[urcu.git] / urcu / uatomic_arch_ppc.h

#ifndef _URCU_ARCH_UATOMIC_PPC_H
#define _URCU_ARCH_UATOMIC_PPC_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__
#ifdef __powerpc64__
#define __SIZEOF_LONG__ 8
#else
#define __SIZEOF_LONG__ 4
#endif
#endif

#ifdef __NO_LWSYNC__
#define LWSYNC_OPCODE	"sync\n"
#else
#define LWSYNC_OPCODE	"lwsync\n"
#endif

#ifndef BITS_PER_LONG
#define BITS_PER_LONG	(__SIZEOF_LONG__ * 8)
#endif

#define ILLEGAL_INSTR	".long	0xd00d00"

#define uatomic_set(addr, v)	STORE_SHARED(*(addr), (v))
#define uatomic_read(addr)	LOAD_SHARED(*(addr))

/*
 * Using a isync as second barrier for exchange to provide acquire semantic.
 * According to uatomic_ops/sysdeps/gcc/powerpc.h, the documentation is "fairly
 * explicit that this also has acquire semantics."
 * Derived from AO_compare_and_swap(), but removed the comparison.
 */

/* xchg */

static inline __attribute__((always_inline))
unsigned long _uatomic_exchange(void *addr, unsigned long val, int len)
{
	switch (len) {
	case 4:
	{
		unsigned int result;

		__asm__ __volatile__(
			LWSYNC_OPCODE
		"1:\t"	"lwarx %0,0,%1\n"	/* load and reserve */
			"stwcx. %2,0,%1\n"	/* else store conditional */
			"bne- 1b\n"	 	/* retry if lost reservation */
			"isync\n"
				: "=&r"(result)
				: "r"(addr), "r"(val)
				: "memory", "cc");

		return result;
	}
#if (BITS_PER_LONG == 64)
	case 8:
	{
		unsigned long result;

		__asm__ __volatile__(
			LWSYNC_OPCODE
		"1:\t"	"ldarx %0,0,%1\n"	/* load and reserve */
			"stdcx. %2,0,%1\n"	/* else store conditional */
			"bne- 1b\n"	 	/* retry if lost reservation */
			"isync\n"
				: "=&r"(result)
				: "r"(addr), "r"(val)
				: "memory", "cc");

		return result;
	}
#endif
	}
	/* generate an illegal instruction. Cannot catch this with linker tricks
	 * when optimizations are disabled. */
	__asm__ __volatile__(ILLEGAL_INSTR);
	return 0;
}

#define uatomic_xchg(addr, v)						    \
	((__typeof__(*(addr))) _uatomic_exchange((addr), (unsigned long)(v), \
						sizeof(*(addr))))
/* cmpxchg */

static inline __attribute__((always_inline))
unsigned long _uatomic_cmpxchg(void *addr, unsigned long old,
			      unsigned long _new, int len)
{
	switch (len) {
	case 4:
	{
		unsigned int old_val;

		__asm__ __volatile__(
			LWSYNC_OPCODE
		"1:\t"	"lwarx %0,0,%1\n"	/* load and reserve */
			"cmpd %0,%3\n"		/* if load is not equal to */
			"bne 2f\n"		/* old, fail */
			"stwcx. %2,0,%1\n"	/* else store conditional */
			"bne- 1b\n"	 	/* retry if lost reservation */
			"isync\n"
		"2:\n"
				: "=&r"(old_val)
				: "r"(addr), "r"((unsigned int)_new),
				  "r"((unsigned int)old)
				: "memory", "cc");

		return old_val;
	}
#if (BITS_PER_LONG == 64)
	case 8:
	{
		unsigned long old_val;

		__asm__ __volatile__(
			LWSYNC_OPCODE
		"1:\t"	"ldarx %0,0,%1\n"	/* load and reserve */
			"cmpd %0,%3\n"		/* if load is not equal to */
			"bne 2f\n"		/* old, fail */
			"stdcx. %2,0,%1\n"	/* else store conditional */
			"bne- 1b\n"	 	/* retry if lost reservation */
			"isync\n"
		"2:\n"
				: "=&r"(old_val),
				: "r"(addr), "r"((unsigned long)_new),
				  "r"((unsigned long)old)
				: "memory", "cc");

		return old_val;
	}
#endif
	}
	/* generate an illegal instruction. Cannot catch this with linker tricks
	 * when optimizations are disabled. */
	__asm__ __volatile__(ILLEGAL_INSTR);
	return 0;
}


#define uatomic_cmpxchg(addr, old, _new)				    \
	((__typeof__(*(addr))) _uatomic_cmpxchg((addr), (unsigned long)(old),\
						(unsigned long)(_new), 	    \
						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 4:
	{
		unsigned int result;

		__asm__ __volatile__(
			LWSYNC_OPCODE
		"1:\t"	"lwarx %0,0,%1\n"	/* load and reserve */
			"add %0,%2,%0\n"	/* add val to value loaded */
			"stwcx. %0,0,%1\n"	/* store conditional */
			"bne- 1b\n"	 	/* retry if lost reservation */
			"isync\n"
				: "=&r"(result)
				: "r"(addr), "r"(val)
				: "memory", "cc");

		return result;
	}
#if (BITS_PER_LONG == 64)
	case 8:
	{
		unsigned long result;

		__asm__ __volatile__(
			LWSYNC_OPCODE
		"1:\t"	"ldarx %0,0,%1\n"	/* load and reserve */
			"add %0,%2,%0\n"	/* add val to value loaded */
			"stdcx. %0,0,%1\n"	/* store conditional */
			"bne- 1b\n"	 	/* retry if lost reservation */
			"isync\n"
				: "=&r"(result)
				: "r"(addr), "r"(val)
				: "memory", "cc");

		return result;
	}
#endif
	}
	/* generate an illegal instruction. Cannot catch this with linker tricks
	 * when optimizations are disabled. */
	__asm__ __volatile__(ILLEGAL_INSTR);
	return 0;
}


#define uatomic_add_return(addr, v)					\
	((__typeof__(*(addr))) _uatomic_add_return((addr),		\
						  (unsigned long)(v),	\
						  sizeof(*(addr))))

/* uatomic_sub_return, uatomic_add, uatomic_sub, uatomic_inc, uatomic_dec */

#define uatomic_sub_return(addr, v)	uatomic_add_return((addr), -(v))

#define uatomic_add(addr, v)		(void)uatomic_add_return((addr), (v))
#define uatomic_sub(addr, v)		(void)uatomic_sub_return((addr), (v))

#define uatomic_inc(addr)		uatomic_add((addr), 1)
#define uatomic_dec(addr)		uatomic_add((addr), -1)

#define compat_uatomic_cmpxchg(ptr, old, _new)	uatomic_cmpxchg(ptr, old, _new)

#ifdef __cplusplus 
}
#endif

#endif /* _URCU_ARCH_UATOMIC_PPC_H */
Commit	Line	Data
ec4e58a3 MD	1	#ifndef _URCU_ARCH_UATOMIC_PPC_H
ec4e58a3 MD	2	#define _URCU_ARCH_UATOMIC_PPC_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>
b46b23cb	24	#include <urcu/system.h>
1315d277	25
36bc70a8 MD	26	#ifdef __cplusplus
	27	extern "C" {
	28	#endif
	29
bcf80111 SM	30	#ifndef __SIZEOF_LONG__
	31	#ifdef __powerpc64__
	32	#define __SIZEOF_LONG__ 8
	33	#else
	34	#define __SIZEOF_LONG__ 4
	35	#endif
	36	#endif
	37
701dd8de SAS	38	#ifdef __NO_LWSYNC__
	39	#define LWSYNC_OPCODE "sync\n"
	40	#else
	41	#define LWSYNC_OPCODE "lwsync\n"
	42	#endif
	43
0114ba7f MD	44	#ifndef BITS_PER_LONG
	45	#define BITS_PER_LONG (__SIZEOF_LONG__ * 8)
	46	#endif
	47
e7061ad2	48	#define ILLEGAL_INSTR ".long 0xd00d00"
0114ba7f	49
b46b23cb MD	50	#define uatomic_set(addr, v) STORE_SHARED(*(addr), (v))
b46b23cb MD	51	#define uatomic_read(addr) LOAD_SHARED(*(addr))
5f8052b4	52
0114ba7f MD	53	/*
0114ba7f MD	54	* Using a isync as second barrier for exchange to provide acquire semantic.
ec4e58a3	55	* According to uatomic_ops/sysdeps/gcc/powerpc.h, the documentation is "fairly
0114ba7f MD	56	* explicit that this also has acquire semantics."
	57	* Derived from AO_compare_and_swap(), but removed the comparison.
	58	*/
	59
f689dcbc MD	60	/* xchg */
f689dcbc MD	61
da1c1635	62	static inline __attribute__((always_inline))
ec4e58a3	63	unsigned long _uatomic_exchange(void *addr, unsigned long val, int len)
0114ba7f	64	{
f689dcbc MD	65	switch (len) {
	66	case 4:
	67	{
	68	unsigned int result;
	69
	70	__asm__ __volatile__(
701dd8de	71	LWSYNC_OPCODE
f689dcbc MD	72	"1:\t" "lwarx %0,0,%1\n" /* load and reserve */
	73	"stwcx. %2,0,%1\n" /* else store conditional */
	74	"bne- 1b\n" /* retry if lost reservation */
	75	"isync\n"
	76	: "=&r"(result)
	77	: "r"(addr), "r"(val)
	78	: "memory", "cc");
	79
	80	return result;
	81	}
	82	#if (BITS_PER_LONG == 64)
	83	case 8:
	84	{
	85	unsigned long result;
	86
	87	__asm__ __volatile__(
701dd8de	88	LWSYNC_OPCODE
f689dcbc MD	89	"1:\t" "ldarx %0,0,%1\n" /* load and reserve */
	90	"stdcx. %2,0,%1\n" /* else store conditional */
	91	"bne- 1b\n" /* retry if lost reservation */
	92	"isync\n"
	93	: "=&r"(result)
	94	: "r"(addr), "r"(val)
	95	: "memory", "cc");
	96
	97	return result;
	98	}
	99	#endif
	100	}
	101	/* generate an illegal instruction. Cannot catch this with linker tricks
	102	* when optimizations are disabled. */
	103	__asm__ __volatile__(ILLEGAL_INSTR);
	104	return 0;
0114ba7f MD	105	}
0114ba7f MD	106
ec4e58a3 MD	107	#define uatomic_xchg(addr, v) \
ec4e58a3 MD	108	((__typeof__(*(addr))) _uatomic_exchange((addr), (unsigned long)(v), \
da1c1635	109	sizeof(*(addr))))
f689dcbc	110	/* cmpxchg */
0114ba7f	111
da1c1635	112	static inline __attribute__((always_inline))
ec4e58a3	113	unsigned long _uatomic_cmpxchg(void *addr, unsigned long old,
f689dcbc	114	unsigned long _new, int len)
0114ba7f	115	{
f689dcbc MD	116	switch (len) {
	117	case 4:
	118	{
	119	unsigned int old_val;
	120
	121	__asm__ __volatile__(
701dd8de	122	LWSYNC_OPCODE
f689dcbc MD	123	"1:\t" "lwarx %0,0,%1\n" /* load and reserve */
	124	"cmpd %0,%3\n" /* if load is not equal to */
	125	"bne 2f\n" /* old, fail */
	126	"stwcx. %2,0,%1\n" /* else store conditional */
	127	"bne- 1b\n" /* retry if lost reservation */
	128	"isync\n"
	129	"2:\n"
e72f4937	130	: "=&r"(old_val)
f689dcbc MD	131	: "r"(addr), "r"((unsigned int)_new),
	132	"r"((unsigned int)old)
	133	: "memory", "cc");
	134
	135	return old_val;
	136	}
	137	#if (BITS_PER_LONG == 64)
	138	case 8:
	139	{
	140	unsigned long old_val;
	141
	142	__asm__ __volatile__(
701dd8de	143	LWSYNC_OPCODE
f689dcbc MD	144	"1:\t" "ldarx %0,0,%1\n" /* load and reserve */
	145	"cmpd %0,%3\n" /* if load is not equal to */
	146	"bne 2f\n" /* old, fail */
	147	"stdcx. %2,0,%1\n" /* else store conditional */
	148	"bne- 1b\n" /* retry if lost reservation */
	149	"isync\n"
	150	"2:\n"
	151	: "=&r"(old_val),
	152	: "r"(addr), "r"((unsigned long)_new),
	153	"r"((unsigned long)old)
	154	: "memory", "cc");
	155
	156	return old_val;
	157	}
	158	#endif
	159	}
	160	/* generate an illegal instruction. Cannot catch this with linker tricks
	161	* when optimizations are disabled. */
	162	__asm__ __volatile__(ILLEGAL_INSTR);
	163	return 0;
0114ba7f MD	164	}
0114ba7f MD	165
da1c1635	166
ec4e58a3 MD	167	#define uatomic_cmpxchg(addr, old, _new) \
ec4e58a3 MD	168	((__typeof__(*(addr))) _uatomic_cmpxchg((addr), (unsigned long)(old),\
da1c1635 MD	169	(unsigned long)(_new), \
da1c1635 MD	170	sizeof(*(addr))))
f689dcbc	171
ec4e58a3	172	/* uatomic_add_return */
0114ba7f	173
da1c1635	174	static inline __attribute__((always_inline))
ec4e58a3	175	unsigned long _uatomic_add_return(void *addr, unsigned long val,
f689dcbc	176	int len)
0114ba7f MD	177	{
0114ba7f MD	178	switch (len) {
f689dcbc MD	179	case 4:
	180	{
	181	unsigned int result;
	182
	183	__asm__ __volatile__(
701dd8de	184	LWSYNC_OPCODE
f689dcbc MD	185	"1:\t" "lwarx %0,0,%1\n" /* load and reserve */
	186	"add %0,%2,%0\n" /* add val to value loaded */
	187	"stwcx. %0,0,%1\n" /* store conditional */
	188	"bne- 1b\n" /* retry if lost reservation */
	189	"isync\n"
	190	: "=&r"(result)
	191	: "r"(addr), "r"(val)
	192	: "memory", "cc");
	193
	194	return result;
	195	}
0114ba7f	196	#if (BITS_PER_LONG == 64)
f689dcbc MD	197	case 8:
	198	{
	199	unsigned long result;
	200
	201	__asm__ __volatile__(
701dd8de	202	LWSYNC_OPCODE
f689dcbc MD	203	"1:\t" "ldarx %0,0,%1\n" /* load and reserve */
	204	"add %0,%2,%0\n" /* add val to value loaded */
	205	"stdcx. %0,0,%1\n" /* store conditional */
	206	"bne- 1b\n" /* retry if lost reservation */
	207	"isync\n"
	208	: "=&r"(result)
	209	: "r"(addr), "r"(val)
	210	: "memory", "cc");
	211
	212	return result;
	213	}
0114ba7f MD	214	#endif
	215	}
	216	/* generate an illegal instruction. Cannot catch this with linker tricks
	217	* when optimizations are disabled. */
	218	__asm__ __volatile__(ILLEGAL_INSTR);
	219	return 0;
	220	}
	221
da1c1635	222
ec4e58a3 MD	223	#define uatomic_add_return(addr, v) \
ec4e58a3 MD	224	((__typeof__(*(addr))) _uatomic_add_return((addr), \
da1c1635 MD	225	(unsigned long)(v), \
da1c1635 MD	226	sizeof(*(addr))))
f689dcbc	227
ec4e58a3	228	/* uatomic_sub_return, uatomic_add, uatomic_sub, uatomic_inc, uatomic_dec */
f689dcbc	229
ec4e58a3	230	#define uatomic_sub_return(addr, v) uatomic_add_return((addr), -(v))
f689dcbc	231
ec4e58a3 MD	232	#define uatomic_add(addr, v) (void)uatomic_add_return((addr), (v))
ec4e58a3 MD	233	#define uatomic_sub(addr, v) (void)uatomic_sub_return((addr), (v))
f689dcbc	234
ec4e58a3 MD	235	#define uatomic_inc(addr) uatomic_add((addr), 1)
ec4e58a3 MD	236	#define uatomic_dec(addr) uatomic_add((addr), -1)
0114ba7f	237
7d413817 MD	238	#define compat_uatomic_cmpxchg(ptr, old, _new) uatomic_cmpxchg(ptr, old, _new)
7d413817 MD	239
36bc70a8 MD	240	#ifdef __cplusplus
	241	}
	242	#endif
	243
ec4e58a3	244	#endif /* _URCU_ARCH_UATOMIC_PPC_H */