From 0fad128bbcd6f99f2a4ebbc976a76ef1e3155d1c Mon Sep 17 00:00:00 2001
From: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Date: Tue, 22 Sep 2009 18:44:26 -0400
Subject: [PATCH] update x86 and ppc atomic ops

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
---
 arch_atomic_x86.h   | 82 ++++++++++++++++++++++++++++++++++++++++++++-
 tests/test_atomic.c | 21 +++++++++---
 2 files changed, 97 insertions(+), 6 deletions(-)

diff --git a/arch_atomic_x86.h b/arch_atomic_x86.h
index 6b33d45..879ce70 100644
--- a/arch_atomic_x86.h
+++ b/arch_atomic_x86.h
@@ -20,6 +20,8 @@
  * Boehm-Demers-Weiser conservative garbage collector.
  */
 
+#include <compiler.h>
+
 #ifndef BITS_PER_LONG
 #define BITS_PER_LONG	(__SIZEOF_LONG__ * 8)
 #endif
@@ -35,16 +37,24 @@ struct __atomic_dummy {
 };
 #define __hp(x)	((struct __atomic_dummy *)(x))
 
+#define atomic_set(addr, v)				\
+do {							\
+	ACCESS_ONCE(*(addr)) = (v);			\
+} while (0)
+
+#define atomic_read(addr)	ACCESS_ONCE(*(addr))
+
 /* cmpxchg */
 
 static inline __attribute__((always_inline))
 unsigned long _atomic_cmpxchg(volatile void *addr, unsigned long old,
-		unsigned long _new, int len)
+			      unsigned long _new, int len)
 {
 	switch (len) {
 	case 1:
 	{
 		unsigned char result = old;
+
 		__asm__ __volatile__(
 		"lock; cmpxchgb %2, %1"
 			: "+a"(result), "+m"(*__hp(addr))
@@ -55,6 +65,7 @@ unsigned long _atomic_cmpxchg(volatile void *addr, unsigned long old,
 	case 2:
 	{
 		unsigned short result = old;
+
 		__asm__ __volatile__(
 		"lock; cmpxchgw %2, %1"
 			: "+a"(result), "+m"(*__hp(addr))
@@ -65,6 +76,7 @@ unsigned long _atomic_cmpxchg(volatile void *addr, unsigned long old,
 	case 4:
 	{
 		unsigned int result = old;
+
 		__asm__ __volatile__(
 		"lock; cmpxchgl %2, %1"
 			: "+a"(result), "+m"(*__hp(addr))
@@ -76,6 +88,7 @@ unsigned long _atomic_cmpxchg(volatile void *addr, unsigned long old,
 	case 8:
 	{
 		unsigned long result = old;
+
 		__asm__ __volatile__(
 		"lock; cmpxchgq %2, %1"
 			: "+a"(result), "+m"(*__hp(addr))
@@ -156,6 +169,73 @@ unsigned long _atomic_exchange(volatile void *addr, unsigned long val, int len)
 	((__typeof__(*(addr))) _atomic_exchange((addr), (unsigned long)(v), \
 						sizeof(*(addr))))
 
+/* atomic_add_return, atomic_sub_return */
+
+static inline __attribute__((always_inline))
+unsigned long _atomic_add_return(volatile void *addr, unsigned long val,
+				 int len)
+{
+	switch (len) {
+	case 1:
+	{
+		unsigned char result = val;
+
+		__asm__ __volatile__(
+		"lock; xaddb %1, %0"
+			: "+m"(*__hp(addr)), "+q" (result)
+			:
+			: "memory");
+		return result + (unsigned char)val;
+	}
+	case 2:
+	{
+		unsigned short result = val;
+
+		__asm__ __volatile__(
+		"lock; xaddw %1, %0"
+			: "+m"(*__hp(addr)), "+r" (result)
+			:
+			: "memory");
+		return result + (unsigned short)val;
+	}
+	case 4:
+	{
+		unsigned int result = val;
+
+		__asm__ __volatile__(
+		"lock; xaddl %1, %0"
+			: "+m"(*__hp(addr)), "+r" (result)
+			:
+			: "memory");
+		return result + (unsigned int)val;
+	}
+#if (BITS_PER_LONG == 64)
+	case 8:
+	{
+		unsigned long result = val;
+
+		__asm__ __volatile__(
+		"lock; xaddq %1, %0"
+			: "+m"(*__hp(addr)), "+r" (result)
+			:
+			: "memory");
+		return result + (unsigned long)val;
+	}
+#endif
+	}
+	/* generate an illegal instruction. Cannot catch this with linker tricks
+	 * when optimizations are disabled. */
+	__asm__ __volatile__("ud2");
+	return 0;
+}
+
+#define atomic_add_return(addr, v)					\
+	((__typeof__(*(addr))) _atomic_add_return((addr),		\
+						  (unsigned long)(v),	\
+						  sizeof(*(addr))))
+
+#define atomic_sub_return(addr, v)	atomic_add_return((addr), -(v))
+
 /* atomic_add, atomic_sub */
 
 static inline __attribute__((always_inline))
diff --git a/tests/test_atomic.c b/tests/test_atomic.c
index 3d2b64e..102012f 100644
--- a/tests/test_atomic.c
+++ b/tests/test_atomic.c
@@ -16,18 +16,29 @@ do {						\
 	__typeof__(*(ptr)) v;			\
 						\
 	atomic_add(ptr, 10);			\
-	assert(*(ptr) == 10);			\
+	assert(atomic_read(ptr) == 10);		\
 	atomic_add(ptr, -11UL);			\
-	assert(*(ptr) == (__typeof__(*(ptr)))-1UL);	\
+	assert(atomic_read(ptr) == (__typeof__(*(ptr)))-1UL);	\
 	v = cmpxchg(ptr, -1UL, 22);		\
-	assert(*(ptr) == 22);			\
+	assert(atomic_read(ptr) == 22);		\
 	assert(v == (__typeof__(*(ptr)))-1UL);	\
 	v = cmpxchg(ptr, 33, 44);		\
-	assert(*(ptr) == 22);			\
+	assert(atomic_read(ptr) == 22);		\
 	assert(v == 22);			\
 	v = xchg(ptr, 55);			\
-	assert(*(ptr) == 55);			\
+	assert(atomic_read(ptr) == 55);		\
 	assert(v == 22);			\
+	atomic_set(ptr, 22);			\
+	atomic_inc(ptr);			\
+	assert(atomic_read(ptr) == 23);		\
+	atomic_dec(ptr);			\
+	assert(atomic_read(ptr) == 22);		\
+	v = atomic_add_return(ptr, 100);	\
+	assert(v == 122);			\
+	assert(atomic_read(ptr) == 122);	\
+	v = atomic_sub_return(ptr, 1);		\
+	assert(v == 121);			\
+	assert(atomic_read(ptr) == 121);	\
 } while (0)
 
 int main(int argc, char **argv)
-- 
2.34.1