[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>

#ifdef __cplusplus
extern "C" {
#endif 

#ifndef __SIZEOF_LONG__
#if defined(__x86_64__) || defined(__amd64__)
#define __SIZEOF_LONG__ 8
#else
#define __SIZEOF_LONG__ 4
#endif
#endif

#ifndef BITS_PER_LONG
#define BITS_PER_LONG	(__SIZEOF_LONG__ * 8)
#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)	STORE_SHARED(*(addr), (v))
#define _uatomic_read(addr)	LOAD_SHARED(*(addr))

/* 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 (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 (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, uatomic_sub_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 (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))))

#define _uatomic_sub_return(addr, v)	_uatomic_add_return((addr), -(v))

/* uatomic_add, uatomic_sub */

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 (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))))

#define _uatomic_sub(addr, v)	_uatomic_add((addr), -(v))


/* 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 (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 (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 ((BITS_PER_LONG != 64) && defined(CONFIG_URCU_COMPAT_ARCH))
extern int __urcu_cas_avail;
extern int __urcu_cas_init(void);

#define UATOMIC_COMPAT(insn)							\
	((likely(__urcu_cas_avail > 0))						\
	? (_uatomic_##insn)							\
		: ((unlikely(__urcu_cas_avail < 0)				\
			? ((__urcu_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_xchg(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_sub_return(addr, v)				       \
		compat_uatomic_add_return((addr), -(v))
#define compat_uatomic_add(addr, v)					       \
		((void)compat_uatomic_add_return((addr), (v)))
#define compat_uatomic_sub(addr, v)					       \
		((void)compat_uatomic_sub_return((addr), (v)))
#define compat_uatomic_inc(addr)					       \
		(compat_uatomic_add((addr), 1))
#define compat_uatomic_dec(addr)					       \
		(compat_uatomic_sub((addr), 1))

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

/* Read is atomic even in compat mode */
#define uatomic_read(addr)	_uatomic_read(addr)

#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_add_return(addr, v)		\
		UATOMIC_COMPAT(add_return(addr, v))
#define uatomic_sub_return(addr, v)		\
		UATOMIC_COMPAT(sub_return(addr, v))
#define uatomic_add(addr, v)	UATOMIC_COMPAT(add(addr, v))
#define uatomic_sub(addr, v)	UATOMIC_COMPAT(sub(addr, v))
#define uatomic_inc(addr)	UATOMIC_COMPAT(inc(addr))
#define uatomic_dec(addr)	UATOMIC_COMPAT(dec(addr))

#ifdef __cplusplus 
}
#endif

#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
36bc70a8 MD	26	#ifdef __cplusplus
	27	extern "C" {
	28	#endif
	29
618c4ae1 MD	30	#ifndef __SIZEOF_LONG__
	31	#if defined(__x86_64__) \|\| defined(__amd64__)
	32	#define __SIZEOF_LONG__ 8
	33	#else
	34	#define __SIZEOF_LONG__ 4
	35	#endif
	36	#endif
	37
0114ba7f MD	38	#ifndef BITS_PER_LONG
	39	#define BITS_PER_LONG (__SIZEOF_LONG__ * 8)
	40	#endif
	41
0114ba7f	42	/*
0114ba7f MD	43	* Derived from AO_compare_and_swap() and AO_test_and_set_full().
	44	*/
	45
ec4e58a3	46	struct __uatomic_dummy {
cc1be41b MD	47	unsigned long v[10];
cc1be41b MD	48	};
ec4e58a3	49	#define __hp(x) ((struct __uatomic_dummy *)(x))
cc1be41b	50
bf9de1b7 MD	51	#define _uatomic_set(addr, v) STORE_SHARED(*(addr), (v))
bf9de1b7 MD	52	#define _uatomic_read(addr) LOAD_SHARED(*(addr))
0fad128b	53
cc1be41b MD	54	/* cmpxchg */
cc1be41b MD	55
5dba80f9	56	static inline __attribute__((always_inline))
bf9de1b7	57	unsigned long __uatomic_cmpxchg(void *addr, unsigned long old,
0fad128b	58	unsigned long _new, int len)
0114ba7f	59	{
cc1be41b MD	60	switch (len) {
	61	case 1:
	62	{
	63	unsigned char result = old;
0fad128b	64
cc1be41b MD	65	__asm__ __volatile__(
	66	"lock; cmpxchgb %2, %1"
	67	: "+a"(result), "+m"(*__hp(addr))
	68	: "q"((unsigned char)_new)
0114ba7f	69	: "memory");
cc1be41b MD	70	return result;
	71	}
	72	case 2:
	73	{
	74	unsigned short result = old;
0fad128b	75
cc1be41b MD	76	__asm__ __volatile__(
	77	"lock; cmpxchgw %2, %1"
	78	: "+a"(result), "+m"(*__hp(addr))
	79	: "r"((unsigned short)_new)
	80	: "memory");
	81	return result;
	82	}
	83	case 4:
	84	{
	85	unsigned int result = old;
0fad128b	86
cc1be41b MD	87	__asm__ __volatile__(
	88	"lock; cmpxchgl %2, %1"
	89	: "+a"(result), "+m"(*__hp(addr))
	90	: "r"((unsigned int)_new)
	91	: "memory");
	92	return result;
	93	}
	94	#if (BITS_PER_LONG == 64)
	95	case 8:
	96	{
6edb297e	97	unsigned long result = old;
0fad128b	98
cc1be41b	99	__asm__ __volatile__(
2c5e5fb3	100	"lock; cmpxchgq %2, %1"
cc1be41b MD	101	: "+a"(result), "+m"(*__hp(addr))
	102	: "r"((unsigned long)_new)
	103	: "memory");
	104	return result;
	105	}
	106	#endif
	107	}
	108	/* generate an illegal instruction. Cannot catch this with linker tricks
	109	* when optimizations are disabled. */
	110	__asm__ __volatile__("ud2");
	111	return 0;
0114ba7f MD	112	}
0114ba7f MD	113
bf9de1b7 MD	114	#define _uatomic_cmpxchg(addr, old, _new) \
	115	((__typeof__(*(addr))) __uatomic_cmpxchg((addr), (unsigned long)(old),\
	116	(unsigned long)(_new), \
cc1be41b MD	117	sizeof(*(addr))))
	118
	119	/* xchg */
0114ba7f	120
5dba80f9	121	static inline __attribute__((always_inline))
bf9de1b7	122	unsigned long __uatomic_exchange(void *addr, unsigned long val, int len)
0114ba7f	123	{
cc1be41b MD	124	/* Note: the "xchg" instruction does not need a "lock" prefix. */
	125	switch (len) {
	126	case 1:
	127	{
	128	unsigned char result;
	129	__asm__ __volatile__(
	130	"xchgb %0, %1"
	131	: "=q"(result), "+m"(*__hp(addr))
	132	: "0" ((unsigned char)val)
	133	: "memory");
	134	return result;
	135	}
	136	case 2:
	137	{
	138	unsigned short result;
	139	__asm__ __volatile__(
	140	"xchgw %0, %1"
	141	: "=r"(result), "+m"(*__hp(addr))
	142	: "0" ((unsigned short)val)
	143	: "memory");
	144	return result;
	145	}
	146	case 4:
	147	{
	148	unsigned int result;
	149	__asm__ __volatile__(
	150	"xchgl %0, %1"
	151	: "=r"(result), "+m"(*__hp(addr))
	152	: "0" ((unsigned int)val)
	153	: "memory");
	154	return result;
	155	}
	156	#if (BITS_PER_LONG == 64)
	157	case 8:
	158	{
	159	unsigned long result;
	160	__asm__ __volatile__(
0114ba7f	161	"xchgq %0, %1"
cc1be41b MD	162	: "=r"(result), "+m"(*__hp(addr))
cc1be41b MD	163	: "0" ((unsigned long)val)
0114ba7f	164	: "memory");
cc1be41b MD	165	return result;
	166	}
	167	#endif
	168	}
	169	/* generate an illegal instruction. Cannot catch this with linker tricks
	170	* when optimizations are disabled. */
	171	__asm__ __volatile__("ud2");
	172	return 0;
0114ba7f MD	173	}
0114ba7f MD	174
bf9de1b7 MD	175	#define _uatomic_xchg(addr, v) \
bf9de1b7 MD	176	((__typeof__(*(addr))) __uatomic_exchange((addr), (unsigned long)(v), \
cc1be41b MD	177	sizeof(*(addr))))
cc1be41b MD	178
ec4e58a3	179	/* uatomic_add_return, uatomic_sub_return */
0fad128b MD	180
0fad128b MD	181	static inline __attribute__((always_inline))
bf9de1b7	182	unsigned long __uatomic_add_return(void *addr, unsigned long val,
0fad128b MD	183	int len)
	184	{
	185	switch (len) {
	186	case 1:
	187	{
	188	unsigned char result = val;
	189
	190	__asm__ __volatile__(
	191	"lock; xaddb %1, %0"
	192	: "+m"(*__hp(addr)), "+q" (result)
	193	:
	194	: "memory");
	195	return result + (unsigned char)val;
	196	}
	197	case 2:
	198	{
	199	unsigned short result = val;
	200
	201	__asm__ __volatile__(
	202	"lock; xaddw %1, %0"
	203	: "+m"(*__hp(addr)), "+r" (result)
	204	:
	205	: "memory");
	206	return result + (unsigned short)val;
	207	}
	208	case 4:
	209	{
	210	unsigned int result = val;
	211
	212	__asm__ __volatile__(
	213	"lock; xaddl %1, %0"
	214	: "+m"(*__hp(addr)), "+r" (result)
	215	:
	216	: "memory");
	217	return result + (unsigned int)val;
	218	}
	219	#if (BITS_PER_LONG == 64)
	220	case 8:
	221	{
	222	unsigned long result = val;
	223
	224	__asm__ __volatile__(
	225	"lock; xaddq %1, %0"
	226	: "+m"(*__hp(addr)), "+r" (result)
	227	:
	228	: "memory");
	229	return result + (unsigned long)val;
	230	}
	231	#endif
	232	}
	233	/* generate an illegal instruction. Cannot catch this with linker tricks
	234	* when optimizations are disabled. */
	235	__asm__ __volatile__("ud2");
	236	return 0;
	237	}
	238
bf9de1b7 MD	239	#define _uatomic_add_return(addr, v) \
bf9de1b7 MD	240	((__typeof__(*(addr))) __uatomic_add_return((addr), \
0fad128b MD	241	(unsigned long)(v), \
	242	sizeof(*(addr))))
	243
bf9de1b7	244	#define _uatomic_sub_return(addr, v) _uatomic_add_return((addr), -(v))
0fad128b	245
ec4e58a3	246	/* uatomic_add, uatomic_sub */
0114ba7f	247
5dba80f9	248	static inline __attribute__((always_inline))
bf9de1b7	249	void __uatomic_add(void *addr, unsigned long val, int len)
0114ba7f MD	250	{
0114ba7f MD	251	switch (len) {
cc1be41b MD	252	case 1:
	253	{
	254	__asm__ __volatile__(
	255	"lock; addb %1, %0"
	256	: "=m"(*__hp(addr))
87322fe8 MD	257	: "iq" ((unsigned char)val)
87322fe8 MD	258	: "memory");
cc1be41b MD	259	return;
	260	}
	261	case 2:
	262	{
	263	__asm__ __volatile__(
	264	"lock; addw %1, %0"
	265	: "=m"(*__hp(addr))
87322fe8 MD	266	: "ir" ((unsigned short)val)
87322fe8 MD	267	: "memory");
cc1be41b MD	268	return;
	269	}
	270	case 4:
	271	{
	272	__asm__ __volatile__(
	273	"lock; addl %1, %0"
	274	: "=m"(*__hp(addr))
87322fe8 MD	275	: "ir" ((unsigned int)val)
87322fe8 MD	276	: "memory");
cc1be41b MD	277	return;
cc1be41b MD	278	}
0114ba7f	279	#if (BITS_PER_LONG == 64)
cc1be41b MD	280	case 8:
	281	{
	282	__asm__ __volatile__(
	283	"lock; addq %1, %0"
	284	: "=m"(*__hp(addr))
87322fe8 MD	285	: "er" ((unsigned long)val)
87322fe8 MD	286	: "memory");
cc1be41b MD	287	return;
cc1be41b MD	288	}
0114ba7f MD	289	#endif
	290	}
	291	/* generate an illegal instruction. Cannot catch this with linker tricks
	292	* when optimizations are disabled. */
	293	__asm__ __volatile__("ud2");
a81b8e5e	294	return;
0114ba7f MD	295	}
0114ba7f MD	296
bf9de1b7 MD	297	#define _uatomic_add(addr, v) \
bf9de1b7 MD	298	(__uatomic_add((addr), (unsigned long)(v), sizeof(*(addr))))
0114ba7f	299
bf9de1b7	300	#define _uatomic_sub(addr, v) _uatomic_add((addr), -(v))
2c5e5fb3 MD	301
2c5e5fb3 MD	302
ec4e58a3	303	/* uatomic_inc */
2c5e5fb3 MD	304
2c5e5fb3 MD	305	static inline __attribute__((always_inline))
bf9de1b7	306	void __uatomic_inc(void *addr, int len)
2c5e5fb3 MD	307	{
	308	switch (len) {
	309	case 1:
	310	{
	311	__asm__ __volatile__(
	312	"lock; incb %0"
	313	: "=m"(*__hp(addr))
	314	:
	315	: "memory");
	316	return;
	317	}
	318	case 2:
	319	{
	320	__asm__ __volatile__(
	321	"lock; incw %0"
	322	: "=m"(*__hp(addr))
	323	:
	324	: "memory");
	325	return;
	326	}
	327	case 4:
	328	{
	329	__asm__ __volatile__(
	330	"lock; incl %0"
	331	: "=m"(*__hp(addr))
	332	:
	333	: "memory");
	334	return;
	335	}
	336	#if (BITS_PER_LONG == 64)
	337	case 8:
	338	{
	339	__asm__ __volatile__(
	340	"lock; incq %0"
	341	: "=m"(*__hp(addr))
	342	:
	343	: "memory");
	344	return;
	345	}
	346	#endif
	347	}
	348	/* generate an illegal instruction. Cannot catch this with linker tricks
	349	* when optimizations are disabled. */
	350	__asm__ __volatile__("ud2");
	351	return;
	352	}
	353
bf9de1b7	354	#define _uatomic_inc(addr) (__uatomic_inc((addr), sizeof(*(addr))))
2c5e5fb3	355
ec4e58a3	356	/* uatomic_dec */
2c5e5fb3 MD	357
2c5e5fb3 MD	358	static inline __attribute__((always_inline))
bf9de1b7	359	void __uatomic_dec(void *addr, int len)
2c5e5fb3 MD	360	{
	361	switch (len) {
	362	case 1:
	363	{
	364	__asm__ __volatile__(
	365	"lock; decb %0"
	366	: "=m"(*__hp(addr))
	367	:
	368	: "memory");
	369	return;
	370	}
	371	case 2:
	372	{
	373	__asm__ __volatile__(
	374	"lock; decw %0"
	375	: "=m"(*__hp(addr))
	376	:
	377	: "memory");
	378	return;
	379	}
	380	case 4:
	381	{
	382	__asm__ __volatile__(
	383	"lock; decl %0"
	384	: "=m"(*__hp(addr))
	385	:
	386	: "memory");
	387	return;
	388	}
	389	#if (BITS_PER_LONG == 64)
	390	case 8:
	391	{
	392	__asm__ __volatile__(
	393	"lock; decq %0"
	394	: "=m"(*__hp(addr))
	395	:
	396	: "memory");
	397	return;
	398	}
	399	#endif
	400	}
	401	/* generate an illegal instruction. Cannot catch this with linker tricks
	402	* when optimizations are disabled. */
	403	__asm__ __volatile__("ud2");
	404	return;
	405	}
	406
bf9de1b7	407	#define _uatomic_dec(addr) (__uatomic_dec((addr), sizeof(*(addr))))
0114ba7f	408
bf9de1b7	409	#if ((BITS_PER_LONG != 64) && defined(CONFIG_URCU_COMPAT_ARCH))
7d413817	410	extern int __urcu_cas_avail;
e1562850	411	extern int __urcu_cas_init(void);
bf9de1b7 MD	412
	413	#define UATOMIC_COMPAT(insn) \
	414	((likely(__urcu_cas_avail > 0)) \
	415	? (_uatomic_##insn) \
	416	: ((unlikely(__urcu_cas_avail < 0) \
	417	? ((__urcu_cas_init() > 0) \
	418	? (_uatomic_##insn) \
	419	: (compat_uatomic_##insn)) \
	420	: (compat_uatomic_##insn))))
	421
	422	extern unsigned long _compat_uatomic_set(void *addr,
	423	unsigned long _new, int len);
	424	#define compat_uatomic_set(addr, _new) \
	425	((__typeof__(*(addr))) _compat_uatomic_set((addr), \
	426	(unsigned long)(_new), \
	427	sizeof(*(addr))))
	428
	429
	430	extern unsigned long _compat_uatomic_xchg(void *addr,
	431	unsigned long _new, int len);
	432	#define compat_uatomic_xchg(addr, _new) \
	433	((__typeof__(*(addr))) _compat_uatomic_xchg((addr), \
	434	(unsigned long)(_new), \
	435	sizeof(*(addr))))
7d413817 MD	436
7d413817 MD	437	extern unsigned long _compat_uatomic_cmpxchg(void *addr, unsigned long old,
bf9de1b7 MD	438	unsigned long _new, int len);
	439	#define compat_uatomic_cmpxchg(addr, old, _new) \
	440	((__typeof__(*(addr))) _compat_uatomic_cmpxchg((addr), \
	441	(unsigned long)(old), \
	442	(unsigned long)(_new), \
	443	sizeof(*(addr))))
7d413817	444
bf9de1b7 MD	445	extern unsigned long _compat_uatomic_xchg(void *addr,
	446	unsigned long _new, int len);
	447	#define compat_uatomic_add_return(addr, v) \
	448	((__typeof__(*(addr))) _compat_uatomic_add_return((addr), \
	449	(unsigned long)(v), \
7d413817	450	sizeof(*(addr))))
bf9de1b7 MD	451
	452	#define compat_uatomic_sub_return(addr, v) \
	453	compat_uatomic_add_return((addr), -(v))
	454	#define compat_uatomic_add(addr, v) \
	455	((void)compat_uatomic_add_return((addr), (v)))
	456	#define compat_uatomic_sub(addr, v) \
	457	((void)compat_uatomic_sub_return((addr), (v)))
	458	#define compat_uatomic_inc(addr) \
	459	(compat_uatomic_add((addr), 1))
	460	#define compat_uatomic_dec(addr) \
	461	(compat_uatomic_sub((addr), 1))
	462
	463	#else
	464	#define UATOMIC_COMPAT(insn) (_uatomic_##insn)
7d413817 MD	465	#endif
7d413817 MD	466
bf9de1b7 MD	467	/* Read is atomic even in compat mode */
	468	#define uatomic_read(addr) _uatomic_read(addr)
	469
	470	#define uatomic_set(addr, v) \
	471	UATOMIC_COMPAT(set(addr, v))
	472	#define uatomic_cmpxchg(addr, old, _new) \
	473	UATOMIC_COMPAT(cmpxchg(addr, old, _new))
	474	#define uatomic_xchg(addr, v) \
	475	UATOMIC_COMPAT(xchg(addr, v))
	476	#define uatomic_add_return(addr, v) \
	477	UATOMIC_COMPAT(add_return(addr, v))
	478	#define uatomic_sub_return(addr, v) \
	479	UATOMIC_COMPAT(sub_return(addr, v))
	480	#define uatomic_add(addr, v) UATOMIC_COMPAT(add(addr, v))
	481	#define uatomic_sub(addr, v) UATOMIC_COMPAT(sub(addr, v))
	482	#define uatomic_inc(addr) UATOMIC_COMPAT(inc(addr))
	483	#define uatomic_dec(addr) UATOMIC_COMPAT(dec(addr))
	484
36bc70a8 MD	485	#ifdef __cplusplus
	486	}
	487	#endif
	488
ec4e58a3	489	#endif /* _URCU_ARCH_UATOMIC_X86_H */