* 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.

14 files changed, 204 insertions, 37 deletions

diff --git a/ChangeLog b/ChangeLog
index 5413c12889..4de00e6a8d 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,16 @@
 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.
+
 	* malloc/malloc.c (do_check_chunk): Correct check for mmaped block
 	not overlapping with arena.
 
diff --git a/elf/dl-close.c b/elf/dl-close.c
index e0fe26ad02..8e5c9fc033 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.  */
@@ -487,6 +488,9 @@ _dl_close_worker (struct link_map *map)
 		ns_msl->r_list[cnt - 1] = ns_msl->r_list[cnt];
 
 	      --ns_msl->r_nlist;
+
+	      if (!RTLD_SINGLE_THREAD_P)
+		THREAD_GSCOPE_WAIT ();
 	    }
 
 	  /* Remove the object from the dtv slotinfo array if it uses TLS.  */
diff --git a/elf/dl-open.c b/elf/dl-open.c
index 583878781e..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>
 
@@ -125,15 +126,25 @@ add_to_global (struct link_map *new)
     {
       /* We have to extend the existing array of link maps in the
 	 main map.  */
+      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,
-		 ((ns->_ns_global_scope_alloc + to_add + 8)
-		  * sizeof (struct link_map *)));
+	malloc (new_nalloc * sizeof (struct link_map *));
       if (new_global == NULL)
 	goto nomem;
 
-      ns->_ns_global_scope_alloc += to_add + 8;
+      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/nptl/ChangeLog b/nptl/ChangeLog
index 894a9bd9fb..db7b86f2fb 100644
--- a/nptl/ChangeLog
+++ b/nptl/ChangeLog
@@ -1,3 +1,16 @@
+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).
+
 2007-05-17  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/init.c b/nptl/init.c
index dddc975a5e..fb4030e249 100644
--- a/nptl/init.c
+++ b/nptl/init.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc.
+/* Copyright (C) 2002,2003,2004,2005,2006,2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
 
@@ -35,27 +35,6 @@
 #include <lowlevellock.h>
 
 
-#ifndef __NR_set_tid_address
-/* XXX For the time being...  Once we can rely on the kernel headers
-   having the definition remove these lines.  */
-#if defined __s390__
-# define __NR_set_tid_address	252
-#elif defined __ia64__
-# define __NR_set_tid_address	1233
-#elif defined __i386__
-# define __NR_set_tid_address	258
-#elif defined __x86_64__
-# define __NR_set_tid_address	218
-#elif defined __powerpc__
-# define __NR_set_tid_address	232
-#elif defined __sparc__
-# define __NR_set_tid_address	166
-#else
-# error "define __NR_set_tid_address"
-#endif
-#endif
-
-
 /* Size and alignment of static TLS block.  */
 size_t __static_tls_size;
 size_t __static_tls_align_m1;
@@ -138,7 +117,8 @@ static const struct pthread_functions pthread_functions =
     .ptr__nptl_deallocate_tsd = __nptl_deallocate_tsd,
     .ptr__nptl_setxid = __nptl_setxid,
     /* For now only the stack cache needs to be freed.  */
-    .ptr_freeres = __free_stack_cache
+    .ptr_freeres = __free_stack_cache,
+    .ptr_wait_lookup_done = __wait_lookup_done
   };
 # define ptr_pthread_functions &pthread_functions
 #else
diff --git a/nptl/pthreadP.h b/nptl/pthreadP.h
index f9634ab0ff..21ce6fe0b7 100644
--- a/nptl/pthreadP.h
+++ b/nptl/pthreadP.h
@@ -1,4 +1,4 @@
-/* Copyright (C) 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc.
+/* Copyright (C) 2002,2003,2004,2005,2006,2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
 
@@ -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/sysdeps/i386/tls.h b/nptl/sysdeps/i386/tls.h
index d5b3797e69..d9044f3fde 100644
--- a/nptl/sysdeps/i386/tls.h
+++ b/nptl/sysdeps/i386/tls.h
@@ -1,5 +1,5 @@
 /* Definition for thread-local data handling.  nptl/i386 version.
-   Copyright (C) 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc.
+   Copyright (C) 2002,2003,2004,2005,2006,2007 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
@@ -27,6 +27,7 @@
 # include <stdint.h>
 # include <stdlib.h>
 # include <list.h>
+# include <sysdep.h>
 
 
 /* Type for the dtv.  */
@@ -51,6 +52,7 @@ 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 +433,35 @@ 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 ("xchg %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)
+#ifdef PTR_DEMANGLE
+# define THREAD_GSCOPE_WAIT() \
+  do { void (*ptr) (void) = GL(dl_wait_lookup_done);			      \
+       PTR_DEMANGLE (ptr);						      \
+       ptr ();								      \
+  } while (0)
+#else
+# define THREAD_GSCOPE_WAIT() \
+  GL(dl_wait_lookup_done) ()
+#endif
+
+
 #endif /* __ASSEMBLER__ */
 
 #endif	/* tls.h */
diff --git a/nptl/sysdeps/pthread/pthread-functions.h b/nptl/sysdeps/pthread/pthread-functions.h
index a13b937032..f0eddd3053 100644
--- a/nptl/sysdeps/pthread/pthread-functions.h
+++ b/nptl/sysdeps/pthread/pthread-functions.h
@@ -97,7 +97,7 @@ struct pthread_functions
   void (*ptr__nptl_deallocate_tsd) (void);
   int (*ptr__nptl_setxid) (struct xid_command *);
   void (*ptr_freeres) (void);
-  void (*ptr_wait_lookup_done) (int);
+  void (*ptr_wait_lookup_done) (void);
 };
 
 /* Variable in libc.so.  */
diff --git a/nptl/sysdeps/unix/sysv/linux/libc_pthread_init.c b/nptl/sysdeps/unix/sysv/linux/libc_pthread_init.c
index 92a188a2f3..25509eb3d6 100644
--- a/nptl/sysdeps/unix/sysv/linux/libc_pthread_init.c
+++ b/nptl/sysdeps/unix/sysv/linux/libc_pthread_init.c
@@ -26,6 +26,7 @@
 #include <pthreadP.h>
 #include <bits/libc-lock.h>
 #include <sysdep.h>
+#include <ldsodefs.h>
 
 
 #ifdef TLS_MULTIPLE_THREADS_IN_TCB
@@ -70,6 +71,12 @@ __libc_pthread_init (ptr, reclaim, functions)
       dest->parr[cnt] = p;
     }
   __libc_pthread_functions_init = 1;
+
+# ifdef RTLD_NOT_MANGLED
+  GL(dl_wait_lookup_done) = functions->ptr_wait_lookup_done;
+# else
+  GL(dl_wait_lookup_done) = __libc_pthread_functions.ptr_wait_lookup_done;
+# endif
 #endif
 
 #ifndef TLS_MULTIPLE_THREADS_IN_TCB
diff --git a/nptl/sysdeps/x86_64/tls.h b/nptl/sysdeps/x86_64/tls.h
index 0b5aeb00ff..00c9abbfcb 100644
--- a/nptl/sysdeps/x86_64/tls.h
+++ b/nptl/sysdeps/x86_64/tls.h
@@ -26,6 +26,7 @@
 # include <stddef.h>
 # include <stdint.h>
 # include <stdlib.h>
+# include <sysdep.h>
 
 
 /* Type for the dtv.  */
@@ -47,6 +48,7 @@ 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 +339,30 @@ 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() \
+  do { void (*ptr) (void) = GL(dl_wait_lookup_done);			      \
+       PTR_DEMANGLE (ptr);						      \
+       ptr ();								      \
+  } while (0)
+
+
 #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__