD_GSCOPE_GET_FLAG, THREAD_GSCOPE_SET_FLAG,
+	THREAD_GSCOPE_RESET_FLAG, THREAD_GSCOPE_WAIT): Define.
+	* sysdeps/i386/tls.h (THREAD_GSCOPE_WAIT): Don't use
+	PTR_DEMANGLE.
+	(THREAD_GSCOPE_GET_FLAG): Define.
+	* sysdeps/x86_64/tls.h (THREAD_GSCOPE_GET_FLAG): Define.
+	* allocatestack.c (__wait_lookup_done): Use THREAD_GSCOPE_GET_FLAG
+	instead of ->header.gscope_flag directly.
+
+2007-05-21  Ulrich Drepper  <drepper@redhat.com>
+
+	* sysdeps/pthread/pthread-functions.h (struct pthread_functions):
+	Remove ptr_wait_lookup_done again.
+	* init.c (pthread_functions): Don't add .ptr_wait_lookup_done here.
+	(__pthread_initialize_minimal_internal): Initialize
+	_dl_wait_lookup_done pointer in _rtld_global directly.
+	* sysdeps/unix/sysv/linux/libc_pthread_init.c (__libc_pthread_init):
+	Remove code to code _dl_wait_lookup_done.
+	* sysdeps/x86_64/tls.h (THREAD_GSCOPE_WAIT): The pointer is not
+	encrypted for now.
+
+2007-05-19  Ulrich Drepper  <drepper@redhat.com>
+
+	* allocatestack.c (__wait_lookup_done): New function.
+	* sysdeps/pthread/pthread-functions.h (struct pthread_functions):
+	Add ptr_wait_lookup_done.
+	* init.c (pthread_functions): Initialize .ptr_wait_lookup_done.
+	* pthreadP.h: Declare __wait_lookup_done.
+	* sysdeps/i386/tls.h (tcbhead_t): Add gscope_flag.
+	Define macros to implement reference handling of global scope.
+	* sysdeps/x86_64/tls.h: Likewise.
+	* sysdeps/unix/sysv/linux/libc_pthread_init.c (__libc_pthread_init):
+	Initialize GL(dl_wait_lookup_done).
+
 2007-05-25  Ulrich Drepper  <drepper@redhat.com>
 
 	* Makefile (tests): Add tst-sem10.
@@ -16,12 +101,6 @@
 	* tst-robust9.c (do_test): Don't fail if ENABLE_PI and
 	pthread_mutex_init failed with ENOTSUP.
 
-2007-01-15  Jakub Jelinek  <jakub@redhat.com>
-
-	* pthread_create.c (__pthread_create_2_1): On the first pthread_create
-	in a process make sure main search list can store at least 256
-	entries.
-
 2007-05-17  Ulrich Drepper  <drepper@redhat.com>
 
 	[BZ #4512]
diff --git a/nptl/allocatestack.c b/nptl/allocatestack.c
index 6b60642042..e556dbac08 100644
--- a/nptl/allocatestack.c
+++ b/nptl/allocatestack.c
@@ -996,3 +996,60 @@ __pthread_init_static_tls (struct link_map *map)
 
   lll_unlock (stack_cache_lock);
 }
+
+
+void
+attribute_hidden
+__wait_lookup_done (void)
+{
+  lll_lock (stack_cache_lock);
+
+  struct pthread *self = THREAD_SELF;
+
+  /* Iterate over the list with system-allocated threads first.  */
+  list_t *runp;
+  list_for_each (runp, &stack_used)
+    {
+      struct pthread *t = list_entry (runp, struct pthread, list);
+      if (t == self || t->header.gscope_flag == THREAD_GSCOPE_FLAG_UNUSED)
+	continue;
+
+      int *const gscope_flagp = &t->header.gscope_flag;
+
+      /* We have to wait until this thread is done with the global
+	 scope.  First tell the thread that we are waiting and
+	 possibly have to be woken.  */
+      if (atomic_compare_and_exchange_bool_acq (gscope_flagp,
+						THREAD_GSCOPE_FLAG_WAIT,
+						THREAD_GSCOPE_FLAG_USED))
+	continue;
+
+      do
+	lll_futex_wait (gscope_flagp, THREAD_GSCOPE_FLAG_WAIT);
+      while (*gscope_flagp == THREAD_GSCOPE_FLAG_WAIT);
+    }
+
+  /* Now the list with threads using user-allocated stacks.  */
+  list_for_each (runp, &__stack_user)
+    {
+      struct pthread *t = list_entry (runp, struct pthread, list);
+      if (t == self || t->header.gscope_flag == THREAD_GSCOPE_FLAG_UNUSED)
+	continue;
+
+      int *const gscope_flagp = &t->header.gscope_flag;
+
+      /* We have to wait until this thread is done with the global
+	 scope.  First tell the thread that we are waiting and
+	 possibly have to be woken.  */
+      if (atomic_compare_and_exchange_bool_acq (gscope_flagp,
+						THREAD_GSCOPE_FLAG_WAIT,
+						THREAD_GSCOPE_FLAG_USED))
+	continue;
+
+      do
+	lll_futex_wait (gscope_flagp, THREAD_GSCOPE_FLAG_WAIT);
+      while (*gscope_flagp == THREAD_GSCOPE_FLAG_WAIT);
+    }
+
+  lll_unlock (stack_cache_lock);
+}
diff --git a/nptl/descr.h b/nptl/descr.h
index 00cad1aa83..74d8c44140 100644
--- a/nptl/descr.h
+++ b/nptl/descr.h
@@ -131,6 +131,7 @@ struct pthread
     struct
     {
       int multiple_threads;
+      int gscope_flag;
     } header;
 #endif
 
diff --git a/nptl/init.c b/nptl/init.c
index dddc975a5e..827e795c2d 100644
--- a/nptl/init.c
+++ b/nptl/init.c
@@ -386,6 +386,8 @@ __pthread_initialize_minimal_internal (void)
 
   GL(dl_init_static_tls) = &__pthread_init_static_tls;
 
+  GL(dl_wait_lookup_done) = &__wait_lookup_done;
+
   /* Register the fork generation counter with the libc.  */
 #ifndef TLS_MULTIPLE_THREADS_IN_TCB
   __libc_multiple_threads_ptr =
diff --git a/nptl/pthreadP.h b/nptl/pthreadP.h
index f9634ab0ff..f560f72e4e 100644
--- a/nptl/pthreadP.h
+++ b/nptl/pthreadP.h
@@ -545,6 +545,8 @@ extern int __nptl_setxid (struct xid_command *cmdp) attribute_hidden;
 
 extern void __free_stack_cache (void) attribute_hidden;
 
+extern void __wait_lookup_done (void) attribute_hidden;
+
 #ifdef SHARED
 # define PTHREAD_STATIC_FN_REQUIRE(name)
 #else
diff --git a/nptl/pthread_create.c b/nptl/pthread_create.c
index 22e8f01804..79729ced03 100644
--- a/nptl/pthread_create.c
+++ b/nptl/pthread_create.c
@@ -462,30 +462,6 @@ __pthread_create_2_1 (newthread, attr, start_routine, arg)
   pd->flags = ((iattr->flags & ~(ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET))
 	       | (self->flags & (ATTR_FLAG_SCHED_SET | ATTR_FLAG_POLICY_SET)));
 
-  /* Hack: realloc the main search list on the first pthread_create call
-     to minimize the number of global search scope reallocations.
-     Wastes at most 1KB on 32-bit and 2KB on 64-bit per process
-     which calls pthread_create.  */
-  if (__builtin_expect (self->header.multiple_threads == 0, 0)
-      && GL(dl_ns)[0]._ns_main_searchlist
-      && GL(dl_ns)[0]._ns_main_searchlist->r_nlist < 256
-      && GL(dl_ns)[0]._ns_global_scope_alloc < 256)
-    {
-      struct link_map **new_global = (struct link_map **)
-	realloc (GL(dl_ns)[0]._ns_global_scope_alloc == 0
-		 ? NULL : GL(dl_ns)[0]._ns_main_searchlist->r_list,
-		 256 * sizeof (struct link_map *));
-      if (new_global != NULL)
-	{
-	  if (GL(dl_ns)[0]._ns_global_scope_alloc == 0)
-	    memcpy (new_global, GL(dl_ns)[0]._ns_main_searchlist->r_list,
-		    GL(dl_ns)[0]._ns_main_searchlist->r_nlist
-		    * sizeof (struct link_map *));
-	  GL(dl_ns)[0]._ns_global_scope_alloc = 256;
-	  GL(dl_ns)[0]._ns_main_searchlist->r_list = new_global;
-	}
-    }
-
   /* Initialize the field for the ID of the thread which is waiting
      for us.  This is a self-reference in case the thread is created
      detached.  */
diff --git a/nptl/sysdeps/alpha/tls.h b/nptl/sysdeps/alpha/tls.h
index be2430f676..9072e2d6fb 100644
--- a/nptl/sysdeps/alpha/tls.h
+++ b/nptl/sysdeps/alpha/tls.h
@@ -121,6 +121,29 @@ typedef struct
 #define THREAD_SETMEM_NC(descr, member, idx, value) \
   descr->member[idx] = (value)
 
+/* Get and set the global scope generation counter in struct pthread.  */
+#define THREAD_GSCOPE_FLAG_UNUSED 0
+#define THREAD_GSCOPE_FLAG_USED   1
+#define THREAD_GSCOPE_FLAG_WAIT   2
+#define THREAD_GSCOPE_RESET_FLAG() \
+  do									     \
+    { int __res								     \
+	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
+			       THREAD_GSCOPE_FLAG_UNUSED);		     \
+      if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1);		     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_SET_FLAG() \
+  do									     \
+    {									     \
+      THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
+      atomic_write_barrier ();						     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_WAIT() \
+  GL(dl_wait_lookup_done) ()
+
 #endif /* __ASSEMBLER__ */
 
 #endif	/* tls.h */
diff --git a/nptl/sysdeps/i386/tls.h b/nptl/sysdeps/i386/tls.h
index d5b3797e69..7c80cd7d14 100644
--- a/nptl/sysdeps/i386/tls.h
+++ b/nptl/sysdeps/i386/tls.h
@@ -51,6 +51,7 @@ typedef struct
   uintptr_t sysinfo;
   uintptr_t stack_guard;
   uintptr_t pointer_guard;
+  int gscope_flag;
 } tcbhead_t;
 
 # define TLS_MULTIPLE_THREADS_IN_TCB 1
@@ -431,6 +432,26 @@ union user_desc_init
    = THREAD_GETMEM (THREAD_SELF, header.pointer_guard))
 
 
+/* Get and set the global scope generation counter in the TCB head.  */
+#define THREAD_GSCOPE_FLAG_UNUSED 0
+#define THREAD_GSCOPE_FLAG_USED   1
+#define THREAD_GSCOPE_FLAG_WAIT   2
+#define THREAD_GSCOPE_RESET_FLAG() \
+  do									      \
+    { int __res;							      \
+      asm volatile ("xchgl %0, %%gs:%P1"				      \
+		    : "=r" (__res)					      \
+		    : "i" (offsetof (struct pthread, header.gscope_flag)),    \
+		      "0" (THREAD_GSCOPE_FLAG_UNUSED));			      \
+      if (__res == THREAD_GSCOPE_FLAG_WAIT)				      \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1);		      \
+    }									      \
+  while (0)
+#define THREAD_GSCOPE_SET_FLAG() \
+  THREAD_SETMEM (THREAD_SELF, header.gscope_flag, THREAD_GSCOPE_FLAG_USED)
+#define THREAD_GSCOPE_WAIT() \
+  GL(dl_wait_lookup_done) ()
+
 #endif /* __ASSEMBLER__ */
 
 #endif	/* tls.h */
diff --git a/nptl/sysdeps/ia64/tls.h b/nptl/sysdeps/ia64/tls.h
index 22a8b08144..4270723b9a 100644
--- a/nptl/sysdeps/ia64/tls.h
+++ b/nptl/sysdeps/ia64/tls.h
@@ -163,6 +163,29 @@ register struct pthread *__thread_self __asm__("r13");
   (((uintptr_t *) ((char *) (descr) + TLS_PRE_TCB_SIZE))[-2] \
    = THREAD_GET_POINTER_GUARD ())
 
+/* Get and set the global scope generation counter in struct pthread.  */
+#define THREAD_GSCOPE_FLAG_UNUSED 0
+#define THREAD_GSCOPE_FLAG_USED   1
+#define THREAD_GSCOPE_FLAG_WAIT   2
+#define THREAD_GSCOPE_RESET_FLAG() \
+  do									     \
+    { int __res								     \
+	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
+			       THREAD_GSCOPE_FLAG_UNUSED);		     \
+      if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1);		     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_SET_FLAG() \
+  do									     \
+    {									     \
+      THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
+      atomic_write_barrier ();						     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_WAIT() \
+  GL(dl_wait_lookup_done) ()
+
 #endif /* __ASSEMBLER__ */
 
 #endif	/* tls.h */
diff --git a/nptl/sysdeps/powerpc/tls.h b/nptl/sysdeps/powerpc/tls.h
index ddaafe23d0..de822833f9 100644
--- a/nptl/sysdeps/powerpc/tls.h
+++ b/nptl/sysdeps/powerpc/tls.h
@@ -180,6 +180,29 @@ register void *__thread_register __asm__ ("r13");
    different value to mean unset l_tls_offset.  */
 # define NO_TLS_OFFSET		-1
 
+/* Get and set the global scope generation counter in struct pthread.  */
+#define THREAD_GSCOPE_FLAG_UNUSED 0
+#define THREAD_GSCOPE_FLAG_USED   1
+#define THREAD_GSCOPE_FLAG_WAIT   2
+#define THREAD_GSCOPE_RESET_FLAG() \
+  do									     \
+    { int __res								     \
+	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
+			       THREAD_GSCOPE_FLAG_UNUSED);		     \
+      if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1);		     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_SET_FLAG() \
+  do									     \
+    {									     \
+      THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
+      atomic_write_barrier ();						     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_WAIT() \
+  GL(dl_wait_lookup_done) ()
+
 #endif /* __ASSEMBLER__ */
 
 #endif	/* tls.h */
diff --git a/nptl/sysdeps/s390/tls.h b/nptl/sysdeps/s390/tls.h
index 6f6f17b975..1c86a19676 100644
--- a/nptl/sysdeps/s390/tls.h
+++ b/nptl/sysdeps/s390/tls.h
@@ -50,6 +50,7 @@ typedef struct
   int multiple_threads;
   uintptr_t sysinfo;
   uintptr_t stack_guard;
+  int gscope_flag;
 } tcbhead_t;
 
 # ifndef __s390x__
@@ -168,6 +169,29 @@ typedef struct
 #define THREAD_SET_POINTER_GUARD(value)
 #define THREAD_COPY_POINTER_GUARD(descr)
 
+/* Get and set the global scope generation counter in struct pthread.  */
+#define THREAD_GSCOPE_FLAG_UNUSED 0
+#define THREAD_GSCOPE_FLAG_USED   1
+#define THREAD_GSCOPE_FLAG_WAIT   2
+#define THREAD_GSCOPE_RESET_FLAG() \
+  do									     \
+    { int __res								     \
+	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
+			       THREAD_GSCOPE_FLAG_UNUSED);		     \
+      if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1);		     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_SET_FLAG() \
+  do									     \
+    {									     \
+      THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
+      atomic_write_barrier ();						     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_WAIT() \
+  GL(dl_wait_lookup_done) ()
+
 #endif /* __ASSEMBLER__ */
 
 #endif	/* tls.h */
diff --git a/nptl/sysdeps/sh/tls.h b/nptl/sysdeps/sh/tls.h
index d9aa1073b8..09522189a6 100644
--- a/nptl/sysdeps/sh/tls.h
+++ b/nptl/sysdeps/sh/tls.h
@@ -150,6 +150,29 @@ typedef struct
      __asm __volatile ("stc gbr,%0" : "=r" (__tcbp));			      \
      ((tcbhead_t *) (descr + 1))->pointer_guard	= __tcbp->pointer_guard;})
 
+/* Get and set the global scope generation counter in struct pthread.  */
+#define THREAD_GSCOPE_FLAG_UNUSED 0
+#define THREAD_GSCOPE_FLAG_USED   1
+#define THREAD_GSCOPE_FLAG_WAIT   2
+#define THREAD_GSCOPE_RESET_FLAG() \
+  do									     \
+    { int __res								     \
+	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
+			       THREAD_GSCOPE_FLAG_UNUSED);		     \
+      if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1);		     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_SET_FLAG() \
+  do									     \
+    {									     \
+      THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
+      atomic_write_barrier ();						     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_WAIT() \
+  GL(dl_wait_lookup_done) ()
+
 #endif /* __ASSEMBLER__ */
 
 #endif	/* tls.h */
diff --git a/nptl/sysdeps/sparc/tls.h b/nptl/sysdeps/sparc/tls.h
index 4fbe426595..8a0c930b3e 100644
--- a/nptl/sysdeps/sparc/tls.h
+++ b/nptl/sysdeps/sparc/tls.h
@@ -141,6 +141,29 @@ register struct pthread *__thread_self __asm__("%g7");
 # define THREAD_COPY_POINTER_GUARD(descr) \
   ((descr)->header.pointer_guard = THREAD_GET_POINTER_GUARD ())
 
+/* Get and set the global scope generation counter in struct pthread.  */
+#define THREAD_GSCOPE_FLAG_UNUSED 0
+#define THREAD_GSCOPE_FLAG_USED   1
+#define THREAD_GSCOPE_FLAG_WAIT   2
+#define THREAD_GSCOPE_RESET_FLAG() \
+  do									     \
+    { int __res								     \
+	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
+			       THREAD_GSCOPE_FLAG_UNUSED);		     \
+      if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1);		     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_SET_FLAG() \
+  do									     \
+    {									     \
+      THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
+      atomic_write_barrier ();						     \
+    }									     \
+  while (0)
+#define THREAD_GSCOPE_WAIT() \
+  GL(dl_wait_lookup_done) ()
+
 #endif /* !ASSEMBLER */
 
 #endif	/* tls.h */
diff --git a/nptl/sysdeps/x86_64/tls.h b/nptl/sysdeps/x86_64/tls.h
index 0b5aeb00ff..4a614c02af 100644
--- a/nptl/sysdeps/x86_64/tls.h
+++ b/nptl/sysdeps/x86_64/tls.h
@@ -47,6 +47,7 @@ typedef struct
   dtv_t *dtv;
   void *self;		/* Pointer to the thread descriptor.  */
   int multiple_threads;
+  int gscope_flag;
   uintptr_t sysinfo;
   uintptr_t stack_guard;
   uintptr_t pointer_guard;
@@ -337,6 +338,26 @@ typedef struct
    = THREAD_GETMEM (THREAD_SELF, header.pointer_guard))
 
 
+/* Get and set the global scope generation counter in the TCB head.  */
+#define THREAD_GSCOPE_FLAG_UNUSED 0
+#define THREAD_GSCOPE_FLAG_USED   1
+#define THREAD_GSCOPE_FLAG_WAIT   2
+#define THREAD_GSCOPE_RESET_FLAG() \
+  do									      \
+    { int __res;							      \
+      asm volatile ("xchgl %0, %%fs:%P1"				      \
+		    : "=r" (__res)					      \
+		    : "i" (offsetof (struct pthread, header.gscope_flag)),    \
+		      "0" (THREAD_GSCOPE_FLAG_UNUSED));			      \
+      if (__res == THREAD_GSCOPE_FLAG_WAIT)				      \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1);		      \
+    }									      \
+  while (0)
+#define THREAD_GSCOPE_SET_FLAG() \
+  THREAD_SETMEM (THREAD_SELF, header.gscope_flag, THREAD_GSCOPE_FLAG_USED)
+#define THREAD_GSCOPE_WAIT() \
+  GL(dl_wait_lookup_done) ()
+
 #endif /* __ASSEMBLER__ */
 
 #endif	/* tls.h */
diff --git a/sysdeps/generic/ldsodefs.h b/sysdeps/generic/ldsodefs.h
index a9d20b2150..5205c41493 100644
--- a/sysdeps/generic/ldsodefs.h
+++ b/sysdeps/generic/ldsodefs.h
@@ -486,6 +486,8 @@ struct rtld_global
 
   EXTERN void (*_dl_init_static_tls) (struct link_map *);
 
+  EXTERN void (*_dl_wait_lookup_done) (void);
+
 #ifdef SHARED
 };
 # define __rtld_global_attribute__