[userspace-rcu.git] / urcu / uatomic / 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_and */

static inline __attribute__((always_inline))
void __uatomic_and(void *addr, unsigned long val, int len)
{
	switch (len) {
	case 1:
	{
		__asm__ __volatile__(
		"lock; andb %1, %0"
			: "=m"(*__hp(addr))
			: "iq" ((unsigned char)val)
			: "memory");
		return;
	}
	case 2:
	{
		__asm__ __volatile__(
		"lock; andw %1, %0"
			: "=m"(*__hp(addr))
			: "ir" ((unsigned short)val)
			: "memory");
		return;
	}
	case 4:
	{
		__asm__ __volatile__(
		"lock; andl %1, %0"
			: "=m"(*__hp(addr))
			: "ir" ((unsigned int)val)
			: "memory");
		return;
	}
#if (CAA_BITS_PER_LONG == 64)
	case 8:
	{
		__asm__ __volatile__(
		"lock; andq %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_and(addr, v)						   \
	(__uatomic_and((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 void _compat_uatomic_and(void *addr, unsigned long _new, int len);
#define compat_uatomic_and(addr, v)				       \
	(_compat_uatomic_and((addr),				       \
			(unsigned long)(v),			       \
			sizeof(*(addr))))

extern void _compat_uatomic_or(void *addr, unsigned long _new, int len);
#define compat_uatomic_or(addr, v)				       \
	(_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_and(addr, v)		\
		UATOMIC_COMPAT(and(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
	1	#ifndef _URCU_ARCH_UATOMIC_X86_H
	2	#define _URCU_ARCH_UATOMIC_X86_H
	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	*
	19	* Code inspired from libuatomic_ops-1.2, inherited in part from the
	20	* Boehm-Demers-Weiser conservative garbage collector.
	21	*/
	22
	23	#include <urcu/compiler.h>
	24	#include <urcu/system.h>
	25
	26	#define UATOMIC_HAS_ATOMIC_BYTE
	27	#define UATOMIC_HAS_ATOMIC_SHORT
	28
	29	#ifdef __cplusplus
	30	extern "C" {
	31	#endif
	32
	33	/*
	34	* Derived from AO_compare_and_swap() and AO_test_and_set_full().
	35	*/
	36
	37	struct __uatomic_dummy {
	38	unsigned long v[10];
	39	};
	40	#define __hp(x) ((struct __uatomic_dummy *)(x))
	41
	42	#define _uatomic_set(addr, v) CMM_STORE_SHARED(*(addr), (v))
	43
	44	/* cmpxchg */
	45
	46	static inline __attribute__((always_inline))
	47	unsigned long __uatomic_cmpxchg(void *addr, unsigned long old,
	48	unsigned long _new, int len)
	49	{
	50	switch (len) {
	51	case 1:
	52	{
	53	unsigned char result = old;
	54
	55	__asm__ __volatile__(
	56	"lock; cmpxchgb %2, %1"
	57	: "+a"(result), "+m"(*__hp(addr))
	58	: "q"((unsigned char)_new)
	59	: "memory");
	60	return result;
	61	}
	62	case 2:
	63	{
	64	unsigned short result = old;
	65
	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;
	76
	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	}
	84	#if (CAA_BITS_PER_LONG == 64)
	85	case 8:
	86	{
	87	unsigned long result = old;
	88
	89	__asm__ __volatile__(
	90	"lock; cmpxchgq %2, %1"
	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;
	102	}
	103
	104	#define _uatomic_cmpxchg(addr, old, _new) \
	105	((__typeof__(*(addr))) __uatomic_cmpxchg((addr), (unsigned long)(old),\
	106	(unsigned long)(_new), \
	107	sizeof(*(addr))))
	108
	109	/* xchg */
	110
	111	static inline __attribute__((always_inline))
	112	unsigned long __uatomic_exchange(void *addr, unsigned long val, int len)
	113	{
	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	}
	146	#if (CAA_BITS_PER_LONG == 64)
	147	case 8:
	148	{
	149	unsigned long result;
	150	__asm__ __volatile__(
	151	"xchgq %0, %1"
	152	: "=r"(result), "+m"(*__hp(addr))
	153	: "0" ((unsigned long)val)
	154	: "memory");
	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;
	163	}
	164
	165	#define _uatomic_xchg(addr, v) \
	166	((__typeof__(*(addr))) __uatomic_exchange((addr), (unsigned long)(v), \
	167	sizeof(*(addr))))
	168
	169	/* uatomic_add_return */
	170
	171	static inline __attribute__((always_inline))
	172	unsigned long __uatomic_add_return(void *addr, unsigned long val,
	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	}
	209	#if (CAA_BITS_PER_LONG == 64)
	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
	229	#define _uatomic_add_return(addr, v) \
	230	((__typeof__(*(addr))) __uatomic_add_return((addr), \
	231	(unsigned long)(v), \
	232	sizeof(*(addr))))
	233
	234	/* uatomic_and */
	235
	236	static inline __attribute__((always_inline))
	237	void __uatomic_and(void *addr, unsigned long val, int len)
	238	{
	239	switch (len) {
	240	case 1:
	241	{
	242	__asm__ __volatile__(
	243	"lock; andb %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; andw %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; andl %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; andq %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_and(addr, v) \
	286	(__uatomic_and((addr), (unsigned long)(v), sizeof(*(addr))))
	287
	288	/* uatomic_or */
	289
	290	static inline __attribute__((always_inline))
	291	void __uatomic_or(void *addr, unsigned long val, int len)
	292	{
	293	switch (len) {
	294	case 1:
	295	{
	296	__asm__ __volatile__(
	297	"lock; orb %1, %0"
	298	: "=m"(*__hp(addr))
	299	: "iq" ((unsigned char)val)
	300	: "memory");
	301	return;
	302	}
	303	case 2:
	304	{
	305	__asm__ __volatile__(
	306	"lock; orw %1, %0"
	307	: "=m"(*__hp(addr))
	308	: "ir" ((unsigned short)val)
	309	: "memory");
	310	return;
	311	}
	312	case 4:
	313	{
	314	__asm__ __volatile__(
	315	"lock; orl %1, %0"
	316	: "=m"(*__hp(addr))
	317	: "ir" ((unsigned int)val)
	318	: "memory");
	319	return;
	320	}
	321	#if (CAA_BITS_PER_LONG == 64)
	322	case 8:
	323	{
	324	__asm__ __volatile__(
	325	"lock; orq %1, %0"
	326	: "=m"(*__hp(addr))
	327	: "er" ((unsigned long)val)
	328	: "memory");
	329	return;
	330	}
	331	#endif
	332	}
	333	/* generate an illegal instruction. Cannot catch this with linker tricks
	334	* when optimizations are disabled. */
	335	__asm__ __volatile__("ud2");
	336	return;
	337	}
	338
	339	#define _uatomic_or(addr, v) \
	340	(__uatomic_or((addr), (unsigned long)(v), sizeof(*(addr))))
	341
	342	/* uatomic_add */
	343
	344	static inline __attribute__((always_inline))
	345	void __uatomic_add(void *addr, unsigned long val, int len)
	346	{
	347	switch (len) {
	348	case 1:
	349	{
	350	__asm__ __volatile__(
	351	"lock; addb %1, %0"
	352	: "=m"(*__hp(addr))
	353	: "iq" ((unsigned char)val)
	354	: "memory");
	355	return;
	356	}
	357	case 2:
	358	{
	359	__asm__ __volatile__(
	360	"lock; addw %1, %0"
	361	: "=m"(*__hp(addr))
	362	: "ir" ((unsigned short)val)
	363	: "memory");
	364	return;
	365	}
	366	case 4:
	367	{
	368	__asm__ __volatile__(
	369	"lock; addl %1, %0"
	370	: "=m"(*__hp(addr))
	371	: "ir" ((unsigned int)val)
	372	: "memory");
	373	return;
	374	}
	375	#if (CAA_BITS_PER_LONG == 64)
	376	case 8:
	377	{
	378	__asm__ __volatile__(
	379	"lock; addq %1, %0"
	380	: "=m"(*__hp(addr))
	381	: "er" ((unsigned long)val)
	382	: "memory");
	383	return;
	384	}
	385	#endif
	386	}
	387	/* generate an illegal instruction. Cannot catch this with linker tricks
	388	* when optimizations are disabled. */
	389	__asm__ __volatile__("ud2");
	390	return;
	391	}
	392
	393	#define _uatomic_add(addr, v) \
	394	(__uatomic_add((addr), (unsigned long)(v), sizeof(*(addr))))
	395
	396
	397	/* uatomic_inc */
	398
	399	static inline __attribute__((always_inline))
	400	void __uatomic_inc(void *addr, int len)
	401	{
	402	switch (len) {
	403	case 1:
	404	{
	405	__asm__ __volatile__(
	406	"lock; incb %0"
	407	: "=m"(*__hp(addr))
	408	:
	409	: "memory");
	410	return;
	411	}
	412	case 2:
	413	{
	414	__asm__ __volatile__(
	415	"lock; incw %0"
	416	: "=m"(*__hp(addr))
	417	:
	418	: "memory");
	419	return;
	420	}
	421	case 4:
	422	{
	423	__asm__ __volatile__(
	424	"lock; incl %0"
	425	: "=m"(*__hp(addr))
	426	:
	427	: "memory");
	428	return;
	429	}
	430	#if (CAA_BITS_PER_LONG == 64)
	431	case 8:
	432	{
	433	__asm__ __volatile__(
	434	"lock; incq %0"
	435	: "=m"(*__hp(addr))
	436	:
	437	: "memory");
	438	return;
	439	}
	440	#endif
	441	}
	442	/* generate an illegal instruction. Cannot catch this with linker tricks
	443	* when optimizations are disabled. */
	444	__asm__ __volatile__("ud2");
	445	return;
	446	}
	447
	448	#define _uatomic_inc(addr) (__uatomic_inc((addr), sizeof(*(addr))))
	449
	450	/* uatomic_dec */
	451
	452	static inline __attribute__((always_inline))
	453	void __uatomic_dec(void *addr, int len)
	454	{
	455	switch (len) {
	456	case 1:
	457	{
	458	__asm__ __volatile__(
	459	"lock; decb %0"
	460	: "=m"(*__hp(addr))
	461	:
	462	: "memory");
	463	return;
	464	}
	465	case 2:
	466	{
	467	__asm__ __volatile__(
	468	"lock; decw %0"
	469	: "=m"(*__hp(addr))
	470	:
	471	: "memory");
	472	return;
	473	}
	474	case 4:
	475	{
	476	__asm__ __volatile__(
	477	"lock; decl %0"
	478	: "=m"(*__hp(addr))
	479	:
	480	: "memory");
	481	return;
	482	}
	483	#if (CAA_BITS_PER_LONG == 64)
	484	case 8:
	485	{
	486	__asm__ __volatile__(
	487	"lock; decq %0"
	488	: "=m"(*__hp(addr))
	489	:
	490	: "memory");
	491	return;
	492	}
	493	#endif
	494	}
	495	/* generate an illegal instruction. Cannot catch this with linker tricks
	496	* when optimizations are disabled. */
	497	__asm__ __volatile__("ud2");
	498	return;
	499	}
	500
	501	#define _uatomic_dec(addr) (__uatomic_dec((addr), sizeof(*(addr))))
	502
	503	#if ((CAA_BITS_PER_LONG != 64) && defined(CONFIG_RCU_COMPAT_ARCH))
	504	extern int __rcu_cas_avail;
	505	extern int __rcu_cas_init(void);
	506
	507	#define UATOMIC_COMPAT(insn) \
	508	((likely(__rcu_cas_avail > 0)) \
	509	? (_uatomic_##insn) \
	510	: ((unlikely(__rcu_cas_avail < 0) \
	511	? ((__rcu_cas_init() > 0) \
	512	? (_uatomic_##insn) \
	513	: (compat_uatomic_##insn)) \
	514	: (compat_uatomic_##insn))))
	515
	516	extern unsigned long _compat_uatomic_set(void *addr,
	517	unsigned long _new, int len);
	518	#define compat_uatomic_set(addr, _new) \
	519	((__typeof__(*(addr))) _compat_uatomic_set((addr), \
	520	(unsigned long)(_new), \
	521	sizeof(*(addr))))
	522
	523
	524	extern unsigned long _compat_uatomic_xchg(void *addr,
	525	unsigned long _new, int len);
	526	#define compat_uatomic_xchg(addr, _new) \
	527	((__typeof__(*(addr))) _compat_uatomic_xchg((addr), \
	528	(unsigned long)(_new), \
	529	sizeof(*(addr))))
	530
	531	extern unsigned long _compat_uatomic_cmpxchg(void *addr, unsigned long old,
	532	unsigned long _new, int len);
	533	#define compat_uatomic_cmpxchg(addr, old, _new) \
	534	((__typeof__(*(addr))) _compat_uatomic_cmpxchg((addr), \
	535	(unsigned long)(old), \
	536	(unsigned long)(_new), \
	537	sizeof(*(addr))))
	538
	539	extern void _compat_uatomic_and(void *addr, unsigned long _new, int len);
	540	#define compat_uatomic_and(addr, v) \
	541	(_compat_uatomic_and((addr), \
	542	(unsigned long)(v), \
	543	sizeof(*(addr))))
	544
	545	extern void _compat_uatomic_or(void *addr, unsigned long _new, int len);
	546	#define compat_uatomic_or(addr, v) \
	547	(_compat_uatomic_or((addr), \
	548	(unsigned long)(v), \
	549	sizeof(*(addr))))
	550
	551	extern unsigned long _compat_uatomic_add_return(void *addr,
	552	unsigned long _new, int len);
	553	#define compat_uatomic_add_return(addr, v) \
	554	((__typeof__(*(addr))) _compat_uatomic_add_return((addr), \
	555	(unsigned long)(v), \
	556	sizeof(*(addr))))
	557
	558	#define compat_uatomic_add(addr, v) \
	559	((void)compat_uatomic_add_return((addr), (v)))
	560	#define compat_uatomic_inc(addr) \
	561	(compat_uatomic_add((addr), 1))
	562	#define compat_uatomic_dec(addr) \
	563	(compat_uatomic_add((addr), -1))
	564
	565	#else
	566	#define UATOMIC_COMPAT(insn) (_uatomic_##insn)
	567	#endif
	568
	569	/* Read is atomic even in compat mode */
	570	#define uatomic_set(addr, v) \
	571	UATOMIC_COMPAT(set(addr, v))
	572
	573	#define uatomic_cmpxchg(addr, old, _new) \
	574	UATOMIC_COMPAT(cmpxchg(addr, old, _new))
	575	#define uatomic_xchg(addr, v) \
	576	UATOMIC_COMPAT(xchg(addr, v))
	577	#define uatomic_and(addr, v) \
	578	UATOMIC_COMPAT(and(addr, v))
	579	#define uatomic_or(addr, v) \
	580	UATOMIC_COMPAT(or(addr, v))
	581	#define uatomic_add_return(addr, v) \
	582	UATOMIC_COMPAT(add_return(addr, v))
	583
	584	#define uatomic_add(addr, v) UATOMIC_COMPAT(add(addr, v))
	585	#define uatomic_inc(addr) UATOMIC_COMPAT(inc(addr))
	586	#define uatomic_dec(addr) UATOMIC_COMPAT(dec(addr))
	587
	588	#ifdef __cplusplus
	589	}
	590	#endif
	591
	592	#include <urcu/uatomic/generic.h>
	593
	594	#endif /* _URCU_ARCH_UATOMIC_X86_H */