* sysdeps/powerpc/bits/atomic.h

	[!MUTEX_HINT_ACQ]: Define MUTEX_HINT_ACQ.
	[!MUTEX_HINT_REL]: Define MUTEX_HINT_REL.
	(__arch_compare_and_exchange_val_32_acq): Add MUTEX_HINT_ACQ to lwarx.
	(__arch_compare_and_exchange_val_32_rel): Add MUTEX_HINT_REL to lwarx.
	(__arch_atomic_exchange_val_32_acq): Add MUTEX_HINT_ACQ to lwarx.
	(__arch_atomic_exchange_rel_32_rel): Add MUTEX_HINT_REL to lwarx.
	* sysdeps/powerpc/powerpc32/bits/atomic.h [_ARCH_PWR6 || _ARCH_PWR6X]:
	Define MUTEX_HINT_ACQ as ",1" and MUTEX_HINT_REL as ",0".
	(__arch_compare_and_exchange_bool_32_acq): Add MUTEX_HINT_ACQ to lwarx.
	(__arch_compare_and_exchange_bool_32_rel): Add MUTEX_HINT_REL to lwarx.
	* sysdeps/powerpc/powerpc64/bits/atomic.h [_ARCH_PWR6 || _ARCH_PWR6D]:
	Define MUTEX_HINT_ACQ as ",1" and MUTEX_HINT_REL as ",0".
	(__arch_compare_and_exchange_bool_32_acq): Add MUTEX_HINT_ACQ to lwarx.
	(__arch_compare_and_exchange_bool_32_rel): Add MUTEX_HINT_REL to lwarx.
	(__arch_compare_and_exchange_bool_64_acq): Add MUTEX_HINT_ACQ to lwarx.
	(__arch_compare_and_exchange_bool_64_rel): Add MUTEX_HINT_REL to lwarx.
	(__arch_compare_and_exchange_val_64_acq): Add MUTEX_HINT_ACQ to lwarx.
	(__arch_compare_and_exchange_val_64_rel): Add MUTEX_HINT_REL to lwarx.
	(__arch_atomic_exchange_val_64_acq): Add MUTEX_HINT_ACQ to lwarx.
	(__arch_atomic_exchange_rel_64_rel): Add MUTEX_HINT_REL to lwarx.

2007-03-20  Jakub Jelinek  <jakub@redhat.com>

5 files changed, 108 insertions, 46 deletions

diff --git a/ChangeLog b/ChangeLog
index 1a5f369b2d..5774d4cb84 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,4 +1,28 @@
-007-03-20  Jakub Jelinek  <jakub@redhat.com>
+2007-03-19  Steven Munroe  <sjmunroe@us.ibm.com>
+
+	* sysdeps/powerpc/bits/atomic.h
+	[!MUTEX_HINT_ACQ]: Define MUTEX_HINT_ACQ.
+	[!MUTEX_HINT_REL]: Define MUTEX_HINT_REL.
+	(__arch_compare_and_exchange_val_32_acq): Add MUTEX_HINT_ACQ to lwarx.
+	(__arch_compare_and_exchange_val_32_rel): Add MUTEX_HINT_REL to lwarx.
+	(__arch_atomic_exchange_val_32_acq): Add MUTEX_HINT_ACQ to lwarx.
+	(__arch_atomic_exchange_rel_32_rel): Add MUTEX_HINT_REL to lwarx.
+	* sysdeps/powerpc/powerpc32/bits/atomic.h [_ARCH_PWR6 || _ARCH_PWR6X]:
+	Define MUTEX_HINT_ACQ as ",1" and MUTEX_HINT_REL as ",0".
+	(__arch_compare_and_exchange_bool_32_acq): Add MUTEX_HINT_ACQ to lwarx.
+	(__arch_compare_and_exchange_bool_32_rel): Add MUTEX_HINT_REL to lwarx.
+	* sysdeps/powerpc/powerpc64/bits/atomic.h [_ARCH_PWR6 || _ARCH_PWR6D]:
+	Define MUTEX_HINT_ACQ as ",1" and MUTEX_HINT_REL as ",0".
+	(__arch_compare_and_exchange_bool_32_acq): Add MUTEX_HINT_ACQ to lwarx.
+	(__arch_compare_and_exchange_bool_32_rel): Add MUTEX_HINT_REL to lwarx.
+	(__arch_compare_and_exchange_bool_64_acq): Add MUTEX_HINT_ACQ to lwarx.
+	(__arch_compare_and_exchange_bool_64_rel): Add MUTEX_HINT_REL to lwarx.
+	(__arch_compare_and_exchange_val_64_acq): Add MUTEX_HINT_ACQ to lwarx.
+	(__arch_compare_and_exchange_val_64_rel): Add MUTEX_HINT_REL to lwarx.
+	(__arch_atomic_exchange_val_64_acq): Add MUTEX_HINT_ACQ to lwarx.
+	(__arch_atomic_exchange_rel_64_rel): Add MUTEX_HINT_REL to lwarx.
+
+2007-03-20  Jakub Jelinek  <jakub@redhat.com>
 
 	* sysdeps/unix/sysv/linux/powerpc/libc-start.c
 	(__cache_line_size): Define the variable here.  Add
diff --git a/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h b/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
index 0136b97595..239f4ddf55 100644
--- a/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
+++ b/nptl/sysdeps/unix/sysv/linux/powerpc/lowlevellock.h
@@ -1,4 +1,4 @@
-/* Copyright (C) 2003, 2004, 2006 Free Software Foundation, Inc.
+/* Copyright (C) 2003, 2004, 2006, 2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Paul Mackerras <paulus@au.ibm.com>, 2003.
 
@@ -25,7 +25,6 @@
 #include <bits/pthreadtypes.h>
 #include <atomic.h>
 
-
 #ifndef __NR_futex
 # define __NR_futex		221
 #endif
@@ -133,7 +132,7 @@
 /* Set *futex to ID if it is 0, atomically.  Returns the old value */
 #define __lll_robust_trylock(futex, id) \
   ({ int __val;								      \
-     __asm __volatile ("1:	lwarx	%0,0,%2\n"			      \
+     __asm __volatile ("1:	lwarx	%0,0,%2" MUTEX_HINT_ACQ "\n"	      \
 		       "	cmpwi	0,%0,0\n"			      \
 		       "	bne	2f\n"				      \
 		       "	stwcx.	%3,0,%2\n"			      \
diff --git a/sysdeps/powerpc/bits/atomic.h b/sysdeps/powerpc/bits/atomic.h
index 31f27e9e10..d71f64e0ac 100644
--- a/sysdeps/powerpc/bits/atomic.h
+++ b/sysdeps/powerpc/bits/atomic.h
@@ -70,6 +70,13 @@ typedef uintmax_t uatomic_max_t;
 # endif
 #endif
 
+#ifndef MUTEX_HINT_ACQ
+# define MUTEX_HINT_ACQ
+#endif
+#ifndef MUTEX_HINT_REL
+# define MUTEX_HINT_REL
+#endif
+
 #define atomic_full_barrier()	__asm ("sync" ::: "memory")
 #define atomic_write_barrier()	__asm ("eieio" ::: "memory")
 
@@ -78,7 +85,7 @@ typedef uintmax_t uatomic_max_t;
       __typeof (*(mem)) __tmp;						      \
       __typeof (mem)  __memp = (mem);					      \
       __asm __volatile (						      \
-		        "1:	lwarx	%0,0,%1\n"			      \
+		        "1:	lwarx	%0,0,%1" MUTEX_HINT_ACQ "\n"	      \
 		        "	cmpw	%0,%2\n"			      \
 		        "	bne	2f\n"				      \
 		        "	stwcx.	%3,0,%1\n"			      \
@@ -95,7 +102,7 @@ typedef uintmax_t uatomic_max_t;
       __typeof (*(mem)) __tmp;						      \
       __typeof (mem)  __memp = (mem);					      \
       __asm __volatile (__ARCH_REL_INSTR "\n"				      \
-		        "1:	lwarx	%0,0,%1\n"			      \
+		        "1:	lwarx	%0,0,%1" MUTEX_HINT_REL "\n"	      \
 		        "	cmpw	%0,%2\n"			      \
 		        "	bne	2f\n"				      \
 		        "	stwcx.	%3,0,%1\n"			      \
@@ -111,7 +118,7 @@ typedef uintmax_t uatomic_max_t;
   ({									      \
     __typeof (*mem) __val;						      \
     __asm __volatile (							      \
-		      "1:	lwarx	%0,0,%2\n"			      \
+		      "1:	lwarx	%0,0,%2" MUTEX_HINT_ACQ "\n"	      \
 		      "		stwcx.	%3,0,%2\n"			      \
 		      "		bne-	1b\n"				      \
 		      "   " __ARCH_ACQ_INSTR				      \
@@ -125,7 +132,7 @@ typedef uintmax_t uatomic_max_t;
   ({									      \
     __typeof (*mem) __val;						      \
     __asm __volatile (__ARCH_REL_INSTR "\n"				      \
-		      "1:	lwarx	%0,0,%2\n"			      \
+		      "1:	lwarx	%0,0,%2" MUTEX_HINT_REL "\n"	      \
 		      "		stwcx.	%3,0,%2\n"			      \
 		      "		bne-	1b"				      \
 		      : "=&r" (__val), "=m" (*mem)			      \
diff --git a/sysdeps/powerpc/powerpc32/bits/atomic.h b/sysdeps/powerpc/powerpc32/bits/atomic.h
index 6fcc669fb1..62cf991b8d 100644
--- a/sysdeps/powerpc/powerpc32/bits/atomic.h
+++ b/sysdeps/powerpc/powerpc32/bits/atomic.h
@@ -1,5 +1,5 @@
 /* Atomic operations.  PowerPC32 version.
-   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 Paul Mackerras <paulus@au.ibm.com>, 2003.
 
@@ -18,17 +18,33 @@
    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
    02111-1307 USA.  */
 
+/*  POWER6 adds a "Mutex Hint" to the Load and Reserve instruction.
+    This is a hint to the hardware to expect additional updates adjacent
+    to the lock word or not.  If we are acquiring a Mutex, the hint
+    should be true. Otherwise we releasing a Mutex or doing a simple
+    atomic operation.  In that case we don't expect addtional updates
+    adjacent to the lock word after the Store Conditional and the hint
+    should be false.  */
+    
+#if defined _ARCH_PWR6 || defined _ARCH_PWR6X
+# define MUTEX_HINT_ACQ	",1"
+# define MUTEX_HINT_REL	",0"
+#else
+# define MUTEX_HINT_ACQ
+# define MUTEX_HINT_REL
+#endif
+
 /*
  * The 32-bit exchange_bool is different on powerpc64 because the subf
  * does signed 64-bit arthmatic while the lwarx is 32-bit unsigned
  * (a load word and zero (high 32) form).  So powerpc64 has a slightly
  * different version in sysdeps/powerpc/powerpc64/bits/atomic.h.
  */
-# define __arch_compare_and_exchange_bool_32_acq(mem, newval, oldval)         \
+#define __arch_compare_and_exchange_bool_32_acq(mem, newval, oldval)         \
 ({									      \
   unsigned int __tmp;							      \
   __asm __volatile (							      \
-		    "1:	lwarx	%0,0,%1\n"				      \
+		    "1:	lwarx	%0,0,%1" MUTEX_HINT_ACQ "\n"		      \
 		    "	subf.	%0,%2,%0\n"				      \
 		    "	bne	2f\n"					      \
 		    "	stwcx.	%3,0,%1\n"				      \
@@ -40,11 +56,11 @@
   __tmp != 0;								      \
 })
 
-# define __arch_compare_and_exchange_bool_32_rel(mem, newval, oldval)	      \
+#define __arch_compare_and_exchange_bool_32_rel(mem, newval, oldval)	      \
 ({									      \
   unsigned int __tmp;							      \
   __asm __volatile (__ARCH_REL_INSTR "\n"				      \
-		    "1:	lwarx	%0,0,%1\n"				      \
+		    "1:	lwarx	%0,0,%1" MUTEX_HINT_REL "\n"		      \
 		    "	subf.	%0,%2,%0\n"				      \
 		    "	bne	2f\n"					      \
 		    "	stwcx.	%3,0,%1\n"				      \
@@ -59,34 +75,34 @@
 /* Powerpc32 processors don't implement the 64-bit (doubleword) forms of
    load and reserve (ldarx) and store conditional (stdcx.) instructions.
    So for powerpc32 we stub out the 64-bit forms.  */
-# define __arch_compare_and_exchange_bool_64_acq(mem, newval, oldval) \
+#define __arch_compare_and_exchange_bool_64_acq(mem, newval, oldval) \
   (abort (), 0)
 
-# define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
+#define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
   (abort (), (__typeof (*mem)) 0)
 
-# define __arch_compare_and_exchange_bool_64_rel(mem, newval, oldval) \
+#define __arch_compare_and_exchange_bool_64_rel(mem, newval, oldval) \
   (abort (), 0)
 
-# define __arch_compare_and_exchange_val_64_rel(mem, newval, oldval) \
+#define __arch_compare_and_exchange_val_64_rel(mem, newval, oldval) \
   (abort (), (__typeof (*mem)) 0)
 
-# define __arch_atomic_exchange_64_acq(mem, value) \
+#define __arch_atomic_exchange_64_acq(mem, value) \
     ({ abort (); (*mem) = (value); })
 
-# define __arch_atomic_exchange_64_rel(mem, value) \
+#define __arch_atomic_exchange_64_rel(mem, value) \
     ({ abort (); (*mem) = (value); })
 
-# define __arch_atomic_exchange_and_add_64(mem, value) \
+#define __arch_atomic_exchange_and_add_64(mem, value) \
     ({ abort (); (*mem) = (value); })
 
-# define __arch_atomic_increment_val_64(mem) \
+#define __arch_atomic_increment_val_64(mem) \
     ({ abort (); (*mem)++; })
 
-# define __arch_atomic_decrement_val_64(mem) \
+#define __arch_atomic_decrement_val_64(mem) \
     ({ abort (); (*mem)--; })
 
-# define __arch_atomic_decrement_if_positive_64(mem) \
+#define __arch_atomic_decrement_if_positive_64(mem) \
     ({ abort (); (*mem)--; })
 
 #ifdef _ARCH_PWR4
diff --git a/sysdeps/powerpc/powerpc64/bits/atomic.h b/sysdeps/powerpc/powerpc64/bits/atomic.h
index e46dc1e4d7..3465bb31b4 100644
--- a/sysdeps/powerpc/powerpc64/bits/atomic.h
+++ b/sysdeps/powerpc/powerpc64/bits/atomic.h
@@ -1,5 +1,5 @@
 /* Atomic operations.  PowerPC64 version.
-   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 Paul Mackerras <paulus@au.ibm.com>, 2003.
 
@@ -18,17 +18,33 @@
    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
    02111-1307 USA.  */
 
+/*  POWER6 adds a "Mutex Hint" to the Load and Reserve instruction.
+    This is a hint to the hardware to expect additional updates adjacent
+    to the lock word or not.  If we are acquiring a Mutex, the hint
+    should be true. Otherwise we releasing a Mutex or doing a simple
+    atomic operation.  In that case we don't expect addtional updates
+    adjacent to the lock word after the Store Conditional and the hint
+    should be false.  */
+
+#if defined _ARCH_PWR6 || defined _ARCH_PWR6X
+# define MUTEX_HINT_ACQ	",1"
+# define MUTEX_HINT_REL	",0"
+#else
+# define MUTEX_HINT_ACQ
+# define MUTEX_HINT_REL
+#endif
+
 /* The 32-bit exchange_bool is different on powerpc64 because the subf
    does signed 64-bit arthmatic while the lwarx is 32-bit unsigned
    (a load word and zero (high 32) form) load.
    In powerpc64 register values are 64-bit by default,  including oldval.
    The value in old val unknown sign extension, lwarx loads the 32-bit
    value as unsigned.  So we explicitly clear the high 32 bits in oldval.  */
-# define __arch_compare_and_exchange_bool_32_acq(mem, newval, oldval) \
+#define __arch_compare_and_exchange_bool_32_acq(mem, newval, oldval) \
 ({									      \
   unsigned int __tmp, __tmp2;						      \
   __asm __volatile ("   clrldi  %1,%1,32\n"				      \
-		    "1:	lwarx	%0,0,%2\n"				      \
+		    "1:	lwarx	%0,0,%2" MUTEX_HINT_ACQ "\n"	 	      \
 		    "	subf.	%0,%1,%0\n"				      \
 		    "	bne	2f\n"					      \
 		    "	stwcx.	%4,0,%2\n"				      \
@@ -40,12 +56,12 @@
   __tmp != 0;								      \
 })
 
-# define __arch_compare_and_exchange_bool_32_rel(mem, newval, oldval) \
+#define __arch_compare_and_exchange_bool_32_rel(mem, newval, oldval) \
 ({									      \
   unsigned int __tmp, __tmp2;						      \
   __asm __volatile (__ARCH_REL_INSTR "\n"				      \
 		    "   clrldi  %1,%1,32\n"				      \
-		    "1:	lwarx	%0,0,%2\n"				      \
+		    "1:	lwarx	%0,0,%2" MUTEX_HINT_REL "\n"		      \
 		    "	subf.	%0,%1,%0\n"				      \
 		    "	bne	2f\n"					      \
 		    "	stwcx.	%4,0,%2\n"				      \
@@ -62,11 +78,11 @@
  * and Store doubleword conditional indexed (stdcx) instructions.  So here
  * we define the 64-bit forms.
  */
-# define __arch_compare_and_exchange_bool_64_acq(mem, newval, oldval) \
+#define __arch_compare_and_exchange_bool_64_acq(mem, newval, oldval) \
 ({									      \
   unsigned long	__tmp;							      \
   __asm __volatile (							      \
-		    "1:	ldarx	%0,0,%1\n"				      \
+		    "1:	ldarx	%0,0,%1" MUTEX_HINT_ACQ "\n"		      \
 		    "	subf.	%0,%2,%0\n"				      \
 		    "	bne	2f\n"					      \
 		    "	stdcx.	%3,0,%1\n"				      \
@@ -78,11 +94,11 @@
   __tmp != 0;								      \
 })
 
-# define __arch_compare_and_exchange_bool_64_rel(mem, newval, oldval) \
+#define __arch_compare_and_exchange_bool_64_rel(mem, newval, oldval) \
 ({									      \
   unsigned long	__tmp;							      \
   __asm __volatile (__ARCH_REL_INSTR "\n"				      \
-		    "1:	ldarx	%0,0,%1\n"				      \
+		    "1:	ldarx	%0,0,%2" MUTEX_HINT_REL "\n"		      \
 		    "	subf.	%0,%2,%0\n"				      \
 		    "	bne	2f\n"					      \
 		    "	stdcx.	%3,0,%1\n"				      \
@@ -99,7 +115,7 @@
       __typeof (*(mem)) __tmp;						      \
       __typeof (mem)  __memp = (mem);					      \
       __asm __volatile (						      \
-		        "1:	ldarx	%0,0,%1\n"			      \
+		        "1:	ldarx	%0,0,%1" MUTEX_HINT_ACQ "\n"	      \
 		        "	cmpd	%0,%2\n"			      \
 		        "	bne	2f\n"				      \
 		        "	stdcx.	%3,0,%1\n"			      \
@@ -116,7 +132,7 @@
       __typeof (*(mem)) __tmp;						      \
       __typeof (mem)  __memp = (mem);					      \
       __asm __volatile (__ARCH_REL_INSTR "\n"				      \
-		        "1:	ldarx	%0,0,%1\n"			      \
+		        "1:	ldarx	%0,0,%1" MUTEX_HINT_REL "\n"	      \
 		        "	cmpd	%0,%2\n"			      \
 		        "	bne	2f\n"				      \
 		        "	stdcx.	%3,0,%1\n"			      \
@@ -128,11 +144,11 @@
       __tmp;								      \
   })
 
-# define __arch_atomic_exchange_64_acq(mem, value) \
+#define __arch_atomic_exchange_64_acq(mem, value) \
     ({									      \
       __typeof (*mem) __val;						      \
       __asm __volatile (__ARCH_REL_INSTR "\n"				      \
-			"1:	ldarx	%0,0,%2\n"			      \
+			"1:	ldarx	%0,0,%2" MUTEX_HINT_ACQ "\n"	      \
 			"	stdcx.	%3,0,%2\n"			      \
 			"	bne-	1b\n"				      \
 		  " " __ARCH_ACQ_INSTR					      \
@@ -142,11 +158,11 @@
       __val;								      \
     })
 
-# define __arch_atomic_exchange_64_rel(mem, value) \
+#define __arch_atomic_exchange_64_rel(mem, value) \
     ({									      \
       __typeof (*mem) __val;						      \
       __asm __volatile (__ARCH_REL_INSTR "\n"				      \
-			"1:	ldarx	%0,0,%2\n"			      \
+			"1:	ldarx	%0,0,%2" MUTEX_HINT_REL "\n"	      \
 			"	stdcx.	%3,0,%2\n"			      \
 			"	bne-	1b"				      \
 			: "=&r" (__val), "=m" (*mem)			      \
@@ -155,7 +171,7 @@
       __val;								      \
     })
 
-# define __arch_atomic_exchange_and_add_64(mem, value) \
+#define __arch_atomic_exchange_and_add_64(mem, value) \
     ({									      \
       __typeof (*mem) __val, __tmp;					      \
       __asm __volatile ("1:	ldarx	%0,0,%3\n"			      \
@@ -168,7 +184,7 @@
       __val;								      \
     })
 
-# define __arch_atomic_increment_val_64(mem) \
+#define __arch_atomic_increment_val_64(mem) \
     ({									      \
       __typeof (*(mem)) __val;						      \
       __asm __volatile ("1:	ldarx	%0,0,%2\n"			      \
@@ -181,7 +197,7 @@
       __val;								      \
     })
 
-# define __arch_atomic_decrement_val_64(mem) \
+#define __arch_atomic_decrement_val_64(mem) \
     ({									      \
       __typeof (*(mem)) __val;						      \
       __asm __volatile ("1:	ldarx	%0,0,%2\n"			      \
@@ -194,7 +210,7 @@
       __val;								      \
     })
 
-# define __arch_atomic_decrement_if_positive_64(mem) \
+#define __arch_atomic_decrement_if_positive_64(mem) \
   ({ int __val, __tmp;							      \
      __asm __volatile ("1:	ldarx	%0,0,%3\n"			      \
 		       "	cmpdi	0,%0,0\n"			      \
@@ -212,13 +228,13 @@
 /*
  * All powerpc64 processors support the new "light weight"  sync (lwsync).
  */
-# define atomic_read_barrier()	__asm ("lwsync" ::: "memory")
+#define atomic_read_barrier()	__asm ("lwsync" ::: "memory")
 /*
  * "light weight" sync can also be used for the release barrier.
  */
-# ifndef UP
-#  define __ARCH_REL_INSTR	"lwsync"
-# endif
+#ifndef UP
+# define __ARCH_REL_INSTR	"lwsync"
+#endif
 
 /*
  * Include the rest of the atomic ops macros which are common to both