Added new 'bits/atomic.h' for most of the architectures and the top-level 'include/atomic.h' to add in new atomic operations for use by NPTL. There are multiple files for PowerPC and Sparc for 'atomic.h'. I will let those architecture maintainers choose the correct file. The files come from glibc in 'sysdeps/ARCH/bits'.

include/atomic.h

@@ -0,0 +1,261 @@
+/* Internal macros for atomic operations for GNU C Library.
+   Copyright (C) 2002, 2003, 2004 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#ifndef _ATOMIC_H
+#define _ATOMIC_H	1
+
+#include <stdlib.h>
+
+#include <bits/atomic.h>
+
+/* Wrapper macros to call pre_NN_post (mem, ...) where NN is the
+   bit width of *MEM.  The calling macro puts parens around MEM
+   and following args.  */
+#define __atomic_val_bysize(pre, post, mem, ...)			      \
+  ({									      \
+    __typeof (*mem) __result;						      \
+    if (sizeof (*mem) == 1)						      \
+      __result = pre##_8_##post (mem, __VA_ARGS__);			      \
+    else if (sizeof (*mem) == 2)					      \
+      __result = pre##_16_##post (mem, __VA_ARGS__);			      \
+    else if (sizeof (*mem) == 4)					      \
+      __result = pre##_32_##post (mem, __VA_ARGS__);			      \
+    else if (sizeof (*mem) == 8)					      \
+      __result = pre##_64_##post (mem, __VA_ARGS__);			      \
+    else								      \
+      abort ();								      \
+    __result;								      \
+  })
+#define __atomic_bool_bysize(pre, post, mem, ...)			      \
+  ({									      \
+    int __result;							      \
+    if (sizeof (*mem) == 1)						      \
+      __result = pre##_8_##post (mem, __VA_ARGS__);			      \
+    else if (sizeof (*mem) == 2)					      \
+      __result = pre##_16_##post (mem, __VA_ARGS__);			      \
+    else if (sizeof (*mem) == 4)					      \
+      __result = pre##_32_##post (mem, __VA_ARGS__);			      \
+    else if (sizeof (*mem) == 8)					      \
+      __result = pre##_64_##post (mem, __VA_ARGS__);			      \
+    else								      \
+      abort ();								      \
+    __result;								      \
+  })
+
+
+/* Atomically store NEWVAL in *MEM if *MEM is equal to OLDVAL.
+   Return the old *MEM value.  */
+#if !defined atomic_compare_and_exchange_val_acq \
+    && defined __arch_compare_and_exchange_val_32_acq
+# define atomic_compare_and_exchange_val_acq(mem, newval, oldval) \
+  __atomic_val_bysize (__arch_compare_and_exchange_val,acq,		      \
+		       mem, newval, oldval)
+#endif
+
+
+#ifndef atomic_compare_and_exchange_val_rel
+# define atomic_compare_and_exchange_val_rel(mem, newval, oldval)	      \
+  atomic_compare_and_exchange_val_acq (mem, newval, oldval)
+#endif
+
+
+/* Atomically store NEWVAL in *MEM if *MEM is equal to OLDVAL.
+   Return zero if *MEM was changed or non-zero if no exchange happened.  */
+#ifndef atomic_compare_and_exchange_bool_acq
+# ifdef __arch_compare_and_exchange_bool_32_acq
+#  define atomic_compare_and_exchange_bool_acq(mem, newval, oldval) \
+  __atomic_bool_bysize (__arch_compare_and_exchange_bool,acq,		      \
+		        mem, newval, oldval)
+#  else
+#   define atomic_compare_and_exchange_bool_acq(mem, newval, oldval) \
+  ({ /* Cannot use __oldval here, because macros later in this file might     \
+	call this macro with __oldval argument.	 */			      \
+     __typeof (oldval) __old = (oldval);				      \
+     atomic_compare_and_exchange_val_acq (mem, newval, __old) != __old;	      \
+  })
+# endif
+#endif
+
+
+#ifndef atomic_compare_and_exchange_bool_rel
+# define atomic_compare_and_exchange_bool_rel(mem, newval, oldval) \
+  atomic_compare_and_exchange_bool_acq (mem, newval, oldval)
+#endif
+
+
+/* Store NEWVALUE in *MEM and return the old value.  */
+#ifndef atomic_exchange_acq
+# define atomic_exchange_acq(mem, newvalue) \
+  ({ __typeof (*(mem)) __oldval;					      \
+     __typeof (mem) __memp = (mem);					      \
+     __typeof (*(mem)) __value = (newvalue);				      \
+									      \
+     do									      \
+       __oldval = (*__memp);						      \
+     while (__builtin_expect (atomic_compare_and_exchange_bool_acq (__memp,   \
+								    __value,  \
+								    __oldval),\
+			      0));					      \
+									      \
+     __oldval; })
+#endif
+
+#ifndef atomic_exchange_rel
+# define atomic_exchange_rel(mem, newvalue) atomic_exchange_acq (mem, newvalue)
+#endif
+
+
+/* Add VALUE to *MEM and return the old value of *MEM.  */
+#ifndef atomic_exchange_and_add
+# define atomic_exchange_and_add(mem, value) \
+  ({ __typeof (*(mem)) __oldval;					      \
+     __typeof (mem) __memp = (mem);					      \
+     __typeof (*(mem)) __value = (value);				      \
+									      \
+     do									      \
+       __oldval = (*__memp);						      \
+     while (__builtin_expect (atomic_compare_and_exchange_bool_acq (__memp,   \
+								    __oldval  \
+								    + __value,\
+								    __oldval),\
+			      0));					      \
+									      \
+     __oldval; })
+#endif
+
+
+#ifndef atomic_add
+# define atomic_add(mem, value) (void) atomic_exchange_and_add ((mem), (value))
+#endif
+
+
+#ifndef atomic_increment
+# define atomic_increment(mem) atomic_add ((mem), 1)
+#endif
+
+
+#ifndef atomic_increment_val
+# define atomic_increment_val(mem) (atomic_exchange_and_add ((mem), 1) + 1)
+#endif
+
+
+/* Add one to *MEM and return true iff it's now zero.  */
+#ifndef atomic_increment_and_test
+# define atomic_increment_and_test(mem) \
+  (atomic_exchange_and_add ((mem), 1) + 1 == 0)
+#endif
+
+
+#ifndef atomic_decrement
+# define atomic_decrement(mem) atomic_add ((mem), -1)
+#endif
+
+
+#ifndef atomic_decrement_val
+# define atomic_decrement_val(mem) (atomic_exchange_and_add ((mem), -1) - 1)
+#endif
+
+
+/* Subtract 1 from *MEM and return true iff it's now zero.  */
+#ifndef atomic_decrement_and_test
+# define atomic_decrement_and_test(mem) \
+  (atomic_exchange_and_add ((mem), -1) == 1)
+#endif
+
+
+/* Decrement *MEM if it is > 0, and return the old value.  */
+#ifndef atomic_decrement_if_positive
+# define atomic_decrement_if_positive(mem) \
+  ({ __typeof (*(mem)) __oldval;					      \
+     __typeof (mem) __memp = (mem);					      \
+									      \
+     do									      \
+       {								      \
+	 __oldval = *__memp;						      \
+	 if (__builtin_expect (__oldval <= 0, 0))			      \
+	   break;							      \
+       }								      \
+     while (__builtin_expect (atomic_compare_and_exchange_bool_acq (__memp,   \
+								    __oldval  \
+								    - 1,      \
+								    __oldval),\
+			      0));\
+     __oldval; })
+#endif
+
+
+#ifndef atomic_add_negative
+# define atomic_add_negative(mem, value)				      \
+  ({ __typeof (value) __aan_value = (value);				      \
+     atomic_exchange_and_add (mem, __aan_value) < -__aan_value; })
+#endif
+
+
+#ifndef atomic_add_zero
+# define atomic_add_zero(mem, value)					      \
+  ({ __typeof (value) __aaz_value = (value);				      \
+     atomic_exchange_and_add (mem, __aaz_value) == -__aaz_value; })
+#endif
+
+
+#ifndef atomic_bit_set
+# define atomic_bit_set(mem, bit) \
+  (void) atomic_bit_test_set(mem, bit)
+#endif
+
+
+#ifndef atomic_bit_test_set
+# define atomic_bit_test_set(mem, bit) \
+  ({ __typeof (*(mem)) __oldval;					      \
+     __typeof (mem) __memp = (mem);					      \
+     __typeof (*(mem)) __mask = ((__typeof (*(mem))) 1 << (bit));	      \
+									      \
+     do									      \
+       __oldval = (*__memp);						      \
+     while (__builtin_expect (atomic_compare_and_exchange_bool_acq (__memp,   \
+								    __oldval  \
+								    | __mask, \
+								    __oldval),\
+			      0));					      \
+									      \
+     __oldval & __mask; })
+#endif
+
+
+#ifndef atomic_full_barrier
+# define atomic_full_barrier() __asm ("" ::: "memory")
+#endif
+
+
+#ifndef atomic_read_barrier
+# define atomic_read_barrier() atomic_full_barrier ()
+#endif
+
+
+#ifndef atomic_write_barrier
+# define atomic_write_barrier() atomic_full_barrier ()
+#endif
+
+
+#ifndef atomic_delay
+# define atomic_delay() do { /* nothing */ } while (0)
+#endif
+
+#endif	/* atomic.h */

libc/sysdeps/linux/alpha/bits/atomic.h

@@ -0,0 +1,369 @@
+/* Copyright (C) 2003 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
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <stdint.h>
+
+typedef int8_t atomic8_t;
+typedef uint8_t uatomic8_t;
+typedef int_fast8_t atomic_fast8_t;
+typedef uint_fast8_t uatomic_fast8_t;
+
+typedef int16_t atomic16_t;
+typedef uint16_t uatomic16_t;
+typedef int_fast16_t atomic_fast16_t;
+typedef uint_fast16_t uatomic_fast16_t;
+
+typedef int32_t atomic32_t;
+typedef uint32_t uatomic32_t;
+typedef int_fast32_t atomic_fast32_t;
+typedef uint_fast32_t uatomic_fast32_t;
+
+typedef int64_t atomic64_t;
+typedef uint64_t uatomic64_t;
+typedef int_fast64_t atomic_fast64_t;
+typedef uint_fast64_t uatomic_fast64_t;
+
+typedef intptr_t atomicptr_t;
+typedef uintptr_t uatomicptr_t;
+typedef intmax_t atomic_max_t;
+typedef uintmax_t uatomic_max_t;
+
+
+#ifdef UP
+# define __MB		/* nothing */
+#else
+# define __MB		"	mb\n"
+#endif
+
+
+/* Compare and exchange.  For all of the "xxx" routines, we expect a
+   "__prev" and a "__cmp" variable to be provided by the enclosing scope,
+   in which values are returned.  */
+
+#define __arch_compare_and_exchange_xxx_8_int(mem, new, old, mb1, mb2)	\
+({									\
+  unsigned long __tmp, __snew, __addr64;				\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"	andnot	%[__addr8],7,%[__addr64]\n"			\
+	"	insbl	%[__new],%[__addr8],%[__snew]\n"		\
+	"1:	ldq_l	%[__tmp],0(%[__addr64])\n"			\
+	"	extbl	%[__tmp],%[__addr8],%[__prev]\n"		\
+	"	cmpeq	%[__prev],%[__old],%[__cmp]\n"			\
+	"	beq	%[__cmp],2f\n"					\
+	"	mskbl	%[__tmp],%[__addr8],%[__tmp]\n"			\
+	"	or	%[__snew],%[__tmp],%[__tmp]\n"			\
+	"	stq_c	%[__tmp],0(%[__addr64])\n"			\
+	"	beq	%[__tmp],1b\n"					\
+		mb2							\
+	"2:"								\
+	: [__prev] "=&r" (__prev),					\
+	  [__snew] "=&r" (__snew),					\
+	  [__tmp] "=&r" (__tmp),					\
+	  [__cmp] "=&r" (__cmp),					\
+	  [__addr64] "=&r" (__addr64)					\
+	: [__addr8] "r" (mem),						\
+	  [__old] "Ir" ((uint64_t)(uint8_t)(uint64_t)(old)),		\
+	  [__new] "r" (new)						\
+	: "memory");							\
+})
+
+#define __arch_compare_and_exchange_xxx_16_int(mem, new, old, mb1, mb2) \
+({									\
+  unsigned long __tmp, __snew, __addr64;				\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"	andnot	%[__addr16],7,%[__addr64]\n"			\
+	"	inswl	%[__new],%[__addr16],%[__snew]\n"		\
+	"1:	ldq_l	%[__tmp],0(%[__addr64])\n"			\
+	"	extwl	%[__tmp],%[__addr16],%[__prev]\n"		\
+	"	cmpeq	%[__prev],%[__old],%[__cmp]\n"			\
+	"	beq	%[__cmp],2f\n"					\
+	"	mskwl	%[__tmp],%[__addr16],%[__tmp]\n"		\
+	"	or	%[__snew],%[__tmp],%[__tmp]\n"			\
+	"	stq_c	%[__tmp],0(%[__addr64])\n"			\
+	"	beq	%[__tmp],1b\n"					\
+		mb2							\
+	"2:"								\
+	: [__prev] "=&r" (__prev),					\
+	  [__snew] "=&r" (__snew),					\
+	  [__tmp] "=&r" (__tmp),					\
+	  [__cmp] "=&r" (__cmp),					\
+	  [__addr64] "=&r" (__addr64)					\
+	: [__addr16] "r" (mem),						\
+	  [__old] "Ir" ((uint64_t)(uint16_t)(uint64_t)(old)),		\
+	  [__new] "r" (new)						\
+	: "memory");							\
+})
+
+#define __arch_compare_and_exchange_xxx_32_int(mem, new, old, mb1, mb2) \
+({									\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"1:	ldl_l	%[__prev],%[__mem]\n"				\
+	"	cmpeq	%[__prev],%[__old],%[__cmp]\n"			\
+	"	beq	%[__cmp],2f\n"					\
+	"	mov	%[__new],%[__cmp]\n"				\
+	"	stl_c	%[__cmp],%[__mem]\n"				\
+	"	beq	%[__cmp],1b\n"					\
+		mb2							\
+	"2:"								\
+	: [__prev] "=&r" (__prev),					\
+	  [__cmp] "=&r" (__cmp)						\
+	: [__mem] "m" (*(mem)),						\
+	  [__old] "Ir" ((uint64_t)(atomic32_t)(uint64_t)(old)),		\
+	  [__new] "Ir" (new)						\
+	: "memory");							\
+})
+
+#define __arch_compare_and_exchange_xxx_64_int(mem, new, old, mb1, mb2) \
+({									\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"1:	ldq_l	%[__prev],%[__mem]\n"				\
+	"	cmpeq	%[__prev],%[__old],%[__cmp]\n"			\
+	"	beq	%[__cmp],2f\n"					\
+	"	mov	%[__new],%[__cmp]\n"				\
+	"	stq_c	%[__cmp],%[__mem]\n"				\
+	"	beq	%[__cmp],1b\n"					\
+		mb2							\
+	"2:"								\
+	: [__prev] "=&r" (__prev),					\
+	  [__cmp] "=&r" (__cmp)						\
+	: [__mem] "m" (*(mem)),						\
+	  [__old] "Ir" ((uint64_t)(old)),				\
+	  [__new] "Ir" (new)						\
+	: "memory");							\
+})
+
+/* For all "bool" routines, we return FALSE if exchange succesful.  */
+
+#define __arch_compare_and_exchange_bool_8_int(mem, new, old, mb1, mb2)	\
+({ unsigned long __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_8_int(mem, new, old, mb1, mb2);	\
+   !__cmp; })
+
+#define __arch_compare_and_exchange_bool_16_int(mem, new, old, mb1, mb2) \
+({ unsigned long __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_16_int(mem, new, old, mb1, mb2);	\
+   !__cmp; })
+
+#define __arch_compare_and_exchange_bool_32_int(mem, new, old, mb1, mb2) \
+({ unsigned long __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_32_int(mem, new, old, mb1, mb2);	\
+   !__cmp; })
+
+#define __arch_compare_and_exchange_bool_64_int(mem, new, old, mb1, mb2) \
+({ unsigned long __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_64_int(mem, new, old, mb1, mb2);	\
+   !__cmp; })
+
+/* For all "val" routines, return the old value whether exchange
+   successful or not.  */
+
+#define __arch_compare_and_exchange_val_8_int(mem, new, old, mb1, mb2)	\
+({ unsigned long __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_8_int(mem, new, old, mb1, mb2);	\
+   (typeof (*mem))__prev; })
+
+#define __arch_compare_and_exchange_val_16_int(mem, new, old, mb1, mb2) \
+({ unsigned long __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_16_int(mem, new, old, mb1, mb2);	\
+   (typeof (*mem))__prev; })
+
+#define __arch_compare_and_exchange_val_32_int(mem, new, old, mb1, mb2) \
+({ unsigned long __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_32_int(mem, new, old, mb1, mb2);	\
+   (typeof (*mem))__prev; })
+
+#define __arch_compare_and_exchange_val_64_int(mem, new, old, mb1, mb2) \
+({ unsigned long __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_64_int(mem, new, old, mb1, mb2);	\
+   (typeof (*mem))__prev; })
+
+/* Compare and exchange with "acquire" semantics, ie barrier after.  */
+
+#define atomic_compare_and_exchange_bool_acq(mem, new, old)	\
+  __atomic_bool_bysize (__arch_compare_and_exchange_bool, int,	\
+		        mem, new, old, "", __MB)
+
+#define atomic_compare_and_exchange_val_acq(mem, new, old)	\
+  __atomic_val_bysize (__arch_compare_and_exchange_val, int,	\
+		       mem, new, old, "", __MB)
+
+/* Compare and exchange with "release" semantics, ie barrier before.  */
+
+#define atomic_compare_and_exchange_bool_rel(mem, new, old)	\
+  __atomic_bool_bysize (__arch_compare_and_exchange_bool, int,	\
+		        mem, new, old, __MB, "")
+
+#define atomic_compare_and_exchange_val_rel(mem, new, old)	\
+  __atomic_val_bysize (__arch_compare_and_exchange_val, int,	\
+		       mem, new, old, __MB, "")
+
+
+/* Atomically store value and return the previous value.  */
+
+#define __arch_exchange_8_int(mem, value, mb1, mb2)			\
+({									\
+  unsigned long __ret, __tmp, __addr64, __sval;				\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"	andnot	%[__addr8],7,%[__addr64]\n"			\
+	"	insbl	%[__value],%[__addr8],%[__sval]\n"		\
+	"1:	ldq_l	%[__tmp],0(%[__addr64])\n"			\
+	"	extbl	%[__tmp],%[__addr8],%[__ret]\n"			\
+	"	mskbl	%[__tmp],%[__addr8],%[__tmp]\n"			\
+	"	or	%[__sval],%[__tmp],%[__tmp]\n"			\
+	"	stq_c	%[__tmp],0(%[__addr64])\n"			\
+	"	beq	%[__tmp],1b\n"					\
+		mb2							\
+	: [__ret] "=&r" (__ret),					\
+	  [__sval] "=&r" (__sval),					\
+	  [__tmp] "=&r" (__tmp),					\
+	  [__addr64] "=&r" (__addr64)					\
+	: [__addr8] "r" (mem),						\
+	  [__value] "r" (value)						\
+	: "memory");							\
+  __ret; })
+
+#define __arch_exchange_16_int(mem, value, mb1, mb2)			\
+({									\
+  unsigned long __ret, __tmp, __addr64, __sval;				\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"	andnot	%[__addr16],7,%[__addr64]\n"			\
+	"	inswl	%[__value],%[__addr16],%[__sval]\n"		\
+	"1:	ldq_l	%[__tmp],0(%[__addr64])\n"			\
+	"	extwl	%[__tmp],%[__addr16],%[__ret]\n"		\
+	"	mskwl	%[__tmp],%[__addr16],%[__tmp]\n"		\
+	"	or	%[__sval],%[__tmp],%[__tmp]\n"			\
+	"	stq_c	%[__tmp],0(%[__addr64])\n"			\
+	"	beq	%[__tmp],1b\n"					\
+		mb2							\
+	: [__ret] "=&r" (__ret),					\
+	  [__sval] "=&r" (__sval),					\
+	  [__tmp] "=&r" (__tmp),					\
+	  [__addr64] "=&r" (__addr64)					\
+	: [__addr16] "r" (mem),						\
+	  [__value] "r" (value)						\
+	: "memory");							\
+  __ret; })
+
+#define __arch_exchange_32_int(mem, value, mb1, mb2)			\
+({									\
+  signed int __ret, __tmp;						\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"1:	ldl_l	%[__ret],%[__mem]\n"				\
+	"	mov	%[__val],%[__tmp]\n"				\
+	"	stl_c	%[__tmp],%[__mem]\n"				\
+	"	beq	%[__tmp],1b\n"					\
+		mb2							\
+	: [__ret] "=&r" (__ret),					\
+	  [__tmp] "=&r" (__tmp)						\
+	: [__mem] "m" (*(mem)),						\
+	  [__val] "Ir" (value)						\
+	: "memory");							\
+  __ret; })
+
+#define __arch_exchange_64_int(mem, value, mb1, mb2)			\
+({									\
+  unsigned long __ret, __tmp;						\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"1:	ldq_l	%[__ret],%[__mem]\n"				\
+	"	mov	%[__val],%[__tmp]\n"				\
+	"	stq_c	%[__tmp],%[__mem]\n"				\
+	"	beq	%[__tmp],1b\n"					\
+		mb2							\
+	: [__ret] "=&r" (__ret),					\
+	  [__tmp] "=&r" (__tmp)						\
+	: [__mem] "m" (*(mem)),						\
+	  [__val] "Ir" (value)						\
+	: "memory");							\
+  __ret; })
+
+#define atomic_exchange_acq(mem, value) \
+  __atomic_val_bysize (__arch_exchange, int, mem, value, "", __MB)
+
+#define atomic_exchange_rel(mem, value) \
+  __atomic_val_bysize (__arch_exchange, int, mem, value, __MB, "")
+
+
+/* Atomically add value and return the previous (unincremented) value.  */
+
+#define __arch_exchange_and_add_8_int(mem, value, mb1, mb2) \
+  ({ __builtin_trap (); 0; })
+
+#define __arch_exchange_and_add_16_int(mem, value, mb1, mb2) \
+  ({ __builtin_trap (); 0; })
+
+#define __arch_exchange_and_add_32_int(mem, value, mb1, mb2)		\
+({									\
+  signed int __ret, __tmp;						\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"1:	ldl_l	%[__ret],%[__mem]\n"				\
+	"	addl	%[__ret],%[__val],%[__tmp]\n"			\
+	"	stl_c	%[__tmp],%[__mem]\n"				\
+	"	beq	%[__tmp],1b\n"					\
+		mb2							\
+	: [__ret] "=&r" (__ret),					\
+	  [__tmp] "=&r" (__tmp)						\
+	: [__mem] "m" (*(mem)),						\
+	  [__val] "Ir" ((signed int)(value))				\
+	: "memory");							\
+  __ret; })
+
+#define __arch_exchange_and_add_64_int(mem, value, mb1, mb2)		\
+({									\
+  unsigned long __ret, __tmp;						\
+  __asm__ __volatile__ (						\
+		mb1							\
+	"1:	ldq_l	%[__ret],%[__mem]\n"				\
+	"	addq	%[__ret],%[__val],%[__tmp]\n"			\
+	"	stq_c	%[__tmp],%[__mem]\n"				\
+	"	beq	%[__tmp],1b\n"					\
+		mb2							\
+	: [__ret] "=&r" (__ret),					\
+	  [__tmp] "=&r" (__tmp)						\
+	: [__mem] "m" (*(mem)),						\
+	  [__val] "Ir" ((unsigned long)(value))				\
+	: "memory");							\
+  __ret; })
+
+/* ??? Barrier semantics for atomic_exchange_and_add appear to be 
+   undefined.  Use full barrier for now, as that's safe.  */
+#define atomic_exchange_and_add(mem, value) \
+  __atomic_val_bysize (__arch_exchange_and_add, int, mem, value, __MB, __MB)
+
+
+/* ??? Blah, I'm lazy.  Implement these later.  Can do better than the
+   compare-and-exchange loop provided by generic code.
+
+#define atomic_decrement_if_positive(mem)
+#define atomic_bit_test_set(mem, bit)
+
+*/
+
+#ifndef UP
+# define atomic_full_barrier()	__asm ("mb" : : : "memory");
+# define atomic_read_barrier()	__asm ("mb" : : : "memory");
+# define atomic_write_barrier()	__asm ("wmb" : : : "memory");
+#endif

libc/sysdeps/linux/common/bits/atomic.h

@@ -0,0 +1,43 @@
+/* Copyright (C) 2003 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Ulrich Drepper <drepper@redhat.com>, 2003.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#ifndef _BITS_ATOMIC_H
+#define _BITS_ATOMIC_H	1
+
+/* We have by default no support for atomic operations.  So define
+   them non-atomic.  If this is a problem somebody will have to come
+   up with real definitions.  */
+
+/* The only basic operation needed is compare and exchange.  */
+#define atomic_compare_and_exchange_val_acq(mem, newval, oldval) \
+  ({ __typeof (mem) __gmemp = (mem);				      \
+     __typeof (*mem) __gret = *__gmemp;				      \
+     __typeof (*mem) __gnewval = (newval);			      \
+								      \
+     if (__gret == (oldval))					      \
+       *__gmemp = __gnewval;					      \
+     __gret; })
+
+#define atomic_compare_and_exchange_bool_acq(mem, newval, oldval) \
+  ({ __typeof (mem) __gmemp = (mem);				      \
+     __typeof (*mem) __gnewval = (newval);			      \
+								      \
+     *__gmemp == (oldval) ? (*__gmemp = __gnewval, 0) : 1; })
+
+#endif	/* bits/atomic.h */

libc/sysdeps/linux/i386/bits/atomic.h

@@ -0,0 +1,366 @@
+/* Copyright (C) 2002, 2003, 2004 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <stdint.h>
+
+
+typedef int8_t atomic8_t;
+typedef uint8_t uatomic8_t;
+typedef int_fast8_t atomic_fast8_t;
+typedef uint_fast8_t uatomic_fast8_t;
+
+typedef int16_t atomic16_t;
+typedef uint16_t uatomic16_t;
+typedef int_fast16_t atomic_fast16_t;
+typedef uint_fast16_t uatomic_fast16_t;
+
+typedef int32_t atomic32_t;
+typedef uint32_t uatomic32_t;
+typedef int_fast32_t atomic_fast32_t;
+typedef uint_fast32_t uatomic_fast32_t;
+
+typedef int64_t atomic64_t;
+typedef uint64_t uatomic64_t;
+typedef int_fast64_t atomic_fast64_t;
+typedef uint_fast64_t uatomic_fast64_t;
+
+typedef intptr_t atomicptr_t;
+typedef uintptr_t uatomicptr_t;
+typedef intmax_t atomic_max_t;
+typedef uintmax_t uatomic_max_t;
+
+
+#ifndef LOCK_PREFIX
+# ifdef UP
+#  define LOCK_PREFIX	/* nothing */
+# else
+#  define LOCK_PREFIX "lock;"
+# endif
+#endif
+
+
+#define __arch_compare_and_exchange_val_8_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret;						      \
+     __asm __volatile (LOCK_PREFIX "cmpxchgb %b2, %1"			      \
+		       : "=a" (ret), "=m" (*mem)			      \
+		       : "q" (newval), "m" (*mem), "0" (oldval));	      \
+     ret; })
+
+#define __arch_compare_and_exchange_val_16_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret;						      \
+     __asm __volatile (LOCK_PREFIX "cmpxchgw %w2, %1"			      \
+		       : "=a" (ret), "=m" (*mem)			      \
+		       : "r" (newval), "m" (*mem), "0" (oldval));	      \
+     ret; })
+
+#define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret;						      \
+     __asm __volatile (LOCK_PREFIX "cmpxchgl %2, %1"			      \
+		       : "=a" (ret), "=m" (*mem)			      \
+		       : "r" (newval), "m" (*mem), "0" (oldval));	      \
+     ret; })
+
+/* XXX We do not really need 64-bit compare-and-exchange.  At least
+   not in the moment.  Using it would mean causing portability
+   problems since not many other 32-bit architectures have support for
+   such an operation.  So don't define any code for now.  If it is
+   really going to be used the code below can be used on Intel Pentium
+   and later, but NOT on i486.  */
+#if 1
+# define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret = *(mem); abort (); ret = (newval); ret = (oldval); })
+#else
+# ifdef __PIC__
+#  define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret;						      \
+     __asm __volatile ("xchgl %2, %%ebx\n\t"				      \
+		       LOCK_PREFIX "cmpxchg8b %1\n\t"			      \
+		       "xchgl %2, %%ebx"				      \
+		       : "=A" (ret), "=m" (*mem)			      \
+		       : "DS" (((unsigned long long int) (newval))	      \
+			       & 0xffffffff),				      \
+			 "c" (((unsigned long long int) (newval)) >> 32),     \
+			 "m" (*mem), "a" (((unsigned long long int) (oldval)) \
+					  & 0xffffffff),		      \
+			 "d" (((unsigned long long int) (oldval)) >> 32));    \
+     ret; })
+# else
+#  define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret;						      \
+     __asm __volatile (LOCK_PREFIX "cmpxchg8b %1"			      \
+		       : "=A" (ret), "=m" (*mem)			      \
+		       : "b" (((unsigned long long int) (newval))	      \
+			      & 0xffffffff),				      \
+			 "c" (((unsigned long long int) (newval)) >> 32),     \
+			 "m" (*mem), "a" (((unsigned long long int) (oldval)) \
+					  & 0xffffffff),		      \
+			 "d" (((unsigned long long int) (oldval)) >> 32));    \
+     ret; })
+# endif
+#endif
+
+
+/* Note that we need no lock prefix.  */
+#define atomic_exchange_acq(mem, newvalue) \
+  ({ __typeof (*mem) result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile ("xchgb %b0, %1"				      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" (newvalue), "m" (*mem));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile ("xchgw %w0, %1"				      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" (newvalue), "m" (*mem));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile ("xchgl %0, %1"					      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" (newvalue), "m" (*mem));			      \
+     else								      \
+       {								      \
+	 result = 0;							      \
+	 abort ();							      \
+       }								      \
+     result; })
+
+
+#define atomic_exchange_and_add(mem, value) \
+  ({ __typeof (*mem) __result;						      \
+     __typeof (value) __addval = (value);				      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "xaddb %b0, %1"			      \
+			 : "=r" (__result), "=m" (*mem)			      \
+			 : "0" (__addval), "m" (*mem));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "xaddw %w0, %1"			      \
+			 : "=r" (__result), "=m" (*mem)			      \
+			 : "0" (__addval), "m" (*mem));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "xaddl %0, %1"			      \
+			 : "=r" (__result), "=m" (*mem)			      \
+			 : "0" (__addval), "m" (*mem));			      \
+     else								      \
+       {								      \
+	 __typeof (mem) __memp = (mem);					      \
+	 __typeof (*mem) __tmpval;					      \
+	 __result = *__memp;						      \
+	 do								      \
+	   __tmpval = __result;						      \
+	 while ((__result = __arch_compare_and_exchange_val_64_acq	      \
+		 (__memp, __result + __addval, __result)) == __tmpval);	      \
+       }								      \
+     __result; })
+
+
+#define atomic_add(mem, value) \
+  (void) ({ if (__builtin_constant_p (value) && (value) == 1)		      \
+	      atomic_increment (mem);					      \
+	    else if (__builtin_constant_p (value) && (value) == -1)	      \
+	      atomic_decrement (mem);					      \
+	    else if (sizeof (*mem) == 1)				      \
+	      __asm __volatile (LOCK_PREFIX "addb %b1, %0"		      \
+				: "=m" (*mem)				      \
+				: "ir" (value), "m" (*mem));		      \
+	    else if (sizeof (*mem) == 2)				      \
+	      __asm __volatile (LOCK_PREFIX "addw %w1, %0"		      \
+				: "=m" (*mem)				      \
+				: "ir" (value), "m" (*mem));		      \
+	    else if (sizeof (*mem) == 4)				      \
+	      __asm __volatile (LOCK_PREFIX "addl %1, %0"		      \
+				: "=m" (*mem)				      \
+				: "ir" (value), "m" (*mem));		      \
+	    else							      \
+	      {								      \
+		__typeof (value) __addval = (value);			      \
+		__typeof (mem) __memp = (mem);				      \
+		__typeof (*mem) __oldval = *__memp;			      \
+		__typeof (*mem) __tmpval;				      \
+		do							      \
+		  __tmpval = __oldval;					      \
+		while ((__oldval = __arch_compare_and_exchange_val_64_acq     \
+		       (__memp, __oldval + __addval, __oldval)) == __tmpval); \
+	      }								      \
+	    })
+
+
+#define atomic_add_negative(mem, value) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "addb %b2, %0; sets %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "iq" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "addw %w2, %0; sets %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "addl %2, %0; sets %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else								      \
+       abort ();							      \
+     __result; })
+
+
+#define atomic_add_zero(mem, value) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "addb %b2, %0; setz %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "addw %w2, %0; setz %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "addl %2, %0; setz %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else								      \
+       abort ();							      \
+     __result; })
+
+
+#define atomic_increment(mem) \
+  (void) ({ if (sizeof (*mem) == 1)					      \
+	      __asm __volatile (LOCK_PREFIX "incb %b0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else if (sizeof (*mem) == 2)				      \
+	      __asm __volatile (LOCK_PREFIX "incw %w0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else if (sizeof (*mem) == 4)				      \
+	      __asm __volatile (LOCK_PREFIX "incl %0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else							      \
+	      {								      \
+		__typeof (mem) __memp = (mem);				      \
+		__typeof (*mem) __oldval = *__memp;			      \
+		__typeof (*mem) __tmpval;				      \
+		do							      \
+		  __tmpval = __oldval;					      \
+		while ((__oldval = __arch_compare_and_exchange_val_64_acq     \
+		       (__memp, __oldval + 1, __oldval)) == __tmpval);	      \
+	      }								      \
+	    })
+
+
+#define atomic_increment_and_test(mem) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "incb %0; sete %b1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "incw %0; sete %w1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "incl %0; sete %1"			      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else								      \
+       abort ();							      \
+     __result; })
+
+
+#define atomic_decrement(mem) \
+  (void) ({ if (sizeof (*mem) == 1)					      \
+	      __asm __volatile (LOCK_PREFIX "decb %b0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else if (sizeof (*mem) == 2)				      \
+	      __asm __volatile (LOCK_PREFIX "decw %w0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else if (sizeof (*mem) == 4)				      \
+	      __asm __volatile (LOCK_PREFIX "decl %0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else							      \
+	      {								      \
+		__typeof (mem) __memp = (mem);				      \
+		__typeof (*mem) __oldval = *__memp;			      \
+		__typeof (*mem) __tmpval;				      \
+		do							      \
+		  __tmpval = __oldval;					      \
+		while ((__oldval = __arch_compare_and_exchange_val_64_acq     \
+		       (__memp, __oldval - 1, __oldval)) == __tmpval); 	      \
+	      }								      \
+	    })
+
+
+#define atomic_decrement_and_test(mem) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "decb %b0; sete %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "decw %w0; sete %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "decl %0; sete %1"			      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else								      \
+       abort ();							      \
+     __result; })
+
+
+#define atomic_bit_set(mem, bit) \
+  (void) ({ if (sizeof (*mem) == 1)					      \
+	      __asm __volatile (LOCK_PREFIX "orb %b2, %0"		      \
+				: "=m" (*mem)				      \
+				: "m" (*mem), "ir" (1 << (bit)));	      \
+	    else if (sizeof (*mem) == 2)				      \
+	      __asm __volatile (LOCK_PREFIX "orw %w2, %0"		      \
+				: "=m" (*mem)				      \
+				: "m" (*mem), "ir" (1 << (bit)));	      \
+	    else if (sizeof (*mem) == 4)				      \
+	      __asm __volatile (LOCK_PREFIX "orl %2, %0"		      \
+				: "=m" (*mem)				      \
+				: "m" (*mem), "ir" (1 << (bit)));	      \
+	    else							      \
+	      abort ();							      \
+	    })
+
+
+#define atomic_bit_test_set(mem, bit) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "btsb %3, %1; setc %0"		      \
+			 : "=q" (__result), "=m" (*mem)			      \
+			 : "m" (*mem), "ir" (bit));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "btsw %3, %1; setc %0"		      \
+			 : "=q" (__result), "=m" (*mem)			      \
+			 : "m" (*mem), "ir" (bit));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "btsl %3, %1; setc %0"		      \
+			 : "=q" (__result), "=m" (*mem)			      \
+			 : "m" (*mem), "ir" (bit));			      \
+     else							      	      \
+       abort ();							      \
+     __result; })
+
+
+#define atomic_delay() asm ("rep; nop")

libc/sysdeps/linux/mips/bits/atomic.h

@@ -0,0 +1,303 @@
+/* Low-level functions for atomic operations. Mips version.
+   Copyright (C) 2005 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
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#ifndef _MIPS_BITS_ATOMIC_H
+#define _MIPS_BITS_ATOMIC_H 1
+
+#include <inttypes.h>
+#include <sgidefs.h>
+
+typedef int32_t atomic32_t;
+typedef uint32_t uatomic32_t;
+typedef int_fast32_t atomic_fast32_t;
+typedef uint_fast32_t uatomic_fast32_t;
+
+typedef int64_t atomic64_t;
+typedef uint64_t uatomic64_t;
+typedef int_fast64_t atomic_fast64_t;
+typedef uint_fast64_t uatomic_fast64_t;
+
+typedef intptr_t atomicptr_t;
+typedef uintptr_t uatomicptr_t;
+typedef intmax_t atomic_max_t;
+typedef uintmax_t uatomic_max_t;
+
+#if _MIPS_SIM == _ABIO32
+#define MIPS_PUSH_MIPS2 ".set	mips2\n\t"
+#else
+#define MIPS_PUSH_MIPS2
+#endif
+
+/* See the comments in <sys/asm.h> about the use of the sync instruction.  */
+#ifndef MIPS_SYNC
+# define MIPS_SYNC	sync
+#endif
+
+#define MIPS_SYNC_STR_2(X) #X
+#define MIPS_SYNC_STR_1(X) MIPS_SYNC_STR_2(X)
+#define MIPS_SYNC_STR MIPS_SYNC_STR_1(MIPS_SYNC)
+
+/* Compare and exchange.  For all of the "xxx" routines, we expect a
+   "__prev" and a "__cmp" variable to be provided by the enclosing scope,
+   in which values are returned.  */
+
+#define __arch_compare_and_exchange_xxx_8_int(mem, newval, oldval, rel, acq) \
+  (abort (), __prev = __cmp = 0)
+
+#define __arch_compare_and_exchange_xxx_16_int(mem, newval, oldval, rel, acq) \
+  (abort (), __prev = __cmp = 0)
+
+#define __arch_compare_and_exchange_xxx_32_int(mem, newval, oldval, rel, acq) \
+     __asm__ __volatile__ (						      \
+     ".set	push\n\t"						      \
+     MIPS_PUSH_MIPS2							      \
+     rel	"\n"							      \
+     "1:\t"								      \
+     "ll	%0,%4\n\t"						      \
+     "move	%1,$0\n\t"						      \
+     "bne	%0,%2,2f\n\t"						      \
+     "move	%1,%3\n\t"						      \
+     "sc	%1,%4\n\t"						      \
+     "beqz	%1,1b\n"						      \
+     acq	"\n\t"							      \
+     ".set	pop\n"							      \
+     "2:\n\t"								      \
+	      : "=&r" (__prev), "=&r" (__cmp)				      \
+	      : "r" (oldval), "r" (newval), "m" (*mem)			      \
+	      : "memory")
+
+#if _MIPS_SIM == _ABIO32
+/* We can't do an atomic 64-bit operation in O32.  */
+#define __arch_compare_and_exchange_xxx_64_int(mem, newval, oldval, rel, acq) \
+  (abort (), __prev = __cmp = 0)
+#else
+#define __arch_compare_and_exchange_xxx_64_int(mem, newval, oldval, rel, acq) \
+     __asm__ __volatile__ ("\n"						      \
+     ".set	push\n\t"						      \
+     MIPS_PUSH_MIPS2							      \
+     rel	"\n"							      \
+     "1:\t"								      \
+     "lld	%0,%4\n\t"						      \
+     "move	%1,$0\n\t"						      \
+     "bne	%0,%2,2f\n\t"						      \
+     "move	%1,%3\n\t"						      \
+     "scd	%1,%4\n\t"						      \
+     "beqz	%1,1b\n"						      \
+     acq	"\n\t"							      \
+     ".set	pop\n"							      \
+     "2:\n\t"								      \
+	      : "=&r" (__prev), "=&r" (__cmp)				      \
+	      : "r" (oldval), "r" (newval), "m" (*mem)			      \
+	      : "memory")
+#endif
+
+/* For all "bool" routines, we return FALSE if exchange succesful.  */
+
+#define __arch_compare_and_exchange_bool_8_int(mem, new, old, rel, acq)	\
+({ typeof (*mem) __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_8_int(mem, new, old, rel, acq);	\
+   !__cmp; })
+
+#define __arch_compare_and_exchange_bool_16_int(mem, new, old, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_16_int(mem, new, old, rel, acq);	\
+   !__cmp; })
+
+#define __arch_compare_and_exchange_bool_32_int(mem, new, old, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_32_int(mem, new, old, rel, acq);	\
+   !__cmp; })
+
+#define __arch_compare_and_exchange_bool_64_int(mem, new, old, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_64_int(mem, new, old, rel, acq);	\
+   !__cmp; })
+
+/* For all "val" routines, return the old value whether exchange
+   successful or not.  */
+
+#define __arch_compare_and_exchange_val_8_int(mem, new, old, rel, acq)	\
+({ typeof (*mem) __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_8_int(mem, new, old, rel, acq);	\
+   (typeof (*mem))__prev; })
+
+#define __arch_compare_and_exchange_val_16_int(mem, new, old, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_16_int(mem, new, old, rel, acq);	\
+   (typeof (*mem))__prev; })
+
+#define __arch_compare_and_exchange_val_32_int(mem, new, old, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_32_int(mem, new, old, rel, acq);	\
+   (typeof (*mem))__prev; })
+
+#define __arch_compare_and_exchange_val_64_int(mem, new, old, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					\
+   __arch_compare_and_exchange_xxx_64_int(mem, new, old, rel, acq);	\
+   (typeof (*mem))__prev; })
+
+/* Compare and exchange with "acquire" semantics, ie barrier after.  */
+
+#define atomic_compare_and_exchange_bool_acq(mem, new, old)	\
+  __atomic_bool_bysize (__arch_compare_and_exchange_bool, int,	\
+		        mem, new, old, "", MIPS_SYNC_STR)
+
+#define atomic_compare_and_exchange_val_acq(mem, new, old)	\
+  __atomic_val_bysize (__arch_compare_and_exchange_val, int,	\
+		       mem, new, old, "", MIPS_SYNC_STR)
+
+/* Compare and exchange with "release" semantics, ie barrier before.  */
+
+#define atomic_compare_and_exchange_bool_rel(mem, new, old)	\
+  __atomic_bool_bysize (__arch_compare_and_exchange_bool, int,	\
+		        mem, new, old, MIPS_SYNC_STR, "")
+
+#define atomic_compare_and_exchange_val_rel(mem, new, old)	\
+  __atomic_val_bysize (__arch_compare_and_exchange_val, int,	\
+		       mem, new, old, MIPS_SYNC_STR, "")
+
+
+
+/* Atomic exchange (without compare).  */
+
+#define __arch_exchange_xxx_8_int(mem, newval, rel, acq) \
+  (abort (), 0)
+
+#define __arch_exchange_xxx_16_int(mem, newval, rel, acq) \
+  (abort (), 0)
+
+#define __arch_exchange_xxx_32_int(mem, newval, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					      \
+     __asm__ __volatile__ ("\n"						      \
+     ".set	push\n\t"						      \
+     MIPS_PUSH_MIPS2							      \
+     rel	"\n"							      \
+     "1:\t"								      \
+     "ll	%0,%3\n\t"						      \
+     "move	%1,%2\n\t"						      \
+     "sc	%1,%3\n\t"						      \
+     "beqz	%1,1b\n"						      \
+     acq	"\n\t"							      \
+     ".set	pop\n"							      \
+     "2:\n\t"								      \
+	      : "=&r" (__prev), "=&r" (__cmp)				      \
+	      : "r" (newval), "m" (*mem)				      \
+	      : "memory");						      \
+  __prev; })
+
+#if _MIPS_SIM == _ABIO32
+/* We can't do an atomic 64-bit operation in O32.  */
+#define __arch_exchange_xxx_64_int(mem, newval, rel, acq) \
+  (abort (), 0)
+#else
+#define __arch_exchange_xxx_64_int(mem, newval, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					      \
+     __asm__ __volatile__ ("\n"						      \
+     ".set	push\n\t"						      \
+     MIPS_PUSH_MIPS2							      \
+     rel	"\n"							      \
+     "1:\n"								      \
+     "lld	%0,%3\n\t"						      \
+     "move	%1,%2\n\t"						      \
+     "scd	%1,%3\n\t"						      \
+     "beqz	%1,1b\n"						      \
+     acq	"\n\t"							      \
+     ".set	pop\n"							      \
+     "2:\n\t"								      \
+	      : "=&r" (__prev), "=&r" (__cmp)				      \
+	      : "r" (newval), "m" (*mem)				      \
+	      : "memory");						      \
+  __prev; })
+#endif
+
+#define atomic_exchange_acq(mem, value) \
+  __atomic_val_bysize (__arch_exchange_xxx, int, mem, value, "", MIPS_SYNC_STR)
+
+#define atomic_exchange_rel(mem, value) \
+  __atomic_val_bysize (__arch_exchange_xxx, int, mem, value, MIPS_SYNC_STR, "")
+
+
+/* Atomically add value and return the previous (unincremented) value.  */
+
+#define __arch_exchange_and_add_8_int(mem, newval, rel, acq) \
+  (abort (), (typeof(*mem)) 0)
+
+#define __arch_exchange_and_add_16_int(mem, newval, rel, acq) \
+  (abort (), (typeof(*mem)) 0)
+
+#define __arch_exchange_and_add_32_int(mem, value, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					      \
+     __asm__ __volatile__ ("\n"						      \
+     ".set	push\n\t"						      \
+     MIPS_PUSH_MIPS2							      \
+     rel	"\n"							      \
+     "1:\t"								      \
+     "ll	%0,%3\n\t"						      \
+     "addu	%1,%0,%2\n\t"						      \
+     "sc	%1,%3\n\t"						      \
+     "beqz	%1,1b\n"						      \
+     acq	"\n\t"							      \
+     ".set	pop\n"							      \
+     "2:\n\t"								      \
+	      : "=&r" (__prev), "=&r" (__cmp)				      \
+	      : "r" (value), "m" (*mem)					      \
+	      : "memory");						      \
+  __prev; })
+
+#if _MIPS_SIM == _ABIO32
+/* We can't do an atomic 64-bit operation in O32.  */
+#define __arch_exchange_and_add_64_int(mem, value, rel, acq) \
+  (abort (), (typeof(*mem)) 0)
+#else
+#define __arch_exchange_and_add_64_int(mem, value, rel, acq) \
+({ typeof (*mem) __prev; int __cmp;					      \
+     __asm__ __volatile__ (						      \
+     ".set	push\n\t"						      \
+     MIPS_PUSH_MIPS2							      \
+     rel	"\n"							      \
+     "1:\t"								      \
+     "lld	%0,%3\n\t"						      \
+     "daddu	%1,%0,%2\n\t"						      \
+     "scd	%1,%3\n\t"						      \
+     "beqz	%1,1b\n"						      \
+     acq	"\n\t"							      \
+     ".set	pop\n"							      \
+     "2:\n\t"								      \
+	      : "=&r" (__prev), "=&r" (__cmp)				      \
+	      : "r" (value), "m" (*mem)					      \
+	      : "memory");						      \
+  __prev; })
+#endif
+
+/* ??? Barrier semantics for atomic_exchange_and_add appear to be 
+   undefined.  Use full barrier for now, as that's safe.  */
+#define atomic_exchange_and_add(mem, value) \
+  __atomic_val_bysize (__arch_exchange_and_add, int, mem, value,	      \
+		       MIPS_SYNC_STR, MIPS_SYNC_STR)
+
+/* TODO: More atomic operations could be implemented efficiently; only the
+   basic requirements are done.  */
+
+#define atomic_full_barrier() \
+  __asm__ __volatile__ (".set push\n\t"					      \
+			MIPS_PUSH_MIPS2					      \
+			MIPS_SYNC_STR "\n\t"				      \
+			".set pop" : : : "memory")
+
+#endif /* bits/atomic.h */

libc/sysdeps/linux/sh/bits/atomic.h

@@ -0,0 +1,419 @@
+/* Atomic operations used inside libc.  Linux/SH version.
+   Copyright (C) 2003 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
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <stdint.h>
+
+
+typedef int8_t atomic8_t;
+typedef uint8_t uatomic8_t;
+typedef int_fast8_t atomic_fast8_t;
+typedef uint_fast8_t uatomic_fast8_t;
+
+typedef int16_t atomic16_t;
+typedef uint16_t uatomic16_t;
+typedef int_fast16_t atomic_fast16_t;
+typedef uint_fast16_t uatomic_fast16_t;
+
+typedef int32_t atomic32_t;
+typedef uint32_t uatomic32_t;
+typedef int_fast32_t atomic_fast32_t;
+typedef uint_fast32_t uatomic_fast32_t;
+
+typedef int64_t atomic64_t;
+typedef uint64_t uatomic64_t;
+typedef int_fast64_t atomic_fast64_t;
+typedef uint_fast64_t uatomic_fast64_t;
+
+typedef intptr_t atomicptr_t;
+typedef uintptr_t uatomicptr_t;
+typedef intmax_t atomic_max_t;
+typedef uintmax_t uatomic_max_t;
+
+/* SH kernel has implemented a gUSA ("g" User Space Atomicity) support
+   for the user space atomicity. The atomicity macros use this scheme.
+
+  Reference:
+    Niibe Yutaka, "gUSA: Simple and Efficient User Space Atomicity
+    Emulation with Little Kernel Modification", Linux Conference 2002,
+    Japan. http://lc.linux.or.jp/lc2002/papers/niibe0919h.pdf (in
+    Japanese).
+
+    B.N. Bershad, D. Redell, and J. Ellis, "Fast Mutual Exclusion for
+    Uniprocessors",  Proceedings of the Fifth Architectural Support for
+    Programming Languages and Operating Systems (ASPLOS), pp. 223-233,
+    October 1992. http://www.cs.washington.edu/homes/bershad/Papers/Rcs.ps
+
+  SuperH ABI:
+      r15:    -(size of atomic instruction sequence) < 0
+      r0:     end point
+      r1:     saved stack pointer
+*/
+
+#define __arch_compare_and_exchange_val_8_acq(mem, newval, oldval) \
+  ({ __typeof (*(mem)) __result; \
+     __asm __volatile ("\
+	.align 2\n\
+	mova 1f,r0\n\
+	nop\n\
+	mov r15,r1\n\
+	mov #-8,r15\n\
+     0: mov.b @%1,%0\n\
+	cmp/eq %0,%3\n\
+	bf 1f\n\
+	mov.b %2,@%1\n\
+     1: mov r1,r15"\
+	: "=&r" (__result) : "r" (mem), "r" (newval), "r" (oldval) \
+	: "r0", "r1", "t", "memory"); \
+     __result; })
+
+#define __arch_compare_and_exchange_val_16_acq(mem, newval, oldval) \
+  ({ __typeof (*(mem)) __result; \
+     __asm __volatile ("\
+	.align 2\n\
+	mova 1f,r0\n\
+	nop\n\
+	mov r15,r1\n\
+	mov #-8,r15\n\
+     0: mov.w @%1,%0\n\
+	cmp/eq %0,%3\n\
+	bf 1f\n\
+	mov.w %2,@%1\n\
+     1: mov r1,r15"\
+	: "=&r" (__result) : "r" (mem), "r" (newval), "r" (oldval) \
+	: "r0", "r1", "t", "memory"); \
+     __result; })
+
+#define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval) \
+  ({ __typeof (*(mem)) __result; \
+     __asm __volatile ("\
+	.align 2\n\
+	mova 1f,r0\n\
+	nop\n\
+	mov r15,r1\n\
+	mov #-8,r15\n\
+     0: mov.l @%1,%0\n\
+	cmp/eq %0,%3\n\
+	bf 1f\n\
+	mov.l %2,@%1\n\
+     1: mov r1,r15"\
+	: "=&r" (__result) : "r" (mem), "r" (newval), "r" (oldval) \
+	: "r0", "r1", "t", "memory"); \
+     __result; })
+
+/* XXX We do not really need 64-bit compare-and-exchange.  At least
+   not in the moment.  Using it would mean causing portability
+   problems since not many other 32-bit architectures have support for
+   such an operation.  So don't define any code for now.  */
+
+# define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
+  (abort (), (__typeof (*mem)) 0)
+
+#define atomic_exchange_and_add(mem, value) \
+  ({ __typeof (*(mem)) __result, __tmp, __value = (value); \
+     if (sizeof (*(mem)) == 1) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  mov r15,r1\n\
+	  mov #-6,r15\n\
+       0: mov.b @%2,%0\n\
+	  add %0,%1\n\
+	  mov.b %1,@%2\n\
+       1: mov r1,r15"\
+	: "=&r" (__result), "=&r" (__tmp) : "r" (mem), "1" (__value) \
+	: "r0", "r1", "memory"); \
+     else if (sizeof (*(mem)) == 2) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  mov r15,r1\n\
+	  mov #-6,r15\n\
+       0: mov.w @%2,%0\n\
+	  add %0,%1\n\
+	  mov.w %1,@%2\n\
+       1: mov r1,r15"\
+	: "=&r" (__result), "=&r" (__tmp) : "r" (mem), "1" (__value) \
+	: "r0", "r1", "memory"); \
+     else if (sizeof (*(mem)) == 4) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  mov r15,r1\n\
+	  mov #-6,r15\n\
+       0: mov.l @%2,%0\n\
+	  add %0,%1\n\
+	  mov.l %1,@%2\n\
+       1: mov r1,r15"\
+	: "=&r" (__result), "=&r" (__tmp) : "r" (mem), "1" (__value) \
+	: "r0", "r1", "memory"); \
+     else \
+       { \
+	 __typeof (mem) memp = (mem); \
+	 do \
+	   __result = *memp; \
+	 while (__arch_compare_and_exchange_val_64_acq \
+		 (memp,	__result + __value, __result) == __result); \
+	 (void) __value; \
+       } \
+     __result; })
+
+#define atomic_add(mem, value) \
+  (void) ({ __typeof (*(mem)) __tmp, __value = (value); \
+	    if (sizeof (*(mem)) == 1) \
+	      __asm __volatile ("\
+		.align 2\n\
+		mova 1f,r0\n\
+		mov r15,r1\n\
+		mov #-6,r15\n\
+	     0: mov.b @%1,r2\n\
+		add r2,%0\n\
+		mov.b %0,@%1\n\
+	     1: mov r1,r15"\
+		: "=&r" (__tmp) : "r" (mem), "0" (__value) \
+		: "r0", "r1", "r2", "memory"); \
+	    else if (sizeof (*(mem)) == 2) \
+	      __asm __volatile ("\
+		.align 2\n\
+		mova 1f,r0\n\
+		mov r15,r1\n\
+		mov #-6,r15\n\
+	     0: mov.w @%1,r2\n\
+		add r2,%0\n\
+		mov.w %0,@%1\n\
+	     1: mov r1,r15"\
+		: "=&r" (__tmp) : "r" (mem), "0" (__value) \
+		: "r0", "r1", "r2", "memory"); \
+	    else if (sizeof (*(mem)) == 4) \
+	      __asm __volatile ("\
+		.align 2\n\
+		mova 1f,r0\n\
+		mov r15,r1\n\
+		mov #-6,r15\n\
+	     0: mov.l @%1,r2\n\
+		add r2,%0\n\
+		mov.l %0,@%1\n\
+	     1: mov r1,r15"\
+		: "=&r" (__tmp) : "r" (mem), "0" (__value) \
+		: "r0", "r1", "r2", "memory"); \
+	    else \
+	      { \
+		__typeof (*(mem)) oldval; \
+		__typeof (mem) memp = (mem); \
+		do \
+		  oldval = *memp; \
+		while (__arch_compare_and_exchange_val_64_acq \
+			(memp, oldval + __value, oldval) == oldval); \
+		(void) __value; \
+	      } \
+	    })
+
+#define atomic_add_negative(mem, value) \
+  ({ unsigned char __result; \
+     __typeof (*(mem)) __tmp, __value = (value); \
+     if (sizeof (*(mem)) == 1) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  mov r15,r1\n\
+	  mov #-6,r15\n\
+       0: mov.b @%2,r2\n\
+	  add r2,%1\n\
+	  mov.b %1,@%2\n\
+       1: mov r1,r15\n\
+	  shal %1\n\
+	  movt %0"\
+	: "=r" (__result), "=&r" (__tmp) : "r" (mem), "1" (__value) \
+	: "r0", "r1", "r2", "t", "memory"); \
+     else if (sizeof (*(mem)) == 2) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  mov r15,r1\n\
+	  mov #-6,r15\n\
+       0: mov.w @%2,r2\n\
+	  add r2,%1\n\
+	  mov.w %1,@%2\n\
+       1: mov r1,r15\n\
+	  shal %1\n\
+	  movt %0"\
+	: "=r" (__result), "=&r" (__tmp) : "r" (mem), "1" (__value) \
+	: "r0", "r1", "r2", "t", "memory"); \
+     else if (sizeof (*(mem)) == 4) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  mov r15,r1\n\
+	  mov #-6,r15\n\
+       0: mov.l @%2,r2\n\
+	  add r2,%1\n\
+	  mov.l %1,@%2\n\
+       1: mov r1,r15\n\
+	  shal %1\n\
+	  movt %0"\
+	: "=r" (__result), "=&r" (__tmp) : "r" (mem), "1" (__value) \
+	: "r0", "r1", "r2", "t", "memory"); \
+     else \
+       abort (); \
+     __result; })
+
+#define atomic_add_zero(mem, value) \
+  ({ unsigned char __result; \
+     __typeof (*(mem)) __tmp, __value = (value); \
+     if (sizeof (*(mem)) == 1) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  mov r15,r1\n\
+	  mov #-6,r15\n\
+       0: mov.b @%2,r2\n\
+	  add r2,%1\n\
+	  mov.b %1,@%2\n\
+       1: mov r1,r15\n\
+	  tst %1,%1\n\
+	  movt %0"\
+	: "=r" (__result), "=&r" (__tmp) : "r" (mem), "1" (__value) \
+	: "r0", "r1", "r2", "t", "memory"); \
+     else if (sizeof (*(mem)) == 2) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  mov r15,r1\n\
+	  mov #-6,r15\n\
+       0: mov.w @%2,r2\n\
+	  add r2,%1\n\
+	  mov.w %1,@%2\n\
+       1: mov r1,r15\n\
+	  tst %1,%1\n\
+	  movt %0"\
+	: "=r" (__result), "=&r" (__tmp) : "r" (mem), "1" (__value) \
+	: "r0", "r1", "r2", "t", "memory"); \
+     else if (sizeof (*(mem)) == 4) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  mov r15,r1\n\
+	  mov #-6,r15\n\
+       0: mov.l @%2,r2\n\
+	  add r2,%1\n\
+	  mov.l %1,@%2\n\
+       1: mov r1,r15\n\
+	  tst %1,%1\n\
+	  movt %0"\
+	: "=r" (__result), "=&r" (__tmp) : "r" (mem), "1" (__value) \
+	: "r0", "r1", "r2", "t", "memory"); \
+     else \
+       abort (); \
+     __result; })
+
+#define atomic_increment_and_test(mem) atomic_add_zero((mem), 1)
+#define atomic_decrement_and_test(mem) atomic_add_zero((mem), -1)
+
+#define atomic_bit_set(mem, bit) \
+  (void) ({ unsigned int __mask = 1 << (bit); \
+	    if (sizeof (*(mem)) == 1) \
+	      __asm __volatile ("\
+		.align 2\n\
+		mova 1f,r0\n\
+		mov r15,r1\n\
+		mov #-6,r15\n\
+	     0: mov.b @%0,r2\n\
+		or %1,r2\n\
+		mov.b r2,@%0\n\
+	     1: mov r1,r15"\
+		: : "r" (mem), "r" (__mask) \
+		: "r0", "r1", "r2", "memory"); \
+	    else if (sizeof (*(mem)) == 2) \
+	      __asm __volatile ("\
+		.align 2\n\
+		mova 1f,r0\n\
+		mov r15,r1\n\
+		mov #-6,r15\n\
+	     0: mov.w @%0,r2\n\
+		or %1,r2\n\
+		mov.w r2,@%0\n\
+	     1: mov r1,r15"\
+		: : "r" (mem), "r" (__mask) \
+		: "r0", "r1", "r2", "memory"); \
+	    else if (sizeof (*(mem)) == 4) \
+	      __asm __volatile ("\
+		.align 2\n\
+		mova 1f,r0\n\
+		mov r15,r1\n\
+		mov #-6,r15\n\
+	     0: mov.l @%0,r2\n\
+		or %1,r2\n\
+		mov.l r2,@%0\n\
+	     1: mov r1,r15"\
+		: : "r" (mem), "r" (__mask) \
+		: "r0", "r1", "r2", "memory"); \
+	    else \
+	      abort (); \
+	    })
+
+#define atomic_bit_test_set(mem, bit) \
+  ({ unsigned int __mask = 1 << (bit); \
+     unsigned int __result = __mask; \
+     if (sizeof (*(mem)) == 1) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  nop\n\
+	  mov r15,r1\n\
+	  mov #-8,r15\n\
+       0: mov.b @%2,r2\n\
+	  or r2,%1\n\
+	  and r2,%0\n\
+	  mov.b %1,@%2\n\
+       1: mov r1,r15"\
+	: "=&r" (__result), "=&r" (__mask) \
+	: "r" (mem), "0" (__result), "1" (__mask) \
+	: "r0", "r1", "r2", "memory"); \
+     else if (sizeof (*(mem)) == 2) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  nop\n\
+	  mov r15,r1\n\
+	  mov #-8,r15\n\
+       0: mov.w @%2,r2\n\
+	  or r2,%1\n\
+	  and r2,%0\n\
+	  mov.w %1,@%2\n\
+       1: mov r1,r15"\
+	: "=&r" (__result), "=&r" (__mask) \
+	: "r" (mem), "0" (__result), "1" (__mask) \
+	: "r0", "r1", "r2", "memory"); \
+     else if (sizeof (*(mem)) == 4) \
+       __asm __volatile ("\
+	  .align 2\n\
+	  mova 1f,r0\n\
+	  nop\n\
+	  mov r15,r1\n\
+	  mov #-8,r15\n\
+       0: mov.l @%2,r2\n\
+	  or r2,%1\n\
+	  and r2,%0\n\
+	  mov.l %1,@%2\n\
+       1: mov r1,r15"\
+	: "=&r" (__result), "=&r" (__mask) \
+	: "r" (mem), "0" (__result), "1" (__mask) \
+	: "r0", "r1", "r2", "memory"); \
+     else \
+       abort (); \
+     __result; })

libc/sysdeps/linux/x86_64/bits/atomic.h

@@ -0,0 +1,324 @@
+/* Copyright (C) 2002, 2003, 2004 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <stdint.h>
+
+
+typedef int8_t atomic8_t;
+typedef uint8_t uatomic8_t;
+typedef int_fast8_t atomic_fast8_t;
+typedef uint_fast8_t uatomic_fast8_t;
+
+typedef int16_t atomic16_t;
+typedef uint16_t uatomic16_t;
+typedef int_fast16_t atomic_fast16_t;
+typedef uint_fast16_t uatomic_fast16_t;
+
+typedef int32_t atomic32_t;
+typedef uint32_t uatomic32_t;
+typedef int_fast32_t atomic_fast32_t;
+typedef uint_fast32_t uatomic_fast32_t;
+
+typedef int64_t atomic64_t;
+typedef uint64_t uatomic64_t;
+typedef int_fast64_t atomic_fast64_t;
+typedef uint_fast64_t uatomic_fast64_t;
+
+typedef intptr_t atomicptr_t;
+typedef uintptr_t uatomicptr_t;
+typedef intmax_t atomic_max_t;
+typedef uintmax_t uatomic_max_t;
+
+
+#ifndef LOCK_PREFIX
+# ifdef UP
+#  define LOCK_PREFIX	/* nothing */
+# else
+#  define LOCK_PREFIX "lock;"
+# endif
+#endif
+
+
+#define __arch_compare_and_exchange_val_8_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret;						      \
+     __asm __volatile (LOCK_PREFIX "cmpxchgb %b2, %1"			      \
+		       : "=a" (ret), "=m" (*mem)			      \
+		       : "q" (newval), "m" (*mem), "0" (oldval));	      \
+     ret; })
+
+#define __arch_compare_and_exchange_val_16_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret;						      \
+     __asm __volatile (LOCK_PREFIX "cmpxchgw %w2, %1"			      \
+		       : "=a" (ret), "=m" (*mem)			      \
+		       : "r" (newval), "m" (*mem), "0" (oldval));	      \
+     ret; })
+
+#define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret;						      \
+     __asm __volatile (LOCK_PREFIX "cmpxchgl %2, %1"			      \
+		       : "=a" (ret), "=m" (*mem)			      \
+		       : "r" (newval), "m" (*mem), "0" (oldval));	      \
+     ret; })
+
+#define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
+  ({ __typeof (*mem) ret;						      \
+     __asm __volatile (LOCK_PREFIX "cmpxchgq %q2, %1"			      \
+		       : "=a" (ret), "=m" (*mem)			      \
+		       : "r" ((long) (newval)), "m" (*mem),		      \
+			 "0" ((long) (oldval)));			      \
+     ret; })
+
+
+/* Note that we need no lock prefix.  */
+#define atomic_exchange_acq(mem, newvalue) \
+  ({ __typeof (*mem) result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile ("xchgb %b0, %1"				      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" (newvalue), "m" (*mem));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile ("xchgw %w0, %1"				      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" (newvalue), "m" (*mem));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile ("xchgl %0, %1"					      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" (newvalue), "m" (*mem));			      \
+     else								      \
+       __asm __volatile ("xchgq %q0, %1"				      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" ((long) (newvalue)), "m" (*mem));	      \
+     result; })
+
+
+#define atomic_exchange_and_add(mem, value) \
+  ({ __typeof (*mem) result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "xaddb %b0, %1"			      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "xaddw %w0, %1"			      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "xaddl %0, %1"			      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" (value), "m" (*mem));			      \
+     else								      \
+       __asm __volatile (LOCK_PREFIX "xaddq %q0, %1"			      \
+			 : "=r" (result), "=m" (*mem)			      \
+			 : "0" ((long) (value)), "m" (*mem));		      \
+     result; })
+
+
+#define atomic_add(mem, value) \
+  (void) ({ if (__builtin_constant_p (value) && (value) == 1)		      \
+	      atomic_increment (mem);					      \
+	    else if (__builtin_constant_p (value) && (value) == 1)	      \
+	      atomic_decrement (mem);					      \
+	    else if (sizeof (*mem) == 1)				      \
+	      __asm __volatile (LOCK_PREFIX "addb %b1, %0"		      \
+				: "=m" (*mem)				      \
+				: "ir" (value), "m" (*mem));		      \
+	    else if (sizeof (*mem) == 2)				      \
+	      __asm __volatile (LOCK_PREFIX "addw %w1, %0"		      \
+				: "=m" (*mem)				      \
+				: "ir" (value), "m" (*mem));		      \
+	    else if (sizeof (*mem) == 4)				      \
+	      __asm __volatile (LOCK_PREFIX "addl %1, %0"		      \
+				: "=m" (*mem)				      \
+				: "ir" (value), "m" (*mem));		      \
+	    else							      \
+	      __asm __volatile (LOCK_PREFIX "addq %q1, %0"		      \
+				: "=m" (*mem)				      \
+				: "ir" ((long) (value)), "m" (*mem));	      \
+	    })
+
+
+#define atomic_add_negative(mem, value) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "addb %b2, %0; sets %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "addw %w2, %0; sets %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "addl %2, %0; sets %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else								      \
+       __asm __volatile (LOCK_PREFIX "addq %q2, %0; sets %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" ((long) (value)), "m" (*mem));		      \
+     __result; })
+
+
+#define atomic_add_zero(mem, value) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "addb %b2, %0; setz %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "addw %w2, %0; setz %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "addl %2, %0; setz %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" (value), "m" (*mem));			      \
+     else								      \
+       __asm __volatile (LOCK_PREFIX "addq %q2, %0; setz %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "ir" ((long) (value)), "m" (*mem));		      \
+     __result; })
+
+
+#define atomic_increment(mem) \
+  (void) ({ if (sizeof (*mem) == 1)					      \
+	      __asm __volatile (LOCK_PREFIX "incb %b0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else if (sizeof (*mem) == 2)				      \
+	      __asm __volatile (LOCK_PREFIX "incw %w0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else if (sizeof (*mem) == 4)				      \
+	      __asm __volatile (LOCK_PREFIX "incl %0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else							      \
+	      __asm __volatile (LOCK_PREFIX "incq %q0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    })
+
+
+#define atomic_increment_and_test(mem) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "incb %b0; sete %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "incw %w0; sete %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "incl %0; sete %1"			      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else								      \
+       __asm __volatile (LOCK_PREFIX "incq %q0; sete %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     __result; })
+
+
+#define atomic_decrement(mem) \
+  (void) ({ if (sizeof (*mem) == 1)					      \
+	      __asm __volatile (LOCK_PREFIX "decb %b0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else if (sizeof (*mem) == 2)				      \
+	      __asm __volatile (LOCK_PREFIX "decw %w0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else if (sizeof (*mem) == 4)				      \
+	      __asm __volatile (LOCK_PREFIX "decl %0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    else							      \
+	      __asm __volatile (LOCK_PREFIX "decq %q0"			      \
+				: "=m" (*mem)				      \
+				: "m" (*mem));				      \
+	    })
+
+
+#define atomic_decrement_and_test(mem) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "decb %b0; sete %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "decw %w0; sete %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "decl %0; sete %1"			      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     else								      \
+       __asm __volatile (LOCK_PREFIX "decq %q0; sete %1"		      \
+			 : "=m" (*mem), "=qm" (__result)		      \
+			 : "m" (*mem));					      \
+     __result; })
+
+
+#define atomic_bit_set(mem, bit) \
+  (void) ({ if (sizeof (*mem) == 1)					      \
+	      __asm __volatile (LOCK_PREFIX "orb %b2, %0"		      \
+				: "=m" (*mem)				      \
+				: "m" (*mem), "ir" (1L << (bit)));	      \
+	    else if (sizeof (*mem) == 2)				      \
+	      __asm __volatile (LOCK_PREFIX "orw %w2, %0"		      \
+				: "=m" (*mem)				      \
+				: "m" (*mem), "ir" (1L << (bit)));	      \
+	    else if (sizeof (*mem) == 4)				      \
+	      __asm __volatile (LOCK_PREFIX "orl %2, %0"		      \
+				: "=m" (*mem)				      \
+				: "m" (*mem), "ir" (1L << (bit)));	      \
+	    else if (__builtin_constant_p (bit) && (bit) < 32)		      \
+	      __asm __volatile (LOCK_PREFIX "orq %2, %0"		      \
+				: "=m" (*mem)				      \
+				: "m" (*mem), "i" (1L << (bit)));	      \
+	    else							      \
+	      __asm __volatile (LOCK_PREFIX "orq %q2, %0"		      \
+				: "=m" (*mem)				      \
+				: "m" (*mem), "r" (1UL << (bit)));	      \
+	    })
+
+
+#define atomic_bit_test_set(mem, bit) \
+  ({ unsigned char __result;						      \
+     if (sizeof (*mem) == 1)						      \
+       __asm __volatile (LOCK_PREFIX "btsb %3, %1; setc %0"		      \
+			 : "=q" (__result), "=m" (*mem)			      \
+			 : "m" (*mem), "ir" (bit));			      \
+     else if (sizeof (*mem) == 2)					      \
+       __asm __volatile (LOCK_PREFIX "btsw %3, %1; setc %0"		      \
+			 : "=q" (__result), "=m" (*mem)			      \
+			 : "m" (*mem), "ir" (bit));			      \
+     else if (sizeof (*mem) == 4)					      \
+       __asm __volatile (LOCK_PREFIX "btsl %3, %1; setc %0"		      \
+			 : "=q" (__result), "=m" (*mem)			      \
+			 : "m" (*mem), "ir" (bit));			      \
+     else							      	      \
+       __asm __volatile (LOCK_PREFIX "btsq %3, %1; setc %0"		      \
+			 : "=q" (__result), "=m" (*mem)			      \
+			 : "m" (*mem), "ir" (bit));			      \
+     __result; })
+
+
+#define atomic_delay() asm ("rep; nop")