* sysdeps/unix/sysv/linux/sh/pthread_rwlock_rdlock.S:

	(__pthread_rwlock_rdlock): Don't use non SH-3/4 instruction.
	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_wrlock.S:
	(__pthread_rwlock_wrlock): Likewise.
	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_timedrdlock.S:
	(pthread_rwlock_timedrdlock): Likewise.
	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_timedwrlock.S:
	(pthread_rwlock_timedwrlock): Likewise.
	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_unlock.S:
	(__pthread_rwlock_unlock): Likewise.

2007-06-10  Kaz Kojima  <kkojima@rr.iij4u.or.jp>

	* sysdeps/sh/tcb-offsets.sym: Add PRIVATE_FUTEX.
	* sysdeps/unix/sysv/linux/sh/bits/pthreadtypes.h: Include endian.h.
	Split __flags into __flags, __shared, __pad1 and __pad2.
	* sysdeps/unix/sysv/linux/sh/libc-lowlevellock.S: Use private
        futexes if they are available.
	* sysdeps/unix/sysv/linux/sh/lowlevellock.S: Adjust so that change
        in libc-lowlevellock.S allow using private futexes.
	* sysdeps/unix/sysv/linux/sh/lowlevellock.h: Define
	FUTEX_PRIVATE_FLAG.  Add additional parameter to lll_futex_wait,
	lll_futex_timed_wait and lll_futex_wake.  Change lll_futex_wait
	to call lll_futex_timed_wait.  Add lll_private_futex_wait,
	lll_private_futex_timed_wait and lll_private_futex_wake.
	(lll_robust_mutex_unlock): Fix typo.
	* sysdeps/unix/sysv/linux/sh/pthread_barrier_wait.S: Use private
        field in futex command setup.
	* sysdeps/unix/sysv/linux/sh/pthread_cond_timedwait.S: Use
	COND_NWAITERS_SHIFT instead of COND_CLOCK_BITS.
	* sysdeps/unix/sysv/linux/sh/pthread_cond_wait.S: Likewise.
	* sysdeps/unix/sysv/linux/sh/pthread_once.S: Use private futexes
        if they are available.  Remove clear_once_control.
	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_rdlock.S: Use private
	futexes if they are available.
	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_timedrdlock.S: Likewise.
	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_timedwrlock.S: Likewise.
	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_unlock.S: Likewise.
	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_wrlock.S: Likewise.
	* sysdeps/unix/sysv/linux/sh/sem_post.S: Add private futex support.
	Wake only when there are waiters.
	* sysdeps/unix/sysv/linux/sh/sem_wait.S: Add private futex
	support.  Indicate that there are waiters.  Remove unnecessary
        extra cancellation test.
	* sysdeps/unix/sysv/linux/sh/sem_timedwait.S: Likewise.  Removed
	left-over duplication of __sem_wait_cleanup.

nptl/ChangeLog

@@ -1,3 +1,52 @@
+2007-06-17  Kaz Kojima  <kkojima@rr.iij4u.or.jp>
+
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_rdlock.S:
+	(__pthread_rwlock_rdlock): Don't use non SH-3/4 instruction.
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_wrlock.S:
+	(__pthread_rwlock_wrlock): Likewise.
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_timedrdlock.S:
+	(pthread_rwlock_timedrdlock): Likewise.
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_timedwrlock.S:
+	(pthread_rwlock_timedwrlock): Likewise.
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_unlock.S:
+	(__pthread_rwlock_unlock): Likewise.
+
+2007-06-10  Kaz Kojima  <kkojima@rr.iij4u.or.jp>
+
+	* sysdeps/sh/tcb-offsets.sym: Add PRIVATE_FUTEX.
+	* sysdeps/unix/sysv/linux/sh/bits/pthreadtypes.h: Include endian.h.
+	Split __flags into __flags, __shared, __pad1 and __pad2.
+	* sysdeps/unix/sysv/linux/sh/libc-lowlevellock.S: Use private
+        futexes if they are available.
+	* sysdeps/unix/sysv/linux/sh/lowlevellock.S: Adjust so that change
+        in libc-lowlevellock.S allow using private futexes.
+	* sysdeps/unix/sysv/linux/sh/lowlevellock.h: Define
+	FUTEX_PRIVATE_FLAG.  Add additional parameter to lll_futex_wait,
+	lll_futex_timed_wait and lll_futex_wake.  Change lll_futex_wait
+	to call lll_futex_timed_wait.  Add lll_private_futex_wait,
+	lll_private_futex_timed_wait and lll_private_futex_wake.
+	(lll_robust_mutex_unlock): Fix typo.
+	* sysdeps/unix/sysv/linux/sh/pthread_barrier_wait.S: Use private
+        field in futex command setup.
+	* sysdeps/unix/sysv/linux/sh/pthread_cond_timedwait.S: Use
+	COND_NWAITERS_SHIFT instead of COND_CLOCK_BITS.
+	* sysdeps/unix/sysv/linux/sh/pthread_cond_wait.S: Likewise.
+	* sysdeps/unix/sysv/linux/sh/pthread_once.S: Use private futexes
+        if they are available.  Remove clear_once_control.
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_rdlock.S: Use private
+	futexes if they are available.
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_timedrdlock.S: Likewise.
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_timedwrlock.S: Likewise.
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_unlock.S: Likewise.
+	* sysdeps/unix/sysv/linux/sh/pthread_rwlock_wrlock.S: Likewise.
+	* sysdeps/unix/sysv/linux/sh/sem_post.S: Add private futex support.
+	Wake only when there are waiters.
+	* sysdeps/unix/sysv/linux/sh/sem_wait.S: Add private futex
+	support.  Indicate that there are waiters.  Remove unnecessary
+        extra cancellation test.
+	* sysdeps/unix/sysv/linux/sh/sem_timedwait.S: Likewise.  Removed
+	left-over duplication of __sem_wait_cleanup.
+
 2007-06-07  Ulrich Drepper  <drepper@redhat.com>
 
 	* sysdeps/unix/sysv/linux/x86_64/lowlevellock.h: Add additional

nptl/sysdeps/sh/tcb-offsets.sym

@@ -10,3 +10,6 @@ MULTIPLE_THREADS_OFFSET	offsetof (struct pthread, header.multiple_threads)
 TLS_PRE_TCB_SIZE	sizeof (struct pthread)
 MUTEX_FUTEX		offsetof (pthread_mutex_t, __data.__lock)
 POINTER_GUARD		offsetof (tcbhead_t, pointer_guard)
+#ifndef __ASSUME_PRIVATE_FUTEX
+PRIVATE_FUTEX		offsetof (struct pthread, header.private_futex)
+#endif

nptl/sysdeps/unix/sysv/linux/sh/bits/pthreadtypes.h

@@ -1,4 +1,5 @@
-/* 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.
 
@@ -20,6 +21,8 @@
 #ifndef _BITS_PTHREADTYPES_H
 #define _BITS_PTHREADTYPES_H	1
 
+#include <endian.h>
+
 #define __SIZEOF_PTHREAD_ATTR_T 36
 #define __SIZEOF_PTHREAD_MUTEX_T 24
 #define __SIZEOF_PTHREAD_MUTEXATTR_T 4
@@ -127,9 +130,21 @@ typedef union
     unsigned int __writer_wakeup;
     unsigned int __nr_readers_queued;
     unsigned int __nr_writers_queued;
+#if __BYTE_ORDER == __BIG_ENDIAN
+    unsigned char __pad1;
+    unsigned char __pad2;
+    unsigned char __shared;
+    /* FLAGS must stay at this position in the structure to maintain
+       binary compatibility.  */
+    unsigned char __flags;
+#else
     /* FLAGS must stay at this position in the structure to maintain
        binary compatibility.  */
-    unsigned int __flags;
+    unsigned char __flags;
+    unsigned char __shared;
+    unsigned char __pad1;
+    unsigned char __pad2;
+#endif
     pthread_t __writer;
   } __data;
   char __size[__SIZEOF_PTHREAD_RWLOCK_T];

nptl/sysdeps/unix/sysv/linux/sh/libc-lowlevellock.S

@@ -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.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -16,4 +16,32 @@
    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
    02111-1307 USA.  */
 
+#include <kernel-features.h>
+
+/* All locks in libc are private.  Use the kernel feature if possible.  */
+#define FUTEX_PRIVATE_FLAG	128
+#ifdef __ASSUME_PRIVATE_FUTEX
+# define FUTEX_WAIT		(0 | FUTEX_PRIVATE_FLAG)
+# define FUTEX_WAKE		(1 | FUTEX_PRIVATE_FLAG)
+#else
+# define LOAD_FUTEX_WAIT(reg,tmp) \
+	stc	gbr, tmp	; \
+	mov.w	99f, reg	; \
+	add	reg, tmp 	; \
+	bra	98f		; \
+	 mov.l	@tmp, reg	; \
+99:	.word	PRIVATE_FUTEX - TLS_PRE_TCB_SIZE ; \
+98:
+	
+# define LOAD_FUTEX_WAKE(reg,tmp) \
+	stc	gbr, tmp	; \
+	mov.w	99f, reg	; \
+	add	reg, tmp 	; \
+	mov.l	@tmp, reg	; \
+	bra	98f		; \
+	 mov	#FUTEX_WAKE, tmp ; \
+99:	.word	PRIVATE_FUTEX - TLS_PRE_TCB_SIZE ; \
+98:	or	tmp, reg
+#endif
+
 #include "lowlevellock.S"

nptl/sysdeps/unix/sysv/linux/sh/lowlevellock.S

@@ -24,8 +24,24 @@
 
 #define SYS_gettimeofday	__NR_gettimeofday
 #define SYS_futex		240
-#define FUTEX_WAIT		0
-#define FUTEX_WAKE		1
+#ifndef FUTEX_WAIT
+# define FUTEX_WAIT		0
+# define FUTEX_WAKE		1
+#endif
+
+#ifndef LOAD_FUTEX_WAIT
+# if FUTEX_WAIT == 0
+#  define LOAD_FUTEX_WAIT(reg,tmp) \
+	xor	reg, reg
+# else
+#  define LOAD_FUTEX_WAIT(reg,tmp) \
+	mov	#FUTEX_WAIT, reg; \
+	extu.b	reg, reg
+# endif
+# define LOAD_FUTEX_WAKE(reg,tmp) \
+	mov	#FUTEX_WAKE, reg; \
+	extu.b	reg, reg
+#endif
 
 
 	.globl	__lll_mutex_lock_wait
@@ -40,7 +56,7 @@ __lll_mutex_lock_wait:
 	mov	r4, r6
 	mov	r5, r8
 	mov	#0, r7		/* No timeout.  */
-	mov	#FUTEX_WAIT, r5
+	LOAD_FUTEX_WAIT (r5, r0)
 
 	mov	#2, r4
 	cmp/eq	r4, r6
@@ -133,7 +149,7 @@ __lll_mutex_timedlock_wait:
 	bt	8f
 
 	mov	r8, r4
-	mov	#FUTEX_WAIT, r5
+	LOAD_FUTEX_WAIT (r5, r0)
 	mov	r10, r6
 	mov	r15, r7
 	mov	#SYS_futex, r3
@@ -192,7 +208,7 @@ __lll_mutex_timedlock_wait:
 	.align	5
 	cfi_startproc
 __lll_mutex_unlock_wake:
-	mov	#FUTEX_WAKE, r5
+	LOAD_FUTEX_WAKE (r5, r0)
 	mov	#1, r6		/* Wake one thread.  */
 	mov	#0, r7
 	mov.l	r7, @r4		/* Stores 0.  */
@@ -261,7 +277,10 @@ __lll_timedwait_tid:
 	bt	4f
 
 	mov	r8, r4
-	mov	#FUTEX_WAIT, r5
+	/* XXX The kernel so far uses global futex for the wakeup at
+	   all times.  */
+	mov	#0, r5
+	extu.b	r5, r5
 	mov	r2, r6
 	mov	r15, r7
 	mov	#SYS_futex, r3

nptl/sysdeps/unix/sysv/linux/sh/lowlevellock.h

@@ -22,13 +22,22 @@
 #include <time.h>
 #include <sys/param.h>
 #include <bits/pthreadtypes.h>
+#include <kernel-features.h>
 
 #define SYS_futex		240
 #define FUTEX_WAIT		0
 #define FUTEX_WAKE		1
 #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
 
 
 /* Initializer for compatibility lock.  */
@@ -251,7 +260,7 @@ extern int __lll_mutex_unlock_wake (int *__futex) attribute_hidden;
 		and %2,%0\n\
 		mov.l %0,@%1\n\
 	     1: mov r1,r15"\
-		: "=&r" (__result) : "r" (__futex), "r" (FUTEX_TID_MASK) \
+		: "=&r" (__result) : "r" (__futex), "r" (FUTEX_WAITERS) \
 		: "r0", "r1", "memory");	\
 	    if (__result) \
 	      __lll_mutex_unlock_wake (__futex); })
@@ -269,7 +278,7 @@ 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); })
+	    lll_futex_wake (__futex, 1, 0); })
 
 #define lll_mutex_islocked(futex) \
   (futex != 0)
@@ -294,14 +303,56 @@ typedef int lll_lock_t;
 	trapa #0x14"
 # endif
 
-#define lll_futex_wait(futex, val) \
+#define lll_futex_wait(futex, val, private) \
+  lll_futex_timed_wait (futex, val, NULL, private)
+
+
+#define lll_futex_timed_wait(futex, val, timeout, private) \
   ({									      \
     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 __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_futex_wake(futex, nr, private) \
+  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;				      \
+    __asm __volatile (SYSCALL_WITH_INST_PAD				      \
+		      : "=z" (__ignore)					      \
+		      : "r" (__r3), "r" (__r4), "r" (__r5),		      \
+			"r" (__r6), "r" (__r7)				      \
+		      : "memory", "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),		      \
@@ -311,14 +362,32 @@ typedef int lll_lock_t;
   })
 
 
-#define lll_futex_timed_wait(futex, val, timeout) \
+# 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") = FUTEX_WAIT;		      \
+    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),		      \
@@ -328,20 +397,22 @@ typedef int lll_lock_t;
   })
 
 
-#define lll_futex_wake(futex, nr) \
+# 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:
@@ -367,7 +438,7 @@ extern int __lll_wait_tid (int *tid) attribute_hidden;
   do {									      \
     __typeof (tid) *__tid = &(tid);					      \
     while (*__tid != 0)							      \
-      lll_futex_wait (__tid, *__tid);					      \
+      lll_futex_wait (__tid, *__tid, 0);				      \
   } while (0)
 
 extern int __lll_timedwait_tid (int *tid, const struct timespec *abstime)

nptl/sysdeps/unix/sysv/linux/sh/pthread_barrier_wait.S

@@ -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
@@ -65,7 +65,13 @@ pthread_barrier_wait:
 #if CURR_EVENT != 0
 	add	#CURR_EVENT, r4
 #endif
+#if FUTEX_WAIT == 0
+	mov.l	@(PRIVATE,r8), r5
+#else
 	mov	#FUTEX_WAIT, r5
+	mov.l	@(PRIVATE,r8), r0
+	or	r0, r5
+#endif
 	mov	#0, r7
 8:
 	mov	#SYS_futex, r3
@@ -118,6 +124,8 @@ pthread_barrier_wait:
 #endif
 	mov	#0, r7
 	mov	#FUTEX_WAKE, r5
+	mov.l	@(PRIVATE,r8), r0
+	or	r0, r5
 	mov	#SYS_futex, r3
 	extu.b	r3, r3
 	trapa	#0x14