* iconvdata/gbk.c (BODY): Make buf and cp char instead of unsigned

	char array resp. pointer.
	* iconvdata/iso-2022-kr.c (BODY): Make buf unsigned char instead of
	char array.
	* iconvdata/cns11643.h (cns11643_to_ucs4): Change first argument
	to const unsigned char **.
	(ucs4_to_cns11643): Change second argument to unsigned char *.
	* iconvdata/euc-tw.c (BODY): Change endp type to
	const unsigned char *.
	* iconvdata/iso-ir-165.h (ucs4_to_isoir165): Change second argument
	to unsigned char *.
	* iconvdata/ibm1008_420.c (LOOP_NEED_FLAGS): Don't define.
	* iconvdata/iso-2022-cn.c (BODY): Change buf to unsigned char array.
	* iconvdata/iso-2022-cn-ext.c (BODY): Change buf, tmpbuf, tmp
	types to unsigned char pointers/arrays instead of char.
	* iconvdata/jis0201.h (ucs4_to_jisx0201): Change second argument
	to unsigned char *.
	* iconvdata/jis0208.h (ucs4_to_jisx0208): Likewise.
	* iconvdata/jis0212.h: Include assert.h.
	(ucs4_to_jisx0212): Change second argument to unsigned char *.
	assert that if cp[0] is not '\0', cp[1] is not '\0' either instead
	of trying to handle that.
	* iconvdata/euc-kr.c (euckr_from_ucs4): Initialize also cp[1] to
	shut up a warning.
	* iconvdata/euc-jp-ms.c (from_ucs4_lat1, from_ucs4_greek,
	from_ucs4_cjk, from_ucs4_cjkcpt, from_ucs4_extra): Change type to
	two dimensional const unsigned char arrays.
	(BODY): Cast "" to (const unsigned char *) for assignment to cp.
	Initialize endp to inptr to shut up a warning.

29 files changed, 362 insertions, 228 deletions

diff --git a/nptl/ChangeLog b/nptl/ChangeLog
index 68a7d3aadb..84681bb334 100644
--- a/nptl/ChangeLog
+++ b/nptl/ChangeLog
@@ -1,3 +1,88 @@
+2007-07-24  Jakub Jelinek  <jakub@redhat.com>
+
+	* allocatestack.c (__nptl_setxid, __wait_lookup_done): Replace
+	lll_private_futex_* (*) with lll_futex_* (*, LLL_PRIVATE).
+	* pthread_create.c (start_thread): Likewise.
+	* init.c (sighandler_setxid): Likewise.
+	* sysdeps/alpha/tls.h (THREAD_GSCOPE_RESET_FLAG): Likewise.
+	* sysdeps/ia64/tls.h (THREAD_GSCOPE_RESET_FLAG): Likewise.
+	* sysdeps/i386/tls.h (THREAD_GSCOPE_RESET_FLAG): Likewise.
+	* sysdeps/s390/tls.h (THREAD_GSCOPE_RESET_FLAG): Likewise.
+	* sysdeps/powerpc/tls.h (THREAD_GSCOPE_RESET_FLAG): Likewise.
+	* sysdeps/x86_64/tls.h (THREAD_GSCOPE_RESET_FLAG): Likewise.
+	* sysdeps/sparc/tls.h (THREAD_GSCOPE_RESET_FLAG): Likewise.
+	* sysdeps/sh/tls.h (THREAD_GSCOPE_RESET_FLAG): Likewise.
+	* sysdeps/pthread/aio_misc.h (AIO_MISC_NOTIFY, AIO_MISC_WAIT):
+	Likewise.
+	* sysdeps/pthread/gai_misc.h (GAI_MISC_NOTIFY, GAI_MISC_WAIT):
+	Likewise.
+	* sysdeps/unix/sysv/linux/unregister-atfork.c (__unregister_atfork):
+	Likewise.
+	* sysdeps/unix/sysv/linux/rtld-lowlevel.h (__rtld_waitzero,
+	__rtld_notify): Likewise.
+	* sysdeps/unix/sysv/linux/fork.c (__libc_fork): Likewise.
+	* sysdeps/unix/sysv/linux/powerpc/pthread_once.c (clear_once_control,
+	__pthread_once): Likewise.
+	* sysdeps/unix/sysv/linux/alpha/pthread_once.c (clear_once_control,
+	__pthread_once): Add LLL_PRIVATE as last argument to lll_futex_*.
+	* sysdeps/unix/sysv/linux/alpha/lowlevellock.h (FUTEX_PRIVATE_FLAG,
+	LLL_PRIVATE, LLL_SHARED, __lll_private_flag): Define.
+	(lll_futex_wait): Add private argument, define as wrapper around
+	lll_futex_timed_wait.
+	(lll_futex_timed_wait, lll_futex_wake): Add private argument,
+	use __lll_private_flag macro.
+	(lll_robust_mutex_dead, __lll_mutex_unlock, __lll_robust_mutex_unlock,
+	__lll_mutex_unlock_force): Pass LLL_SHARED as last arg to lll_futex_*.
+	* sysdeps/unix/sysv/linux/ia64/pthread_once.c (clear_once_control,
+	__pthread_once): Add LLL_PRIVATE as last argument to lll_futex_*.
+	* sysdeps/unix/sysv/linux/ia64/lowlevellock.h (FUTEX_PRIVATE_FLAG,
+	LLL_PRIVATE, LLL_SHARED, __lll_private_flag): Define.
+	(lll_futex_wait): Add private argument, define as wrapper around
+	lll_futex_timed_wait.
+	(lll_futex_timed_wait, lll_futex_wake): Add private argument,
+	use __lll_private_flag macro.
+	(__lll_mutex_unlock, __lll_robust_mutex_unlock, lll_wait_tid,
+	__lll_mutex_unlock_force): Pass LLL_SHARED as last arg to lll_futex_*.
+	* sysdeps/unix/sysv/linux/i386/lowlevellock.h (__lll_private_flag):
+	Define.
+	(lll_futex_timed_wait, lll_futex_wake): Use it.
+	(lll_private_futex_wait, lll_private_futex_timed_wait,
+	lll_private_futex_wake): Removed.
+	* sysdeps/unix/sysv/linux/s390/pthread_once.c (clear_once_control,
+	__pthread_once): Add LLL_PRIVATE as last argument to lll_futex_*.
+	* sysdeps/unix/sysv/linux/s390/lowlevellock.h (FUTEX_PRIVATE_FLAG,
+	LLL_PRIVATE, LLL_SHARED, __lll_private_flag): Define.
+	(lll_futex_wait): Add private argument, define as wrapper around
+	lll_futex_timed_wait.
+	(lll_futex_timed_wait, lll_futex_wake): Add private argument,
+	use __lll_private_flag macro.
+	(lll_robust_mutex_dead, __lll_mutex_unlock, __lll_robust_mutex_unlock,
+	lll_wait_tid, __lll_mutex_unlock_force): Pass LLL_SHARED as last arg
+	to lll_futex_*.
+	* sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
+	(lll_private_futex_wait, lll_private_futex_timed_wait,
+	lll_private_futex_wake): Removed.
+	* sysdeps/unix/sysv/linux/x86_64/lowlevellock.h (__lll_private_flag):
+	Fix !__ASSUME_PRIVATE_FUTEX non-constant private case.
+	(lll_private_futex_wait, lll_private_futex_timed_wait,
+	lll_private_futex_wake): Removed.
+	* sysdeps/unix/sysv/linux/sparc/pthread_once.c (clear_once_control,
+	__pthread_once): Add LLL_PRIVATE as last argument to lll_futex_*.
+	* sysdeps/unix/sysv/linux/sparc/lowlevellock.h (FUTEX_PRIVATE_FLAG,
+	LLL_PRIVATE, LLL_SHARED, __lll_private_flag): Define.
+	(lll_futex_wait): Add private argument, define as wrapper around
+	lll_futex_timed_wait.
+	(lll_futex_timed_wait, lll_futex_wake): Add private argument,
+	use __lll_private_flag macro.
+	(lll_robust_mutex_dead, __lll_mutex_unlock, __lll_robust_mutex_unlock,
+	lll_wait_tid, __lll_mutex_unlock_force): Pass LLL_SHARED as last arg
+	to lll_futex_*.
+	* sysdeps/unix/sysv/linux/sh/lowlevellock.h (__lll_private_flag):
+	Define.
+	(lll_futex_timed_wait, lll_futex_wake): Use it.
+	(lll_private_futex_wait, lll_private_futex_timed_wait,
+	lll_private_futex_wake): Removed.
+
 2007-07-27  Jakub Jelinek  <jakub@redhat.com>
 
 	* sysdeps/sparc/tls.h (tcbhead_t): Move gscope_flag to the end
diff --git a/nptl/allocatestack.c b/nptl/allocatestack.c
index 145fe0a35f..ddf91e5c10 100644
--- a/nptl/allocatestack.c
+++ b/nptl/allocatestack.c
@@ -951,7 +951,7 @@ __nptl_setxid (struct xid_command *cmdp)
   int cur = cmdp->cntr;
   while (cur != 0)
     {
-      lll_private_futex_wait (&cmdp->cntr, cur);
+      lll_futex_wait (&cmdp->cntr, cur, LLL_PRIVATE);
       cur = cmdp->cntr;
     }
 
@@ -1037,7 +1037,7 @@ __wait_lookup_done (void)
 	continue;
 
       do
-	lll_private_futex_wait (gscope_flagp, THREAD_GSCOPE_FLAG_WAIT);
+	lll_futex_wait (gscope_flagp, THREAD_GSCOPE_FLAG_WAIT, LLL_PRIVATE);
       while (*gscope_flagp == THREAD_GSCOPE_FLAG_WAIT);
     }
 
@@ -1059,7 +1059,7 @@ __wait_lookup_done (void)
 	continue;
 
       do
-	lll_private_futex_wait (gscope_flagp, THREAD_GSCOPE_FLAG_WAIT);
+	lll_futex_wait (gscope_flagp, THREAD_GSCOPE_FLAG_WAIT, LLL_PRIVATE);
       while (*gscope_flagp == THREAD_GSCOPE_FLAG_WAIT);
     }
 
diff --git a/nptl/init.c b/nptl/init.c
index a1aec6b8f7..95b9c7be5d 100644
--- a/nptl/init.c
+++ b/nptl/init.c
@@ -216,7 +216,7 @@ sighandler_setxid (int sig, siginfo_t *si, void *ctx)
 			__xidcmd->id[1], __xidcmd->id[2]);
 
   if (atomic_decrement_val (&__xidcmd->cntr) == 0)
-    lll_private_futex_wake (&__xidcmd->cntr, 1);
+    lll_futex_wake (&__xidcmd->cntr, 1, LLL_PRIVATE);
 
   /* Reset the SETXID flag.  */
   struct pthread *self = THREAD_SELF;
@@ -225,7 +225,7 @@ sighandler_setxid (int sig, siginfo_t *si, void *ctx)
 
   /* And release the futex.  */
   self->setxid_futex = 1;
-  lll_private_futex_wake (&self->setxid_futex, 1);
+  lll_futex_wake (&self->setxid_futex, 1, LLL_PRIVATE);
 }
 
 
diff --git a/nptl/pthread_create.c b/nptl/pthread_create.c
index b1f852dc95..ca55903c22 100644
--- a/nptl/pthread_create.c
+++ b/nptl/pthread_create.c
@@ -385,7 +385,7 @@ start_thread (void *arg)
       /* Some other thread might call any of the setXid functions and expect
 	 us to reply.  In this case wait until we did that.  */
       do
-	lll_private_futex_wait (&pd->setxid_futex, 0);
+	lll_futex_wait (&pd->setxid_futex, 0, LLL_PRIVATE);
       while (pd->cancelhandling & SETXID_BITMASK);
 
       /* Reset the value so that the stack can be reused.  */
diff --git a/nptl/sysdeps/alpha/tls.h b/nptl/sysdeps/alpha/tls.h
index 388a399c73..e77b1ffca9 100644
--- a/nptl/sysdeps/alpha/tls.h
+++ b/nptl/sysdeps/alpha/tls.h
@@ -131,7 +131,7 @@ typedef struct
 	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
 			       THREAD_GSCOPE_FLAG_UNUSED);		     \
       if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
-	lll_private_futex_wake (&THREAD_SELF->header.gscope_flag, 1);	     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1, LLL_PRIVATE);   \
     }									     \
   while (0)
 #define THREAD_GSCOPE_SET_FLAG() \
diff --git a/nptl/sysdeps/i386/tls.h b/nptl/sysdeps/i386/tls.h
index 1ffd4cf494..b5127420cf 100644
--- a/nptl/sysdeps/i386/tls.h
+++ b/nptl/sysdeps/i386/tls.h
@@ -449,7 +449,7 @@ union user_desc_init
 		    : "i" (offsetof (struct pthread, header.gscope_flag)),    \
 		      "0" (THREAD_GSCOPE_FLAG_UNUSED));			      \
       if (__res == THREAD_GSCOPE_FLAG_WAIT)				      \
-	lll_private_futex_wake (&THREAD_SELF->header.gscope_flag, 1);	      \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1, LLL_PRIVATE);    \
     }									      \
   while (0)
 #define THREAD_GSCOPE_SET_FLAG() \
diff --git a/nptl/sysdeps/ia64/tls.h b/nptl/sysdeps/ia64/tls.h
index a5b7b76e2a..936ff01a72 100644
--- a/nptl/sysdeps/ia64/tls.h
+++ b/nptl/sysdeps/ia64/tls.h
@@ -173,7 +173,7 @@ register struct pthread *__thread_self __asm__("r13");
 	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
 			       THREAD_GSCOPE_FLAG_UNUSED);		     \
       if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
-	lll_private_futex_wake (&THREAD_SELF->header.gscope_flag, 1);	     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1, LLL_PRIVATE);   \
     }									     \
   while (0)
 #define THREAD_GSCOPE_SET_FLAG() \
diff --git a/nptl/sysdeps/powerpc/tls.h b/nptl/sysdeps/powerpc/tls.h
index 06c60e8ba2..0f4d5290dd 100644
--- a/nptl/sysdeps/powerpc/tls.h
+++ b/nptl/sysdeps/powerpc/tls.h
@@ -190,7 +190,7 @@ register void *__thread_register __asm__ ("r13");
 	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
 			       THREAD_GSCOPE_FLAG_UNUSED);		     \
       if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
-	lll_private_futex_wake (&THREAD_SELF->header.gscope_flag, 1);	     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1, LLL_PRIVATE);   \
     }									     \
   while (0)
 #define THREAD_GSCOPE_SET_FLAG() \
diff --git a/nptl/sysdeps/pthread/aio_misc.h b/nptl/sysdeps/pthread/aio_misc.h
index 25ab74e272..f36825e06f 100644
--- a/nptl/sysdeps/pthread/aio_misc.h
+++ b/nptl/sysdeps/pthread/aio_misc.h
@@ -30,7 +30,7 @@
 #define AIO_MISC_NOTIFY(waitlist) \
   do {									      \
     if (*waitlist->counterp > 0 && --*waitlist->counterp == 0)		      \
-      lll_private_futex_wake (waitlist->counterp, 1);			      \
+      lll_futex_wake (waitlist->counterp, 1, LLL_PRIVATE);		      \
   } while (0)
 
 #define AIO_MISC_WAIT(result, futex, timeout, cancel)			      \
@@ -49,8 +49,8 @@
 	int status;							      \
 	do								      \
 	  {								      \
-	    status = lll_private_futex_timed_wait (futexaddr, oldval,	      \
-						   timeout);		      \
+	    status = lll_futex_timed_wait (futexaddr, oldval, timeout,	      \
+					   LLL_PRIVATE);		      \
 	    if (status != -EWOULDBLOCK)					      \
 	      break;							      \
 									      \
diff --git a/nptl/sysdeps/pthread/gai_misc.h b/nptl/sysdeps/pthread/gai_misc.h
index 5f6990040a..0a2686cb27 100644
--- a/nptl/sysdeps/pthread/gai_misc.h
+++ b/nptl/sysdeps/pthread/gai_misc.h
@@ -31,7 +31,7 @@
 #define GAI_MISC_NOTIFY(waitlist) \
   do {									      \
     if (*waitlist->counterp > 0 && --*waitlist->counterp == 0)		      \
-      lll_private_futex_wake (waitlist->counterp, 1);			      \
+      lll_futex_wake (waitlist->counterp, 1, LLL_PRIVATE);		      \
   } while (0)
 
 #define GAI_MISC_WAIT(result, futex, timeout, cancel) \
@@ -50,8 +50,8 @@
 	int status;							      \
 	do								      \
 	  {								      \
-	    status = lll_private_futex_timed_wait (futexaddr, oldval,	      \
-						   timeout);		      \
+	    status = lll_futex_timed_wait (futexaddr, oldval, timeout,	      \
+					   LLL_PRIVATE);		      \
 	    if (status != -EWOULDBLOCK)					      \
 	      break;							      \
 									      \
diff --git a/nptl/sysdeps/s390/tls.h b/nptl/sysdeps/s390/tls.h
index f10db8f227..3be459e32c 100644
--- a/nptl/sysdeps/s390/tls.h
+++ b/nptl/sysdeps/s390/tls.h
@@ -183,7 +183,7 @@ typedef struct
 	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
 			       THREAD_GSCOPE_FLAG_UNUSED);		     \
       if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
-	lll_private_futex_wake (&THREAD_SELF->header.gscope_flag, 1);	     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1, LLL_PRIVATE);   \
     }									     \
   while (0)
 #define THREAD_GSCOPE_SET_FLAG() \
diff --git a/nptl/sysdeps/sh/tls.h b/nptl/sysdeps/sh/tls.h
index e07e29f786..a2d4d565ff 100644
--- a/nptl/sysdeps/sh/tls.h
+++ b/nptl/sysdeps/sh/tls.h
@@ -164,7 +164,7 @@ typedef struct
 	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
 			       THREAD_GSCOPE_FLAG_UNUSED);		     \
       if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
-	lll_private_futex_wake (&THREAD_SELF->header.gscope_flag, 1);	     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1, LLL_PRIVATE);   \
     }									     \
   while (0)
 #define THREAD_GSCOPE_SET_FLAG() \
diff --git a/nptl/sysdeps/sparc/tls.h b/nptl/sysdeps/sparc/tls.h
index c4741c8fc8..601b53732b 100644
--- a/nptl/sysdeps/sparc/tls.h
+++ b/nptl/sysdeps/sparc/tls.h
@@ -157,7 +157,7 @@ register struct pthread *__thread_self __asm__("%g7");
 	= atomic_exchange_rel (&THREAD_SELF->header.gscope_flag,	     \
 			       THREAD_GSCOPE_FLAG_UNUSED);		     \
       if (__res == THREAD_GSCOPE_FLAG_WAIT)				     \
-	lll_private_futex_wake (&THREAD_SELF->header.gscope_flag, 1);	     \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1, LLL_PRIVATE);   \
     }									     \
   while (0)
 #define THREAD_GSCOPE_SET_FLAG() \
diff --git a/nptl/sysdeps/unix/sysv/linux/alpha/lowlevellock.h b/nptl/sysdeps/unix/sysv/linux/alpha/lowlevellock.h
index 04ac006400..5f08673c43 100644
--- a/nptl/sysdeps/unix/sysv/linux/alpha/lowlevellock.h
+++ b/nptl/sysdeps/unix/sysv/linux/alpha/lowlevellock.h
@@ -36,34 +36,63 @@
 #define FUTEX_LOCK_PI		6
 #define FUTEX_UNLOCK_PI		7
 #define FUTEX_TRYLOCK_PI	8
+#define FUTEX_PRIVATE_FLAG	128
+
+/* Values for 'private' parameter of locking macros.  Yes, the
+   definition seems to be backwards.  But it is not.  The bit will be
+   reversed before passing to the system call.  */
+#define LLL_PRIVATE	0
+#define LLL_SHARED	FUTEX_PRIVATE_FLAG
+
+
+#if !defined NOT_IN_libc || defined IS_IN_rtld
+/* In libc.so or ld.so all futexes are private.  */
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  ((fl) | FUTEX_PRIVATE_FLAG)
+# else
+#  define __lll_private_flag(fl, private) \
+  ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))
+# endif
+#else
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  (((fl) | FUTEX_PRIVATE_FLAG) ^ (private))
+# else
+#  define __lll_private_flag(fl, private) \
+  (__builtin_constant_p (private)					      \
+   ? ((private) == 0							      \
+      ? ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))	      \
+      : (fl))								      \
+   : ((fl) | (((private) ^ FUTEX_PRIVATE_FLAG)				      \
+	      & THREAD_GETMEM (THREAD_SELF, header.private_futex))))
+# endif	      
+#endif
+
 
 /* Initializer for compatibility lock.	*/
 #define LLL_MUTEX_LOCK_INITIALIZER (0)
 
-#define lll_futex_wait(futexp, val) \
-  ({									      \
-    INTERNAL_SYSCALL_DECL (__err);					      \
-    long int __ret;							      \
-    __ret = INTERNAL_SYSCALL (futex, __err, 4,				      \
-			      (futexp), FUTEX_WAIT, (val), 0);		      \
-    INTERNAL_SYSCALL_ERROR_P (__ret, __err)? -__ret : __ret;		      \
-  })
+#define lll_futex_wait(futexp, val, private) \
+  lll_futex_timed_wait (futexp, val, NULL, private)
 
-#define lll_futex_timed_wait(futexp, val, timespec) \
+#define lll_futex_timed_wait(futexp, val, timespec, private) \
   ({									      \
     INTERNAL_SYSCALL_DECL (__err);					      \
     long int __ret;							      \
-    __ret = INTERNAL_SYSCALL (futex, __err, 4,				      \
-			      (futexp), FUTEX_WAIT, (val), (timespec));	      \
+    __ret = INTERNAL_SYSCALL (futex, __err, 4, (futexp),		      \
+			      __lll_private_flag (FUTEX_WAIT, private),	      \
+			      (val), (timespec));			      \
     INTERNAL_SYSCALL_ERROR_P (__ret, __err)? -__ret : __ret;		      \
   })
 
-#define lll_futex_wake(futexp, nr) \
+#define lll_futex_wake(futexp, nr, private) \
   ({									      \
     INTERNAL_SYSCALL_DECL (__err);					      \
     long int __ret;							      \
-    __ret = INTERNAL_SYSCALL (futex, __err, 4,				      \
-			      (futexp), FUTEX_WAKE, (nr), 0);		      \
+    __ret = INTERNAL_SYSCALL (futex, __err, 4, (futexp),		      \
+			      __lll_private_flag (FUTEX_WAKE, private),	      \
+			      (nr), 0);					      \
     INTERNAL_SYSCALL_ERROR_P (__ret, __err)? -__ret : __ret;		      \
   })
 
@@ -72,7 +101,7 @@
     {									      \
       int *__futexp = &(futexv);					      \
       atomic_or (__futexp, FUTEX_OWNER_DIED);				      \
-      lll_futex_wake (__futexp, 1);					      \
+      lll_futex_wake (__futexp, 1, LLL_SHARED);				      \
     }									      \
   while (0)
 
@@ -198,7 +227,7 @@ __lll_mutex_unlock (int *futex)
 {
   int val = atomic_exchange_rel (futex, 0);
   if (__builtin_expect (val > 1, 0))
-    lll_futex_wake (futex, 1);
+    lll_futex_wake (futex, 1, LLL_SHARED);
 }
 #define lll_mutex_unlock(futex) __lll_mutex_unlock(&(futex))
 
@@ -208,7 +237,7 @@ __lll_robust_mutex_unlock (int *futex, int mask)
 {
   int val = atomic_exchange_rel (futex, 0);
   if (__builtin_expect (val & mask, 0))
-    lll_futex_wake (futex, 1);
+    lll_futex_wake (futex, 1, LLL_SHARED);
 }
 #define lll_robust_mutex_unlock(futex) \
   __lll_robust_mutex_unlock(&(futex), FUTEX_WAITERS)
@@ -218,7 +247,7 @@ static inline void __attribute__ ((always_inline))
 __lll_mutex_unlock_force (int *futex)
 {
   (void) atomic_exchange_rel (futex, 0);
-  lll_futex_wake (futex, 1);
+  lll_futex_wake (futex, 1, LLL_SHARED);
 }
 #define lll_mutex_unlock_force(futex) __lll_mutex_unlock_force(&(futex))
 
@@ -252,10 +281,10 @@ typedef int lll_lock_t;
    thread ID while the clone is running and is reset to zero
    afterwards.	*/
 #define lll_wait_tid(tid) \
-  do {					\
-    __typeof (tid) __tid;		\
-    while ((__tid = (tid)) != 0)	\
-      lll_futex_wait (&(tid), __tid);	\
+  do {							\
+    __typeof (tid) __tid;				\
+    while ((__tid = (tid)) != 0)			\
+      lll_futex_wait (&(tid), __tid, LLL_SHARED);	\
   } while (0)
 
 extern int __lll_timedwait_tid (int *, const struct timespec *)
diff --git a/nptl/sysdeps/unix/sysv/linux/alpha/pthread_once.c b/nptl/sysdeps/unix/sysv/linux/alpha/pthread_once.c
index 79a3c47aed..0e7e9790dd 100644
--- a/nptl/sysdeps/unix/sysv/linux/alpha/pthread_once.c
+++ b/nptl/sysdeps/unix/sysv/linux/alpha/pthread_once.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 2003, 2004 Free Software Foundation, Inc.
+/* Copyright (C) 2003, 2004, 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
@@ -28,7 +28,7 @@ clear_once_control (void *arg)
   pthread_once_t *once_control = (pthread_once_t *) arg;
 
   *once_control = 0;
-  lll_futex_wake (once_control, INT_MAX);
+  lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
 }
 
 int
@@ -72,7 +72,7 @@ __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
 	break;
 
       /* Same generation, some other thread was faster. Wait.  */
-      lll_futex_wait (once_control, oldval);
+      lll_futex_wait (once_control, oldval, LLL_PRIVATE);
     }
 
   /* This thread is the first here.  Do the initialization.
@@ -88,7 +88,7 @@ __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
   atomic_increment (once_control);
 
   /* Wake up all other threads.  */
-  lll_futex_wake (once_control, INT_MAX);
+  lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
 
   return 0;
 }
diff --git a/nptl/sysdeps/unix/sysv/linux/fork.c b/nptl/sysdeps/unix/sysv/linux/fork.c
index c6dadb5683..f9913c343e 100644
--- a/nptl/sysdeps/unix/sysv/linux/fork.c
+++ b/nptl/sysdeps/unix/sysv/linux/fork.c
@@ -203,7 +203,7 @@ __libc_fork (void)
 
 	  if (atomic_decrement_and_test (&allp->handler->refcntr)
 	      && allp->handler->need_signal)
-	    lll_private_futex_wake (allp->handler->refcntr, 1);
+	    lll_futex_wake (allp->handler->refcntr, 1, LLL_PRIVATE);
 
 	  allp = allp->next;
 	}
diff --git a/nptl/sysdeps/unix/sysv/linux/ia64/lowlevellock.h b/nptl/sysdeps/unix/sysv/linux/ia64/lowlevellock.h
index 17093471d3..095f0e8aca 100644
--- a/nptl/sysdeps/unix/sysv/linux/ia64/lowlevellock.h
+++ b/nptl/sysdeps/unix/sysv/linux/ia64/lowlevellock.h
@@ -36,6 +36,39 @@
 #define FUTEX_LOCK_PI		6
 #define FUTEX_UNLOCK_PI		7
 #define FUTEX_TRYLOCK_PI	8
+#define FUTEX_PRIVATE_FLAG	128
+
+/* Values for 'private' parameter of locking macros.  Yes, the
+   definition seems to be backwards.  But it is not.  The bit will be
+   reversed before passing to the system call.  */
+#define LLL_PRIVATE	0
+#define LLL_SHARED	FUTEX_PRIVATE_FLAG
+
+
+#if !defined NOT_IN_libc || defined IS_IN_rtld
+/* In libc.so or ld.so all futexes are private.  */
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  ((fl) | FUTEX_PRIVATE_FLAG)
+# else
+#  define __lll_private_flag(fl, private) \
+  ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))
+# endif
+#else
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  (((fl) | FUTEX_PRIVATE_FLAG) ^ (private))
+# else
+#  define __lll_private_flag(fl, private) \
+  (__builtin_constant_p (private)					      \
+   ? ((private) == 0							      \
+      ? ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))	      \
+      : (fl))								      \
+   : ((fl) | (((private) ^ FUTEX_PRIVATE_FLAG)				      \
+	      & THREAD_GETMEM (THREAD_SELF, header.private_futex))))
+# endif	      
+#endif
+
 
 /* Delay in spinlock loop.  */
 #define BUSY_WAIT_NOP          asm ("hint @pause")
@@ -43,18 +76,22 @@
 /* Initializer for compatibility lock.	*/
 #define LLL_MUTEX_LOCK_INITIALIZER (0)
 
-#define lll_futex_wait(futex, val) lll_futex_timed_wait (futex, val, 0)
+#define lll_futex_wait(futex, val, private) \
+  lll_futex_timed_wait (futex, val, NULL, private)
 
-#define lll_futex_timed_wait(ftx, val, timespec)			\
+#define lll_futex_timed_wait(ftx, val, timespec, private)		\
 ({									\
-   DO_INLINE_SYSCALL(futex, 4, (long) (ftx), FUTEX_WAIT, (int) (val),	\
-		     (long) (timespec));				\
+   DO_INLINE_SYSCALL(futex, 4, (long) (ftx),				\
+		     __lll_private_flag (FUTEX_WAIT, private),		\
+		     (int) (val), (long) (timespec));			\
    _r10 == -1 ? -_retval : _retval;					\
 })
 
-#define lll_futex_wake(ftx, nr)						\
+#define lll_futex_wake(ftx, nr, private)				\
 ({									\
-   DO_INLINE_SYSCALL(futex, 3, (long) (ftx), FUTEX_WAKE, (int) (nr));	\
+   DO_INLINE_SYSCALL(futex, 3, (long) (ftx),				\
+		     __lll_private_flag (FUTEX_WAKE, private),		\
+		     (int) (nr));					\
    _r10 == -1 ? -_retval : _retval;					\
 })
 
@@ -188,7 +225,7 @@ extern int __lll_robust_timedlock_wait (int *futex, const struct timespec *)
     int __val = atomic_exchange_rel (__futex, 0);	\
 							\
     if (__builtin_expect (__val > 1, 0))		\
-      lll_futex_wake (__futex, 1);			\
+      lll_futex_wake (__futex, 1, LLL_SHARED);		\
   }))
 #define lll_mutex_unlock(futex) \
   __lll_mutex_unlock(&(futex))
@@ -200,7 +237,7 @@ extern int __lll_robust_timedlock_wait (int *futex, const struct timespec *)
     int __val = atomic_exchange_rel (__futex, 0);	\
 							\
     if (__builtin_expect (__val & FUTEX_WAITERS, 0))	\
-      lll_futex_wake (__futex, 1);			\
+      lll_futex_wake (__futex, 1, LLL_SHARED);		\
   }))
 #define lll_robust_mutex_unlock(futex) \
   __lll_robust_mutex_unlock(&(futex))
@@ -210,7 +247,7 @@ extern int __lll_robust_timedlock_wait (int *futex, const struct timespec *)
   ((void) ({					\
     int *__futex = (futex);			\
     (void) atomic_exchange_rel (__futex, 0);	\
-    lll_futex_wake (__futex, 1);		\
+    lll_futex_wake (__futex, 1, LLL_SHARED);	\
   }))
 #define lll_mutex_unlock_force(futex) \
   __lll_mutex_unlock_force(&(futex))
@@ -241,12 +278,12 @@ typedef int lll_lock_t;
    thread ID while the clone is running and is reset to zero
    afterwards.	*/
 #define lll_wait_tid(tid) \
-  do						\
-    {						\
-      __typeof (tid) __tid;			\
-      while ((__tid = (tid)) != 0)		\
-	lll_futex_wait (&(tid), __tid);		\
-    }						\
+  do							\
+    {							\
+      __typeof (tid) __tid;				\
+      while ((__tid = (tid)) != 0)			\
+	lll_futex_wait (&(tid), __tid, LLL_SHARED);	\
+    }							\
   while (0)
 
 extern int __lll_timedwait_tid (int *, const struct timespec *)
diff --git a/nptl/sysdeps/unix/sysv/linux/ia64/pthread_once.c b/nptl/sysdeps/unix/sysv/linux/ia64/pthread_once.c
index 3b07cc127d..22e2dd3c0c 100644
--- a/nptl/sysdeps/unix/sysv/linux/ia64/pthread_once.c
+++ b/nptl/sysdeps/unix/sysv/linux/ia64/pthread_once.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 2003, 2004 Free Software Foundation, Inc.
+/* Copyright (C) 2003, 2004, 2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Jakub Jelinek <jakub@redhat.com>, 2003.
 
@@ -30,7 +30,7 @@ clear_once_control (void *arg)
   pthread_once_t *once_control = (pthread_once_t *) arg;
 
   *once_control = 0;
-  lll_futex_wake (once_control, INT_MAX);
+  lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
 }
 
 
@@ -65,7 +65,7 @@ __pthread_once (once_control, init_routine)
 	  if (((oldval ^ newval) & -4) == 0)
 	    {
 	      /* Same generation, some other thread was faster. Wait.  */
-	      lll_futex_wait (once_control, newval);
+	      lll_futex_wait (once_control, newval, LLL_PRIVATE);
 	      continue;
 	    }
 	}
@@ -84,7 +84,7 @@ __pthread_once (once_control, init_routine)
       atomic_increment (once_control);
 
       /* Wake up all other threads.  */
-      lll_futex_wake (once_control, INT_MAX);
+      lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
       break;
     }
 
diff --git a/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h b/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
index ab795348ee..cf91f21483 100644
--- a/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
+++ b/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
@@ -134,15 +134,6 @@
   })
   
   
-#define lll_private_futex_wait(futexp, val) \
-  lll_futex_timed_wait (futexp, val, NULL, LLL_PRIVATE)
-
-#define lll_private_futex_timed_wait(futexp, val, timeout) \
-  lll_futex_timed_wait (futexp, val, timeout, LLL_PRIVATE)
-
-#define lll_private_futex_wake(futexp, val) \
-  lll_futex_wake (futexp, val, LLL_PRIVATE)
-
 #ifdef UP
 # define __lll_acq_instr	""
 # define __lll_rel_instr	""
diff --git a/nptl/sysdeps/unix/sysv/linux/powerpc/pthread_once.c b/nptl/sysdeps/unix/sysv/linux/powerpc/pthread_once.c
index 3224568fb7..969078094d 100644
--- a/nptl/sysdeps/unix/sysv/linux/powerpc/pthread_once.c
+++ b/nptl/sysdeps/unix/sysv/linux/powerpc/pthread_once.c
@@ -30,7 +30,7 @@ clear_once_control (void *arg)
   pthread_once_t *once_control = (pthread_once_t *) arg;
 
   *once_control = 0;
-  lll_private_futex_wake (once_control, INT_MAX);
+  lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
 }
 
 
@@ -74,7 +74,7 @@ __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
 	break;
 
       /* Same generation, some other thread was faster. Wait.  */
-      lll_private_futex_wait (once_control, oldval);
+      lll_futex_wait (once_control, oldval, LLL_PRIVATE);
     }
 
 
@@ -92,7 +92,7 @@ __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
   atomic_increment (once_control);
 
   /* Wake up all other threads.  */
-  lll_private_futex_wake (once_control, INT_MAX);
+  lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
 
   return 0;
 }
diff --git a/nptl/sysdeps/unix/sysv/linux/rtld-lowlevel.h b/nptl/sysdeps/unix/sysv/linux/rtld-lowlevel.h
index 1aeff8fb39..4a49925588 100644
--- a/nptl/sysdeps/unix/sysv/linux/rtld-lowlevel.h
+++ b/nptl/sysdeps/unix/sysv/linux/rtld-lowlevel.h
@@ -33,12 +33,12 @@
 	int val = word;							      \
 	if (val == 0)							      \
 	  break;							      \
-	lll_private_futex_wait (&(word), val);				      \
+	lll_futex_wait (&(word), val, LLL_PRIVATE);			      \
       }									      \
   } while (0)
 
 
 #define __rtld_notify(word) \
-  lll_private_futex_wake (&(word), 1)
+  lll_futex_wake (&(word), 1, LLL_PRIVATE)
 
 #endif
diff --git a/nptl/sysdeps/unix/sysv/linux/s390/lowlevellock.h b/nptl/sysdeps/unix/sysv/linux/s390/lowlevellock.h
index d915facba1..4758b63bd0 100644
--- a/nptl/sysdeps/unix/sysv/linux/s390/lowlevellock.h
+++ b/nptl/sysdeps/unix/sysv/linux/s390/lowlevellock.h
@@ -35,31 +35,50 @@
 #define FUTEX_LOCK_PI		6
 #define FUTEX_UNLOCK_PI		7
 #define FUTEX_TRYLOCK_PI	8
+#define FUTEX_PRIVATE_FLAG	128
+
+/* Values for 'private' parameter of locking macros.  Yes, the
+   definition seems to be backwards.  But it is not.  The bit will be
+   reversed before passing to the system call.  */
+#define LLL_PRIVATE	0
+#define LLL_SHARED	FUTEX_PRIVATE_FLAG
+
+
+#if !defined NOT_IN_libc || defined IS_IN_rtld
+/* In libc.so or ld.so all futexes are private.  */
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  ((fl) | FUTEX_PRIVATE_FLAG)
+# else
+#  define __lll_private_flag(fl, private) \
+  ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))
+# endif
+#else
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  (((fl) | FUTEX_PRIVATE_FLAG) ^ (private))
+# else
+#  define __lll_private_flag(fl, private) \
+  (__builtin_constant_p (private)					      \
+   ? ((private) == 0							      \
+      ? ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))	      \
+      : (fl))								      \
+   : ((fl) | (((private) ^ FUTEX_PRIVATE_FLAG)				      \
+	      & THREAD_GETMEM (THREAD_SELF, header.private_futex))))
+# endif	      
+#endif
 
 /* Initializer for compatibility lock.	*/
 #define LLL_MUTEX_LOCK_INITIALIZER (0)
 
-#define lll_futex_wait(futex, val) \
-  ({									      \
-     register unsigned long int __r2 asm ("2") = (unsigned long int) (futex); \
-     register unsigned long int __r3 asm ("3") = FUTEX_WAIT;		      \
-     register unsigned long int __r4 asm ("4") = (unsigned long int) (val);   \
-     register unsigned long int __r5 asm ("5") = 0ul;			      \
-     register unsigned long __result asm ("2");				      \
-									      \
-    __asm __volatile ("svc %b1"						      \
-		      : "=d" (__result)					      \
-		      : "i" (SYS_futex), "0" (__r2), "d" (__r3),	      \
-			"d" (__r4), "d" (__r5)				      \
-		      : "cc", "memory" );				      \
-    __result;								      \
-  })
-
+#define lll_futex_wait(futex, val, private) \
+  lll_futex_timed_wait (futex, val, NULL, private)
 
-#define lll_futex_timed_wait(futex, val, timespec) \
+#define lll_futex_timed_wait(futex, val, timespec, private) \
   ({									      \
     register unsigned long int __r2 asm ("2") = (unsigned long int) (futex);  \
-    register unsigned long int __r3 asm ("3") = FUTEX_WAIT;		      \
+    register unsigned long int __r3 asm ("3")				      \
+      = __lll_private_flag (FUTEX_WAIT, private);			      \
     register unsigned long int __r4 asm ("4") = (unsigned long int) (val);    \
     register unsigned long int __r5 asm ("5") = (unsigned long int)(timespec);\
     register unsigned long int __result asm ("2");			      \
@@ -73,10 +92,11 @@
   })
 
 
-#define lll_futex_wake(futex, nr) \
+#define lll_futex_wake(futex, nr, private) \
   ({									      \
     register unsigned long int __r2 asm ("2") = (unsigned long int) (futex);  \
-    register unsigned long int __r3 asm ("3") = FUTEX_WAKE;		      \
+    register unsigned long int __r3 asm ("3")				      \
+      __lll_private_flag (FUTEX_WAKE, private);				      \
     register unsigned long int __r4 asm ("4") = (unsigned long int) (nr);     \
     register unsigned long int __result asm ("2");			      \
 									      \
@@ -94,7 +114,7 @@
       int *__futexp = &(futexv);					      \
 									      \
       atomic_or (__futexp, FUTEX_OWNER_DIED);				      \
-      lll_futex_wake (__futexp, 1);					      \
+      lll_futex_wake (__futexp, 1, LLL_SHARED);				      \
     }									      \
   while (0)
 
@@ -271,7 +291,7 @@ __lll_mutex_unlock (int *futex)
 
   lll_compare_and_swap (futex, oldval, newval, "slr %2,%2");
   if (oldval > 1)
-    lll_futex_wake (futex, 1);
+    lll_futex_wake (futex, 1, LLL_SHARED);
 }
 #define lll_mutex_unlock(futex) \
   __lll_mutex_unlock(&(futex))
@@ -286,7 +306,7 @@ __lll_robust_mutex_unlock (int *futex, int mask)
 
   lll_compare_and_swap (futex, oldval, newval, "slr %2,%2");
   if (oldval & mask)
-    lll_futex_wake (futex, 1);
+    lll_futex_wake (futex, 1, LLL_SHARED);
 }
 #define lll_robust_mutex_unlock(futex) \
   __lll_robust_mutex_unlock(&(futex), FUTEX_WAITERS)
@@ -297,7 +317,7 @@ __attribute__ ((always_inline))
 __lll_mutex_unlock_force (int *futex)
 {
   *futex = 0;
-  lll_futex_wake (futex, 1);
+  lll_futex_wake (futex, 1, LLL_SHARED);
 }
 #define lll_mutex_unlock_force(futex) \
   __lll_mutex_unlock_force(&(futex))
@@ -338,7 +358,7 @@ __lll_wait_tid (int *ptid)
   int tid;
 
   while ((tid = *ptid) != 0)
-    lll_futex_wait (ptid, tid);
+    lll_futex_wait (ptid, tid, LLL_SHARED);
 }
 #define lll_wait_tid(tid) __lll_wait_tid(&(tid))
 
diff --git a/nptl/sysdeps/unix/sysv/linux/s390/pthread_once.c b/nptl/sysdeps/unix/sysv/linux/s390/pthread_once.c
index f29e23fd4b..0012e9ae27 100644
--- a/nptl/sysdeps/unix/sysv/linux/s390/pthread_once.c
+++ b/nptl/sysdeps/unix/sysv/linux/s390/pthread_once.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 2003 Free Software Foundation, Inc.
+/* Copyright (C) 2003, 2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Martin Schwidefsky <schwidefsky@de.ibm.com>, 2003.
 
@@ -30,7 +30,7 @@ clear_once_control (void *arg)
   pthread_once_t *once_control = (pthread_once_t *) arg;
 
   *once_control = 0;
-  lll_futex_wake (once_control, INT_MAX);
+  lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
 }
 
 
@@ -76,7 +76,7 @@ __pthread_once (once_control, init_routine)
 	  if (((oldval ^ newval) & -4) == 0)
 	    {
 	      /* Same generation, some other thread was faster. Wait.  */
-	      lll_futex_wait (once_control, newval);
+	      lll_futex_wait (once_control, newval, LLL_PRIVATE);
 	      continue;
 	    }
 	}
@@ -101,7 +101,7 @@ __pthread_once (once_control, init_routine)
 			: "m" (*once_control) : "cc" );
 
       /* Wake up all other threads.  */
-      lll_futex_wake (once_control, INT_MAX);
+      lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
       break;
     }
 
diff --git a/nptl/sysdeps/unix/sysv/linux/sh/lowlevellock.h b/nptl/sysdeps/unix/sysv/linux/sh/lowlevellock.h
index 3092b27642..be47bd752e 100644
--- a/nptl/sysdeps/unix/sysv/linux/sh/lowlevellock.h
+++ b/nptl/sysdeps/unix/sysv/linux/sh/lowlevellock.h
@@ -40,6 +40,31 @@
 #define LLL_SHARED     FUTEX_PRIVATE_FLAG
 
 
+#if !defined NOT_IN_libc || defined IS_IN_rtld
+/* In libc.so or ld.so all futexes are private.  */
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  ((fl) | FUTEX_PRIVATE_FLAG)
+# else
+#  define __lll_private_flag(fl, private) \
+  ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))
+# endif
+#else
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  (((fl) | FUTEX_PRIVATE_FLAG) ^ (private))
+# else
+#  define __lll_private_flag(fl, private) \
+  (__builtin_constant_p (private)					      \
+   ? ((private) == 0							      \
+      ? ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))	      \
+      : (fl))								      \
+   : ((fl) | (((private) ^ FUTEX_PRIVATE_FLAG)				      \
+	      & THREAD_GETMEM (THREAD_SELF, header.private_futex))))
+# endif	      
+#endif
+
+
 /* Initializer for compatibility lock.  */
 #define LLL_MUTEX_LOCK_INITIALIZER		(0)
 #define LLL_MUTEX_LOCK_INITIALIZER_LOCKED	(1)
@@ -278,7 +303,7 @@ extern int __lll_mutex_unlock_wake (int *__futex) attribute_hidden;
 	     1: mov r1,r15"\
 		: "=&r" (__ignore) : "r" (__futex), "r" (FUTEX_OWNER_DIED) \
 		: "r0", "r1", "memory");	\
-	    lll_futex_wake (__futex, 1, 0); })
+	    lll_futex_wake (__futex, 1, LLL_SHARED); })
 
 #define lll_mutex_islocked(futex) \
   (futex != 0)
@@ -312,7 +337,8 @@ typedef int lll_lock_t;
     int __status;							      \
     register unsigned long __r3 asm ("r3") = SYS_futex;			      \
     register unsigned long __r4 asm ("r4") = (unsigned long) (futex);	      \
-    register unsigned long __r5 asm ("r5") = FUTEX_WAIT;		      \
+    register unsigned long __r5 asm ("r5")				      \
+      = __lll_private_flag (FUTEX_WAIT, private);			      \
     register unsigned long __r6 asm ("r6") = (unsigned long) (val);	      \
     register unsigned long __r7 asm ("r7") = (timeout);			      \
     __asm __volatile (SYSCALL_WITH_INST_PAD				      \
@@ -329,7 +355,8 @@ typedef int lll_lock_t;
     int __ignore;							      \
     register unsigned long __r3 asm ("r3") = SYS_futex;			      \
     register unsigned long __r4 asm ("r4") = (unsigned long) (futex);	      \
-    register unsigned long __r5 asm ("r5") = FUTEX_WAKE;		      \
+    register unsigned long __r5 asm ("r5")				      \
+      = __lll_private_flag (FUTEX_WAKE, private);			      \
     register unsigned long __r6 asm ("r6") = (unsigned long) (nr);	      \
     register unsigned long __r7 asm ("r7") = 0;				      \
     __asm __volatile (SYSCALL_WITH_INST_PAD				      \
@@ -340,81 +367,6 @@ typedef int lll_lock_t;
   } while (0)
 
 
-#define lll_private_futex_wait(futex, val) \
-  lll_private_futex_timed_wait (futex, val, NULL)
-
-
-#ifdef __ASSUME_PRIVATE_FUTEX
-# define lll_private_futex_timed_wait(futex, val, timeout) \
-  ({									      \
-    int __status;							      \
-    register unsigned long __r3 asm ("r3") = SYS_futex;			      \
-    register unsigned long __r4 asm ("r4") = (unsigned long) (futex);	      \
-    register unsigned long __r5 asm ("r5") = FUTEX_WAIT | FUTEX_PRIVATE_FLAG; \
-    register unsigned long __r6 asm ("r6") = (unsigned long) (val);	      \
-    register unsigned long __r7 asm ("r7") = (timeout);			      \
-    __asm __volatile (SYSCALL_WITH_INST_PAD				      \
-		      : "=z" (__status)					      \
-		      : "r" (__r3), "r" (__r4), "r" (__r5),		      \
-			"r" (__r6), "r" (__r7)				      \
-		      : "memory", "t");					      \
-    __status;								      \
-  })
-
-
-# define lll_private_futex_wake(futex, nr) \
-  do {									      \
-    int __ignore;							      \
-    register unsigned long __r3 asm ("r3") = SYS_futex;			      \
-    register unsigned long __r4 asm ("r4") = (unsigned long) (futex);	      \
-    register unsigned long __r5 asm ("r5") = FUTEX_WAKE | FUTEX_PRIVATE_FLAG; \
-    register unsigned long __r6 asm ("r6") = (unsigned long) (nr);	      \
-    register unsigned long __r7 asm ("r7") = 0;				      \
-    __asm __volatile (SYSCALL_WITH_INST_PAD				      \
-		      : "=z" (__ignore)					      \
-		      : "r" (__r3), "r" (__r4), "r" (__r5),		      \
-			"r" (__r6), "r" (__r7)				      \
-		      : "memory", "t");					      \
-  } while (0)
-
-
-#else
-# define lll_private_futex_timed_wait(futex, val, timeout) \
-  ({									      \
-    int __status;							      \
-    register unsigned long __r3 asm ("r3") = SYS_futex;			      \
-    register unsigned long __r4 asm ("r4") = (unsigned long) (futex);	      \
-    register unsigned long __r5 asm ("r5");				      \
-    register unsigned long __r6 asm ("r6") = (unsigned long) (val);	      \
-    register unsigned long __r7 asm ("r7") = (timeout);			      \
-    __r5 = THREAD_GETMEM (THREAD_SELF, header.private_futex);		      \
-    __asm __volatile (SYSCALL_WITH_INST_PAD				      \
-		      : "=z" (__status)					      \
-		      : "r" (__r3), "r" (__r4), "r" (__r5),		      \
-			"r" (__r6), "r" (__r7)				      \
-		      : "memory", "t");					      \
-    __status;								      \
-  })
-
-
-# define lll_private_futex_wake(futex, nr) \
-  do {									      \
-    int __ignore;							      \
-    register unsigned long __r3 asm ("r3") = SYS_futex;			      \
-    register unsigned long __r4 asm ("r4") = (unsigned long) (futex);	      \
-    register unsigned long __r5 asm ("r5") = FUTEX_WAKE;		      \
-    register unsigned long __r6 asm ("r6") = (unsigned long) (nr);	      \
-    register unsigned long __r7 asm ("r7") = 0;				      \
-    __r5 |= THREAD_GETMEM (THREAD_SELF,	header.private_futex);		      \
-    __asm __volatile (SYSCALL_WITH_INST_PAD				      \
-		      : "=z" (__ignore)					      \
-		      : "r" (__r3), "r" (__r4), "r" (__r5),		      \
-			"r" (__r6), "r" (__r7)				      \
-		      : "memory", "t");					      \
-  } while (0)
-#endif
-
-
 /* The states of a lock are:
     0  -  untaken
     1  -  taken by one user
@@ -438,7 +390,7 @@ extern int __lll_wait_tid (int *tid) attribute_hidden;
   do {									      \
     __typeof (tid) *__tid = &(tid);					      \
     while (*__tid != 0)							      \
-      lll_futex_wait (__tid, *__tid, 0);				      \
+      lll_futex_wait (__tid, *__tid, LLL_SHARED);			      \
   } while (0)
 
 extern int __lll_timedwait_tid (int *tid, const struct timespec *abstime)
diff --git a/nptl/sysdeps/unix/sysv/linux/sparc/lowlevellock.h b/nptl/sysdeps/unix/sysv/linux/sparc/lowlevellock.h
index 6548970663..922e3c21d3 100644
--- a/nptl/sysdeps/unix/sysv/linux/sparc/lowlevellock.h
+++ b/nptl/sysdeps/unix/sysv/linux/sparc/lowlevellock.h
@@ -35,37 +35,66 @@
 #define FUTEX_LOCK_PI		6
 #define FUTEX_UNLOCK_PI		7
 #define FUTEX_TRYLOCK_PI	8
+#define FUTEX_PRIVATE_FLAG	128
+
+
+/* Values for 'private' parameter of locking macros.  Yes, the
+   definition seems to be backwards.  But it is not.  The bit will be
+   reversed before passing to the system call.  */
+#define LLL_PRIVATE	0
+#define LLL_SHARED	FUTEX_PRIVATE_FLAG
+
+
+#if !defined NOT_IN_libc || defined IS_IN_rtld
+/* In libc.so or ld.so all futexes are private.  */
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  ((fl) | FUTEX_PRIVATE_FLAG)
+# else
+#  define __lll_private_flag(fl, private) \
+  ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))
+# endif
+#else
+# ifdef __ASSUME_PRIVATE_FUTEX
+#  define __lll_private_flag(fl, private) \
+  (((fl) | FUTEX_PRIVATE_FLAG) ^ (private))
+# else
+#  define __lll_private_flag(fl, private) \
+  (__builtin_constant_p (private)					      \
+   ? ((private) == 0							      \
+      ? ((fl) | THREAD_GETMEM (THREAD_SELF, header.private_futex))	      \
+      : (fl))								      \
+   : ((fl) | (((private) ^ FUTEX_PRIVATE_FLAG)				      \
+	      & THREAD_GETMEM (THREAD_SELF, header.private_futex))))
+# endif	      
+#endif
+
 
 /* Initializer for compatibility lock.	*/
 #define LLL_MUTEX_LOCK_INITIALIZER (0)
 
-#define lll_futex_wait(futexp, val) \
-  ({									      \
-    INTERNAL_SYSCALL_DECL (__err);					      \
-    long int __ret;							      \
-									      \
-    __ret = INTERNAL_SYSCALL (futex, __err, 4,				      \
-			      (futexp), FUTEX_WAIT, (val), 0);		      \
-    __ret;								      \
-  })
+#define lll_futex_wait(futexp, val, private) \
+  lll_futex_timed_wait (futexp, val, NULL, private)
 
-#define lll_futex_timed_wait(futexp, val, timespec) \
+#define lll_futex_timed_wait(futexp, val, timespec, private) \
   ({									      \
     INTERNAL_SYSCALL_DECL (__err);					      \
     long int __ret;							      \
 									      \
-    __ret = INTERNAL_SYSCALL (futex, __err, 4,				      \
-			      (futexp), FUTEX_WAIT, (val), (timespec));	      \
+    __ret = INTERNAL_SYSCALL (futex, __err, 4, (futexp),		      \
+			      __lll_private_flag (FUTEX_WAIT, private),	      \
+			      (val), (timespec));			      \
     __ret;								      \
   })
 
-#define lll_futex_wake(futexp, nr) \
+#define lll_futex_wake(futexp, nr, private) \
   ({									      \
     INTERNAL_SYSCALL_DECL (__err);					      \
     long int __ret;							      \
 									      \
-    __ret = INTERNAL_SYSCALL (futex, __err, 4,				      \
-			      (futexp), FUTEX_WAKE, (nr), 0);		      \
+    __ret = INTERNAL_SYSCALL (futex, __err, 4, (futexp),		      \
+			      __lll_private_flag (FUTEX_WAKE, private),	      \
+			      (nr), 0);					      \
     __ret;								      \
   })
 
@@ -86,7 +115,7 @@
     {									      \
       int *__futexp = &(futexv);					      \
       atomic_or (__futexp, FUTEX_OWNER_DIED);				      \
-      lll_futex_wake (__futexp, 1);					      \
+      lll_futex_wake (__futexp, 1, LLL_SHARED);				      \
     }									      \
   while (0)
 
@@ -212,7 +241,7 @@ __lll_robust_mutex_timedlock (int *futex, const struct timespec *abstime,
     int *__futex = &(lock);						      \
     int __val = atomic_exchange_24_rel (__futex, 0);			      \
     if (__builtin_expect (__val > 1, 0))				      \
-      lll_futex_wake (__futex, 1);					      \
+      lll_futex_wake (__futex, 1, LLL_SHARED);				      \
   }))
 
 #define lll_robust_mutex_unlock(lock) \
@@ -220,14 +249,14 @@ __lll_robust_mutex_timedlock (int *futex, const struct timespec *abstime,
     int *__futex = &(lock);						      \
     int __val = atomic_exchange_rel (__futex, 0);			      \
     if (__builtin_expect (__val & FUTEX_WAITERS, 0))			      \
-      lll_futex_wake (__futex, 1);					      \
+      lll_futex_wake (__futex, 1, LLL_SHARED);				      \
   }))
 
 #define lll_mutex_unlock_force(lock) \
   ((void) ({								      \
     int *__futex = &(lock);						      \
     (void) atomic_exchange_24_rel (__futex, 0);				      \
-    lll_futex_wake (__futex, 1);					      \
+    lll_futex_wake (__futex, 1, LLL_SHARED);				      \
   }))
 
 #define lll_mutex_islocked(futex) \
@@ -255,12 +284,12 @@ typedef int lll_lock_t;
    thread ID while the clone is running and is reset to zero
    afterwards.	*/
 #define lll_wait_tid(tid) \
-  do						\
-    {						\
-      __typeof (tid) __tid;			\
-      while ((__tid = (tid)) != 0)		\
-	lll_futex_wait (&(tid), __tid);		\
-    }						\
+  do							\
+    {							\
+      __typeof (tid) __tid;				\
+      while ((__tid = (tid)) != 0)			\
+	lll_futex_wait (&(tid), __tid, LLL_SHARED);	\
+    }							\
   while (0)
 
 extern int __lll_timedwait_tid (int *, const struct timespec *)
diff --git a/nptl/sysdeps/unix/sysv/linux/sparc/pthread_once.c b/nptl/sysdeps/unix/sysv/linux/sparc/pthread_once.c
index 83fedef8e8..22e2dd3c0c 100644
--- a/nptl/sysdeps/unix/sysv/linux/sparc/pthread_once.c
+++ b/nptl/sysdeps/unix/sysv/linux/sparc/pthread_once.c
@@ -30,7 +30,7 @@ clear_once_control (void *arg)
   pthread_once_t *once_control = (pthread_once_t *) arg;
 
   *once_control = 0;
-  lll_private_futex_wake (once_control, INT_MAX);
+  lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
 }
 
 
@@ -65,7 +65,7 @@ __pthread_once (once_control, init_routine)
 	  if (((oldval ^ newval) & -4) == 0)
 	    {
 	      /* Same generation, some other thread was faster. Wait.  */
-	      lll_futex_wait (once_control, newval);
+	      lll_futex_wait (once_control, newval, LLL_PRIVATE);
 	      continue;
 	    }
 	}
@@ -84,7 +84,7 @@ __pthread_once (once_control, init_routine)
       atomic_increment (once_control);
 
       /* Wake up all other threads.  */
-      lll_private_futex_wake (once_control, INT_MAX);
+      lll_futex_wake (once_control, INT_MAX, LLL_PRIVATE);
       break;
     }
 
diff --git a/nptl/sysdeps/unix/sysv/linux/unregister-atfork.c b/nptl/sysdeps/unix/sysv/linux/unregister-atfork.c
index 240ce597f9..bc5b7537a3 100644
--- a/nptl/sysdeps/unix/sysv/linux/unregister-atfork.c
+++ b/nptl/sysdeps/unix/sysv/linux/unregister-atfork.c
@@ -104,7 +104,7 @@ __unregister_atfork (dso_handle)
       atomic_decrement (&deleted->handler->refcntr);
       unsigned int val;
       while ((val = deleted->handler->refcntr) != 0)
-	lll_private_futex_wait (&deleted->handler->refcntr, val);
+	lll_futex_wait (&deleted->handler->refcntr, val, LLL_PRIVATE);
 
       deleted = deleted->next;
     }
diff --git a/nptl/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h b/nptl/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
index 6323d4850d..7ec7deff17 100644
--- a/nptl/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
+++ b/nptl/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
@@ -71,8 +71,8 @@
       : (fl))								      \
    : ({ unsigned int __fl = ((private) ^ FUTEX_PRIVATE_FLAG);		      \
 	asm ("andl %%fs:%P1, %0" : "+r" (__fl)				      \
-	     : "i" offsetof (struct pthread, header.private_futex));	      \
-	__fl | (fl); })
+	     : "i" (offsetof (struct pthread, header.private_futex)));	      \
+	__fl | (fl); }))
 # endif	      
 #endif
 
@@ -215,15 +215,6 @@ LLL_STUB_UNWIND_INFO_END
   } while (0)
 
 
-#define lll_private_futex_wait(futex, val) \
-  lll_futex_timed_wait (futex, val, NULL, LLL_PRIVATE)
-
-#define lll_private_futex_timed_wait(futex, val, timeout) \
-  lll_futex_timed_wait (futex, val, timeout, LLL_PRIVATE)
-
-#define lll_private_futex_wake(futex, nr) \
-  lll_futex_wake (futex, nr, LLL_PRIVATE)
-
 
 /* Does not preserve %eax and %ecx.  */
 extern int __lll_mutex_lock_wait (int *__futex, int __val) attribute_hidden;
diff --git a/nptl/sysdeps/x86_64/tls.h b/nptl/sysdeps/x86_64/tls.h
index 029848a8fa..79db61c709 100644
--- a/nptl/sysdeps/x86_64/tls.h
+++ b/nptl/sysdeps/x86_64/tls.h
@@ -355,7 +355,7 @@ typedef struct
 		    : "i" (offsetof (struct pthread, header.gscope_flag)),    \
 		      "0" (THREAD_GSCOPE_FLAG_UNUSED));			      \
       if (__res == THREAD_GSCOPE_FLAG_WAIT)				      \
-	lll_private_futex_wake (&THREAD_SELF->header.gscope_flag, 1);	      \
+	lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1, LLL_PRIVATE);    \
     }									      \
   while (0)
 #define THREAD_GSCOPE_SET_FLAG() \