[urcu.git] / urcu / uatomic_arch_x86.h

#ifndef _URCU_ARCH_UATOMIC_X86_H
#define _URCU_ARCH_UATOMIC_X86_H

/* 
 * Copyright (c) 1991-1994 by Xerox Corporation.  All rights reserved.
 * Copyright (c) 1996-1999 by Silicon Graphics.  All rights reserved.
 * Copyright (c) 1999-2004 Hewlett-Packard Development Company, L.P.
 * Copyright (c) 2009      Mathieu Desnoyers
 *
 * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
 * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
 *
 * Permission is hereby granted to use or copy this program
 * for any purpose,  provided the above notices are retained on all copies.
 * Permission to modify the code and to distribute modified code is granted,
 * provided the above notices are retained, and a notice that the code was
 * modified is included with the above copyright notice.
 *
 * Code inspired from libuatomic_ops-1.2, inherited in part from the
 * Boehm-Demers-Weiser conservative garbage collector.
 */

#include <urcu/compiler.h>
#include <urcu/system.h>

#define UATOMIC_HAS_ATOMIC_BYTE
#define UATOMIC_HAS_ATOMIC_SHORT

#ifdef __cplusplus
extern "C" {
#endif 

/*
 * Derived from AO_compare_and_swap() and AO_test_and_set_full().
 */

struct __uatomic_dummy {
	unsigned long v[10];
};
#define __hp(x)	((struct __uatomic_dummy *)(x))

#define _uatomic_set(addr, v)	CMM_STORE_SHARED(*(addr), (v))

/* cmpxchg */

static inline __attribute__((always_inline))
unsigned long __uatomic_cmpxchg(void *addr, unsigned long old,
			      unsigned long _new, int len)
{
	switch (len) {
	case 1:
	{
		unsigned char result = old;

		__asm__ __volatile__(
		"lock; cmpxchgb %2, %1"
			: "+a"(result), "+m"(*__hp(addr))
			: "q"((unsigned char)_new)
			: "memory");
		return result;
	}
	case 2:
	{
		unsigned short result = old;

		__asm__ __volatile__(
		"lock; cmpxchgw %2, %1"
			: "+a"(result), "+m"(*__hp(addr))
			: "r"((unsigned short)_new)
			: "memory");
		return result;
	}
	case 4:
	{
		unsigned int result = old;

		__asm__ __volatile__(
		"lock; cmpxchgl %2, %1"
			: "+a"(result), "+m"(*__hp(addr))
			: "r"((unsigned int)_new)
			: "memory");
		return result;
	}
#if (CAA_BITS_PER_LONG == 64)
	case 8:
	{
		unsigned long result = old;

		__asm__ __volatile__(
		"lock; cmpxchgq %2, %1"
			: "+a"(result), "+m"(*__hp(addr))
			: "r"((unsigned long)_new)
			: "memory");
		return result;
	}
#endif
	}
	/* generate an illegal instruction. Cannot catch this with linker tricks
	 * when optimizations are disabled. */
	__asm__ __volatile__("ud2");
	return 0;
}

#define _uatomic_cmpxchg(addr, old, _new)				      \
	((__typeof__(*(addr))) __uatomic_cmpxchg((addr), (unsigned long)(old),\
						(unsigned long)(_new), 	      \
						sizeof(*(addr))))

/* xchg */

static inline __attribute__((always_inline))
unsigned long __uatomic_exchange(void *addr, unsigned long val, int len)
{
	/* Note: the "xchg" instruction does not need a "lock" prefix. */
	switch (len) {
	case 1:
	{
		unsigned char result;
		__asm__ __volatile__(
		"xchgb %0, %1"
			: "=q"(result), "+m"(*__hp(addr))
			: "0" ((unsigned char)val)
			: "memory");
		return result;
	}
	case 2:
	{
		unsigned short result;
		__asm__ __volatile__(
		"xchgw %0, %1"
			: "=r"(result), "+m"(*__hp(addr))
			: "0" ((unsigned short)val)
			: "memory");
		return result;
	}
	case 4:
	{
		unsigned int result;
		__asm__ __volatile__(
		"xchgl %0, %1"
			: "=r"(result), "+m"(*__hp(addr))
			: "0" ((unsigned int)val)
			: "memory");
		return result;
	}
#if (CAA_BITS_PER_LONG == 64)
	case 8:
	{
		unsigned long result;
		__asm__ __volatile__(
		"xchgq %0, %1"
			: "=r"(result), "+m"(*__hp(addr))
			: "0" ((unsigned long)val)
			: "memory");
		return result;
	}
#endif
	}
	/* generate an illegal instruction. Cannot catch this with linker tricks
	 * when optimizations are disabled. */
	__asm__ __volatile__("ud2");
	return 0;
}

#define _uatomic_xchg(addr, v)						      \
	((__typeof__(*(addr))) __uatomic_exchange((addr), (unsigned long)(v), \
						sizeof(*(addr))))

/* uatomic_add_return */

static inline __attribute__((always_inline))
unsigned long __uatomic_add_return(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 (CAA_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 _uatomic_add_return(addr, v)					\
	((__typeof__(*(addr))) __uatomic_add_return((addr),		\
						  (unsigned long)(v),	\
						  sizeof(*(addr))))

/* uatomic_or */

static inline __attribute__((always_inline))
void __uatomic_or(void *addr, unsigned long val, int len)
{
	switch (len) {
	case 1:
	{
		__asm__ __volatile__(
		"lock; orb %1, %0"
			: "=m"(*__hp(addr))
			: "iq" ((unsigned char)val)
			: "memory");
		return;
	}
	case 2:
	{
		__asm__ __volatile__(
		"lock; orw %1, %0"
			: "=m"(*__hp(addr))
			: "ir" ((unsigned short)val)
			: "memory");
		return;
	}
	case 4:
	{
		__asm__ __volatile__(
		"lock; orl %1, %0"
			: "=m"(*__hp(addr))
			: "ir" ((unsigned int)val)
			: "memory");
		return;
	}
#if (CAA_BITS_PER_LONG == 64)
	case 8:
	{
		__asm__ __volatile__(
		"lock; orq %1, %0"
			: "=m"(*__hp(addr))
			: "er" ((unsigned long)val)
			: "memory");
		return;
	}
#endif
	}
	/* generate an illegal instruction. Cannot catch this with linker tricks
	 * when optimizations are disabled. */
	__asm__ __volatile__("ud2");
	return;
}

#define _uatomic_or(addr, v)						   \
	(__uatomic_or((addr), (unsigned long)(v), sizeof(*(addr))))

/* uatomic_add */

static inline __attribute__((always_inline))
void __uatomic_add(void *addr, unsigned long val, int len)
{
	switch (len) {
	case 1:
	{
		__asm__ __volatile__(
		"lock; addb %1, %0"
			: "=m"(*__hp(addr))
			: "iq" ((unsigned char)val)
			: "memory");
		return;
	}
	case 2:
	{
		__asm__ __volatile__(
		"lock; addw %1, %0"
			: "=m"(*__hp(addr))
			: "ir" ((unsigned short)val)
			: "memory");
		return;
	}
	case 4:
	{
		__asm__ __volatile__(
		"lock; addl %1, %0"
			: "=m"(*__hp(addr))
			: "ir" ((unsigned int)val)
			: "memory");
		return;
	}
#if (CAA_BITS_PER_LONG == 64)
	case 8:
	{
		__asm__ __volatile__(
		"lock; addq %1, %0"
			: "=m"(*__hp(addr))
			: "er" ((unsigned long)val)
			: "memory");
		return;
	}
#endif
	}
	/* generate an illegal instruction. Cannot catch this with linker tricks
	 * when optimizations are disabled. */
	__asm__ __volatile__("ud2");
	return;
}

#define _uatomic_add(addr, v)						   \
	(__uatomic_add((addr), (unsigned long)(v), sizeof(*(addr))))


/* uatomic_inc */

static inline __attribute__((always_inline))
void __uatomic_inc(void *addr, int len)
{
	switch (len) {
	case 1:
	{
		__asm__ __volatile__(
		"lock; incb %0"
			: "=m"(*__hp(addr))
			:
			: "memory");
		return;
	}
	case 2:
	{
		__asm__ __volatile__(
		"lock; incw %0"
			: "=m"(*__hp(addr))
			:
			: "memory");
		return;
	}
	case 4:
	{
		__asm__ __volatile__(
		"lock; incl %0"
			: "=m"(*__hp(addr))
			:
			: "memory");
		return;
	}
#if (CAA_BITS_PER_LONG == 64)
	case 8:
	{
		__asm__ __volatile__(
		"lock; incq %0"
			: "=m"(*__hp(addr))
			:
			: "memory");
		return;
	}
#endif
	}
	/* generate an illegal instruction. Cannot catch this with linker tricks
	 * when optimizations are disabled. */
	__asm__ __volatile__("ud2");
	return;
}

#define _uatomic_inc(addr)	(__uatomic_inc((addr), sizeof(*(addr))))

/* uatomic_dec */

static inline __attribute__((always_inline))
void __uatomic_dec(void *addr, int len)
{
	switch (len) {
	case 1:
	{
		__asm__ __volatile__(
		"lock; decb %0"
			: "=m"(*__hp(addr))
			:
			: "memory");
		return;
	}
	case 2:
	{
		__asm__ __volatile__(
		"lock; decw %0"
			: "=m"(*__hp(addr))
			:
			: "memory");
		return;
	}
	case 4:
	{
		__asm__ __volatile__(
		"lock; decl %0"
			: "=m"(*__hp(addr))
			:
			: "memory");
		return;
	}
#if (CAA_BITS_PER_LONG == 64)
	case 8:
	{
		__asm__ __volatile__(
		"lock; decq %0"
			: "=m"(*__hp(addr))
			:
			: "memory");
		return;
	}
#endif
	}
	/* generate an illegal instruction. Cannot catch this with linker tricks
	 * when optimizations are disabled. */
	__asm__ __volatile__("ud2");
	return;
}

#define _uatomic_dec(addr)	(__uatomic_dec((addr), sizeof(*(addr))))

#if ((CAA_BITS_PER_LONG != 64) && defined(CONFIG_RCU_COMPAT_ARCH))
extern int __rcu_cas_avail;
extern int __rcu_cas_init(void);

#define UATOMIC_COMPAT(insn)							\
	((likely(__rcu_cas_avail > 0))						\
	? (_uatomic_##insn)							\
		: ((unlikely(__rcu_cas_avail < 0)				\
			? ((__rcu_cas_init() > 0)				\
				? (_uatomic_##insn)				\
				: (compat_uatomic_##insn))			\
			: (compat_uatomic_##insn))))

extern unsigned long _compat_uatomic_set(void *addr,
					 unsigned long _new, int len);
#define compat_uatomic_set(addr, _new)				     	       \
	((__typeof__(*(addr))) _compat_uatomic_set((addr),		       \
						(unsigned long)(_new), 	       \
						sizeof(*(addr))))


extern unsigned long _compat_uatomic_xchg(void *addr,
					  unsigned long _new, int len);
#define compat_uatomic_xchg(addr, _new)					       \
	((__typeof__(*(addr))) _compat_uatomic_xchg((addr),		       \
						(unsigned long)(_new), 	       \
						sizeof(*(addr))))

extern unsigned long _compat_uatomic_cmpxchg(void *addr, unsigned long old,
					     unsigned long _new, int len);
#define compat_uatomic_cmpxchg(addr, old, _new)				       \
	((__typeof__(*(addr))) _compat_uatomic_cmpxchg((addr),		       \
						(unsigned long)(old),	       \
						(unsigned long)(_new), 	       \
						sizeof(*(addr))))

extern unsigned long _compat_uatomic_or(void *addr,
				        unsigned long _new, int len);
#define compat_uatomic_or(addr, v)				       \
	((__typeof__(*(addr))) _compat_uatomic_or((addr),	       \
						  (unsigned long)(v),  \
						  sizeof(*(addr))))

extern unsigned long _compat_uatomic_add_return(void *addr,
						unsigned long _new, int len);
#define compat_uatomic_add_return(addr, v)				       \
	((__typeof__(*(addr))) _compat_uatomic_add_return((addr),	       \
						(unsigned long)(v), 	       \
						sizeof(*(addr))))

#define compat_uatomic_add(addr, v)					       \
		((void)compat_uatomic_add_return((addr), (v)))
#define compat_uatomic_inc(addr)					       \
		(compat_uatomic_add((addr), 1))
#define compat_uatomic_dec(addr)					       \
		(compat_uatomic_add((addr), -1))

#else
#define UATOMIC_COMPAT(insn)	(_uatomic_##insn)
#endif

/* Read is atomic even in compat mode */
#define uatomic_set(addr, v)			\
		UATOMIC_COMPAT(set(addr, v))

#define uatomic_cmpxchg(addr, old, _new)	\
		UATOMIC_COMPAT(cmpxchg(addr, old, _new))
#define uatomic_xchg(addr, v)			\
		UATOMIC_COMPAT(xchg(addr, v))
#define uatomic_or(addr, v)		\
		UATOMIC_COMPAT(or(addr, v))
#define uatomic_add_return(addr, v)		\
		UATOMIC_COMPAT(add_return(addr, v))

#define uatomic_add(addr, v)	UATOMIC_COMPAT(add(addr, v))
#define uatomic_inc(addr)	UATOMIC_COMPAT(inc(addr))
#define uatomic_dec(addr)	UATOMIC_COMPAT(dec(addr))

#ifdef __cplusplus 
}
#endif

#include <urcu/uatomic_generic.h>

#endif /* _URCU_ARCH_UATOMIC_X86_H */
Commit	Line	Data
ec4e58a3 MD	1	#ifndef _URCU_ARCH_UATOMIC_X86_H
ec4e58a3 MD	2	#define _URCU_ARCH_UATOMIC_X86_H
0114ba7f MD	3
	4	/*
	5	* Copyright (c) 1991-1994 by Xerox Corporation. All rights reserved.
	6	* Copyright (c) 1996-1999 by Silicon Graphics. All rights reserved.
	7	* Copyright (c) 1999-2004 Hewlett-Packard Development Company, L.P.
	8	* Copyright (c) 2009 Mathieu Desnoyers
	9	*
	10	* THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
	11	* OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
	12	*
	13	* Permission is hereby granted to use or copy this program
	14	* for any purpose, provided the above notices are retained on all copies.
	15	* Permission to modify the code and to distribute modified code is granted,
	16	* provided the above notices are retained, and a notice that the code was
	17	* modified is included with the above copyright notice.
	18	*
ec4e58a3	19	* Code inspired from libuatomic_ops-1.2, inherited in part from the
0114ba7f MD	20	* Boehm-Demers-Weiser conservative garbage collector.
	21	*/
	22
ec4e58a3	23	#include <urcu/compiler.h>
bf9de1b7	24	#include <urcu/system.h>
0fad128b	25
f469d839 PB	26	#define UATOMIC_HAS_ATOMIC_BYTE
	27	#define UATOMIC_HAS_ATOMIC_SHORT
	28
36bc70a8 MD	29	#ifdef __cplusplus
	30	extern "C" {
	31	#endif
	32
0114ba7f	33	/*
0114ba7f MD	34	* Derived from AO_compare_and_swap() and AO_test_and_set_full().
	35	*/
	36
ec4e58a3	37	struct __uatomic_dummy {
cc1be41b MD	38	unsigned long v[10];
cc1be41b MD	39	};
ec4e58a3	40	#define __hp(x) ((struct __uatomic_dummy *)(x))
cc1be41b	41
6cf3827c	42	#define _uatomic_set(addr, v) CMM_STORE_SHARED(*(addr), (v))
0fad128b	43
cc1be41b MD	44	/* cmpxchg */
cc1be41b MD	45
5dba80f9	46	static inline __attribute__((always_inline))
bf9de1b7	47	unsigned long __uatomic_cmpxchg(void *addr, unsigned long old,
0fad128b	48	unsigned long _new, int len)
0114ba7f	49	{
cc1be41b MD	50	switch (len) {
	51	case 1:
	52	{
	53	unsigned char result = old;
0fad128b	54
cc1be41b MD	55	__asm__ __volatile__(
	56	"lock; cmpxchgb %2, %1"
	57	: "+a"(result), "+m"(*__hp(addr))
	58	: "q"((unsigned char)_new)
0114ba7f	59	: "memory");
cc1be41b MD	60	return result;
	61	}
	62	case 2:
	63	{
	64	unsigned short result = old;
0fad128b	65
cc1be41b MD	66	__asm__ __volatile__(
	67	"lock; cmpxchgw %2, %1"
	68	: "+a"(result), "+m"(*__hp(addr))
	69	: "r"((unsigned short)_new)
	70	: "memory");
	71	return result;
	72	}
	73	case 4:
	74	{
	75	unsigned int result = old;
0fad128b	76
cc1be41b MD	77	__asm__ __volatile__(
	78	"lock; cmpxchgl %2, %1"
	79	: "+a"(result), "+m"(*__hp(addr))
	80	: "r"((unsigned int)_new)
	81	: "memory");
	82	return result;
	83	}
e040d717	84	#if (CAA_BITS_PER_LONG == 64)
cc1be41b MD	85	case 8:
cc1be41b MD	86	{
6edb297e	87	unsigned long result = old;
0fad128b	88
cc1be41b	89	__asm__ __volatile__(
2c5e5fb3	90	"lock; cmpxchgq %2, %1"
cc1be41b MD	91	: "+a"(result), "+m"(*__hp(addr))
	92	: "r"((unsigned long)_new)
	93	: "memory");
	94	return result;
	95	}
	96	#endif
	97	}
	98	/* generate an illegal instruction. Cannot catch this with linker tricks
	99	* when optimizations are disabled. */
	100	__asm__ __volatile__("ud2");
	101	return 0;
0114ba7f MD	102	}
0114ba7f MD	103
bf9de1b7 MD	104	#define _uatomic_cmpxchg(addr, old, _new) \
	105	((__typeof__(*(addr))) __uatomic_cmpxchg((addr), (unsigned long)(old),\
	106	(unsigned long)(_new), \
cc1be41b MD	107	sizeof(*(addr))))
	108
	109	/* xchg */
0114ba7f	110
5dba80f9	111	static inline __attribute__((always_inline))
bf9de1b7	112	unsigned long __uatomic_exchange(void *addr, unsigned long val, int len)
0114ba7f	113	{
cc1be41b MD	114	/* Note: the "xchg" instruction does not need a "lock" prefix. */
	115	switch (len) {
	116	case 1:
	117	{
	118	unsigned char result;
	119	__asm__ __volatile__(
	120	"xchgb %0, %1"
	121	: "=q"(result), "+m"(*__hp(addr))
	122	: "0" ((unsigned char)val)
	123	: "memory");
	124	return result;
	125	}
	126	case 2:
	127	{
	128	unsigned short result;
	129	__asm__ __volatile__(
	130	"xchgw %0, %1"
	131	: "=r"(result), "+m"(*__hp(addr))
	132	: "0" ((unsigned short)val)
	133	: "memory");
	134	return result;
	135	}
	136	case 4:
	137	{
	138	unsigned int result;
	139	__asm__ __volatile__(
	140	"xchgl %0, %1"
	141	: "=r"(result), "+m"(*__hp(addr))
	142	: "0" ((unsigned int)val)
	143	: "memory");
	144	return result;
	145	}
e040d717	146	#if (CAA_BITS_PER_LONG == 64)
cc1be41b MD	147	case 8:
	148	{
	149	unsigned long result;
	150	__asm__ __volatile__(
0114ba7f	151	"xchgq %0, %1"
cc1be41b MD	152	: "=r"(result), "+m"(*__hp(addr))
cc1be41b MD	153	: "0" ((unsigned long)val)
0114ba7f	154	: "memory");
cc1be41b MD	155	return result;
	156	}
	157	#endif
	158	}
	159	/* generate an illegal instruction. Cannot catch this with linker tricks
	160	* when optimizations are disabled. */
	161	__asm__ __volatile__("ud2");
	162	return 0;
0114ba7f MD	163	}
0114ba7f MD	164
bf9de1b7 MD	165	#define _uatomic_xchg(addr, v) \
bf9de1b7 MD	166	((__typeof__(*(addr))) __uatomic_exchange((addr), (unsigned long)(v), \
cc1be41b MD	167	sizeof(*(addr))))
cc1be41b MD	168
8760d94e	169	/* uatomic_add_return */
0fad128b MD	170
0fad128b MD	171	static inline __attribute__((always_inline))
bf9de1b7	172	unsigned long __uatomic_add_return(void *addr, unsigned long val,
0fad128b MD	173	int len)
	174	{
	175	switch (len) {
	176	case 1:
	177	{
	178	unsigned char result = val;
	179
	180	__asm__ __volatile__(
	181	"lock; xaddb %1, %0"
	182	: "+m"(*__hp(addr)), "+q" (result)
	183	:
	184	: "memory");
	185	return result + (unsigned char)val;
	186	}
	187	case 2:
	188	{
	189	unsigned short result = val;
	190
	191	__asm__ __volatile__(
	192	"lock; xaddw %1, %0"
	193	: "+m"(*__hp(addr)), "+r" (result)
	194	:
	195	: "memory");
	196	return result + (unsigned short)val;
	197	}
	198	case 4:
	199	{
	200	unsigned int result = val;
	201
	202	__asm__ __volatile__(
	203	"lock; xaddl %1, %0"
	204	: "+m"(*__hp(addr)), "+r" (result)
	205	:
	206	: "memory");
	207	return result + (unsigned int)val;
	208	}
e040d717	209	#if (CAA_BITS_PER_LONG == 64)
0fad128b MD	210	case 8:
	211	{
	212	unsigned long result = val;
	213
	214	__asm__ __volatile__(
	215	"lock; xaddq %1, %0"
	216	: "+m"(*__hp(addr)), "+r" (result)
	217	:
	218	: "memory");
	219	return result + (unsigned long)val;
	220	}
	221	#endif
	222	}
	223	/* generate an illegal instruction. Cannot catch this with linker tricks
	224	* when optimizations are disabled. */
	225	__asm__ __volatile__("ud2");
	226	return 0;
	227	}
	228
bf9de1b7 MD	229	#define _uatomic_add_return(addr, v) \
bf9de1b7 MD	230	((__typeof__(*(addr))) __uatomic_add_return((addr), \
0fad128b MD	231	(unsigned long)(v), \
	232	sizeof(*(addr))))
	233
985b35b1 PB	234	/* uatomic_or */
	235
	236	static inline __attribute__((always_inline))
	237	void __uatomic_or(void *addr, unsigned long val, int len)
	238	{
	239	switch (len) {
	240	case 1:
	241	{
	242	__asm__ __volatile__(
	243	"lock; orb %1, %0"
	244	: "=m"(*__hp(addr))
	245	: "iq" ((unsigned char)val)
	246	: "memory");
	247	return;
	248	}
	249	case 2:
	250	{
	251	__asm__ __volatile__(
	252	"lock; orw %1, %0"
	253	: "=m"(*__hp(addr))
	254	: "ir" ((unsigned short)val)
	255	: "memory");
	256	return;
	257	}
	258	case 4:
	259	{
	260	__asm__ __volatile__(
	261	"lock; orl %1, %0"
	262	: "=m"(*__hp(addr))
	263	: "ir" ((unsigned int)val)
	264	: "memory");
	265	return;
	266	}
	267	#if (CAA_BITS_PER_LONG == 64)
	268	case 8:
	269	{
	270	__asm__ __volatile__(
	271	"lock; orq %1, %0"
	272	: "=m"(*__hp(addr))
	273	: "er" ((unsigned long)val)
	274	: "memory");
	275	return;
	276	}
	277	#endif
	278	}
	279	/* generate an illegal instruction. Cannot catch this with linker tricks
	280	* when optimizations are disabled. */
	281	__asm__ __volatile__("ud2");
	282	return;
	283	}
	284
	285	#define _uatomic_or(addr, v) \
	286	(__uatomic_or((addr), (unsigned long)(v), sizeof(*(addr))))
	287
8760d94e	288	/* uatomic_add */
0114ba7f	289
5dba80f9	290	static inline __attribute__((always_inline))
bf9de1b7	291	void __uatomic_add(void *addr, unsigned long val, int len)
0114ba7f MD	292	{
0114ba7f MD	293	switch (len) {
cc1be41b MD	294	case 1:
	295	{
	296	__asm__ __volatile__(
	297	"lock; addb %1, %0"
	298	: "=m"(*__hp(addr))
87322fe8 MD	299	: "iq" ((unsigned char)val)
87322fe8 MD	300	: "memory");
cc1be41b MD	301	return;
	302	}
	303	case 2:
	304	{
	305	__asm__ __volatile__(
	306	"lock; addw %1, %0"
	307	: "=m"(*__hp(addr))
87322fe8 MD	308	: "ir" ((unsigned short)val)
87322fe8 MD	309	: "memory");
cc1be41b MD	310	return;
	311	}
	312	case 4:
	313	{
	314	__asm__ __volatile__(
	315	"lock; addl %1, %0"
	316	: "=m"(*__hp(addr))
87322fe8 MD	317	: "ir" ((unsigned int)val)
87322fe8 MD	318	: "memory");
cc1be41b MD	319	return;
cc1be41b MD	320	}
e040d717	321	#if (CAA_BITS_PER_LONG == 64)
cc1be41b MD	322	case 8:
	323	{
	324	__asm__ __volatile__(
	325	"lock; addq %1, %0"
	326	: "=m"(*__hp(addr))
87322fe8 MD	327	: "er" ((unsigned long)val)
87322fe8 MD	328	: "memory");
cc1be41b MD	329	return;
cc1be41b MD	330	}
0114ba7f MD	331	#endif
	332	}
	333	/* generate an illegal instruction. Cannot catch this with linker tricks
	334	* when optimizations are disabled. */
	335	__asm__ __volatile__("ud2");
a81b8e5e	336	return;
0114ba7f MD	337	}
0114ba7f MD	338
bf9de1b7 MD	339	#define _uatomic_add(addr, v) \
bf9de1b7 MD	340	(__uatomic_add((addr), (unsigned long)(v), sizeof(*(addr))))
0114ba7f	341
2c5e5fb3	342
ec4e58a3	343	/* uatomic_inc */
2c5e5fb3 MD	344
2c5e5fb3 MD	345	static inline __attribute__((always_inline))
bf9de1b7	346	void __uatomic_inc(void *addr, int len)
2c5e5fb3 MD	347	{
	348	switch (len) {
	349	case 1:
	350	{
	351	__asm__ __volatile__(
	352	"lock; incb %0"
	353	: "=m"(*__hp(addr))
	354	:
	355	: "memory");
	356	return;
	357	}
	358	case 2:
	359	{
	360	__asm__ __volatile__(
	361	"lock; incw %0"
	362	: "=m"(*__hp(addr))
	363	:
	364	: "memory");
	365	return;
	366	}
	367	case 4:
	368	{
	369	__asm__ __volatile__(
	370	"lock; incl %0"
	371	: "=m"(*__hp(addr))
	372	:
	373	: "memory");
	374	return;
	375	}
e040d717	376	#if (CAA_BITS_PER_LONG == 64)
2c5e5fb3 MD	377	case 8:
	378	{
	379	__asm__ __volatile__(
	380	"lock; incq %0"
	381	: "=m"(*__hp(addr))
	382	:
	383	: "memory");
	384	return;
	385	}
	386	#endif
	387	}
	388	/* generate an illegal instruction. Cannot catch this with linker tricks
	389	* when optimizations are disabled. */
	390	__asm__ __volatile__("ud2");
	391	return;
	392	}
	393
bf9de1b7	394	#define _uatomic_inc(addr) (__uatomic_inc((addr), sizeof(*(addr))))
2c5e5fb3	395
ec4e58a3	396	/* uatomic_dec */
2c5e5fb3 MD	397
2c5e5fb3 MD	398	static inline __attribute__((always_inline))
bf9de1b7	399	void __uatomic_dec(void *addr, int len)
2c5e5fb3 MD	400	{
	401	switch (len) {
	402	case 1:
	403	{
	404	__asm__ __volatile__(
	405	"lock; decb %0"
	406	: "=m"(*__hp(addr))
	407	:
	408	: "memory");
	409	return;
	410	}
	411	case 2:
	412	{
	413	__asm__ __volatile__(
	414	"lock; decw %0"
	415	: "=m"(*__hp(addr))
	416	:
	417	: "memory");
	418	return;
	419	}
	420	case 4:
	421	{
	422	__asm__ __volatile__(
	423	"lock; decl %0"
	424	: "=m"(*__hp(addr))
	425	:
	426	: "memory");
	427	return;
	428	}
e040d717	429	#if (CAA_BITS_PER_LONG == 64)
2c5e5fb3 MD	430	case 8:
	431	{
	432	__asm__ __volatile__(
	433	"lock; decq %0"
	434	: "=m"(*__hp(addr))
	435	:
	436	: "memory");
	437	return;
	438	}
	439	#endif
	440	}
	441	/* generate an illegal instruction. Cannot catch this with linker tricks
	442	* when optimizations are disabled. */
	443	__asm__ __volatile__("ud2");
	444	return;
	445	}
	446
bf9de1b7	447	#define _uatomic_dec(addr) (__uatomic_dec((addr), sizeof(*(addr))))
0114ba7f	448
e040d717	449	#if ((CAA_BITS_PER_LONG != 64) && defined(CONFIG_RCU_COMPAT_ARCH))
02be5561 MD	450	extern int __rcu_cas_avail;
02be5561 MD	451	extern int __rcu_cas_init(void);
bf9de1b7 MD	452
bf9de1b7 MD	453	#define UATOMIC_COMPAT(insn) \
02be5561	454	((likely(__rcu_cas_avail > 0)) \
bf9de1b7	455	? (_uatomic_##insn) \
02be5561 MD	456	: ((unlikely(__rcu_cas_avail < 0) \
02be5561 MD	457	? ((__rcu_cas_init() > 0) \
bf9de1b7 MD	458	? (_uatomic_##insn) \
	459	: (compat_uatomic_##insn)) \
	460	: (compat_uatomic_##insn))))
	461
	462	extern unsigned long _compat_uatomic_set(void *addr,
	463	unsigned long _new, int len);
	464	#define compat_uatomic_set(addr, _new) \
	465	((__typeof__(*(addr))) _compat_uatomic_set((addr), \
	466	(unsigned long)(_new), \
	467	sizeof(*(addr))))
	468
	469
	470	extern unsigned long _compat_uatomic_xchg(void *addr,
	471	unsigned long _new, int len);
	472	#define compat_uatomic_xchg(addr, _new) \
	473	((__typeof__(*(addr))) _compat_uatomic_xchg((addr), \
	474	(unsigned long)(_new), \
	475	sizeof(*(addr))))
7d413817 MD	476
7d413817 MD	477	extern unsigned long _compat_uatomic_cmpxchg(void *addr, unsigned long old,
bf9de1b7 MD	478	unsigned long _new, int len);
	479	#define compat_uatomic_cmpxchg(addr, old, _new) \
	480	((__typeof__(*(addr))) _compat_uatomic_cmpxchg((addr), \
	481	(unsigned long)(old), \
	482	(unsigned long)(_new), \
	483	sizeof(*(addr))))
7d413817	484
985b35b1 PB	485	extern unsigned long _compat_uatomic_or(void *addr,
	486	unsigned long _new, int len);
	487	#define compat_uatomic_or(addr, v) \
	488	((__typeof__(*(addr))) _compat_uatomic_or((addr), \
	489	(unsigned long)(v), \
	490	sizeof(*(addr))))
	491
28ca843d PB	492	extern unsigned long _compat_uatomic_add_return(void *addr,
28ca843d PB	493	unsigned long _new, int len);
bf9de1b7 MD	494	#define compat_uatomic_add_return(addr, v) \
	495	((__typeof__(*(addr))) _compat_uatomic_add_return((addr), \
	496	(unsigned long)(v), \
7d413817	497	sizeof(*(addr))))
bf9de1b7	498
bf9de1b7 MD	499	#define compat_uatomic_add(addr, v) \
bf9de1b7 MD	500	((void)compat_uatomic_add_return((addr), (v)))
bf9de1b7 MD	501	#define compat_uatomic_inc(addr) \
	502	(compat_uatomic_add((addr), 1))
	503	#define compat_uatomic_dec(addr) \
8760d94e	504	(compat_uatomic_add((addr), -1))
bf9de1b7 MD	505
	506	#else
	507	#define UATOMIC_COMPAT(insn) (_uatomic_##insn)
7d413817 MD	508	#endif
7d413817 MD	509
bf9de1b7	510	/* Read is atomic even in compat mode */
bf9de1b7 MD	511	#define uatomic_set(addr, v) \
bf9de1b7 MD	512	UATOMIC_COMPAT(set(addr, v))
8760d94e	513
bf9de1b7 MD	514	#define uatomic_cmpxchg(addr, old, _new) \
	515	UATOMIC_COMPAT(cmpxchg(addr, old, _new))
	516	#define uatomic_xchg(addr, v) \
	517	UATOMIC_COMPAT(xchg(addr, v))
985b35b1 PB	518	#define uatomic_or(addr, v) \
985b35b1 PB	519	UATOMIC_COMPAT(or(addr, v))
bf9de1b7 MD	520	#define uatomic_add_return(addr, v) \
bf9de1b7 MD	521	UATOMIC_COMPAT(add_return(addr, v))
8760d94e	522
bf9de1b7	523	#define uatomic_add(addr, v) UATOMIC_COMPAT(add(addr, v))
bf9de1b7 MD	524	#define uatomic_inc(addr) UATOMIC_COMPAT(inc(addr))
	525	#define uatomic_dec(addr) UATOMIC_COMPAT(dec(addr))
	526
36bc70a8 MD	527	#ifdef __cplusplus
	528	}
	529	#endif
	530
8760d94e PB	531	#include <urcu/uatomic_generic.h>
8760d94e PB	532
ec4e58a3	533	#endif /* _URCU_ARCH_UATOMIC_X86_H */