Add powerpc atomic operations
[urcu.git] / arch_atomic_ppc.h
1 #ifndef _ARCH_ATOMIC_PPC_H
2 #define _ARCH_ATOMIC_PPC_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 libatomic_ops-1.2, inherited in part from the
20 * Boehm-Demers-Weiser conservative garbage collector.
21 */
22
23 #ifndef __SIZEOF_LONG__
24 #ifdef __powerpc64__
25 #define __SIZEOF_LONG__ 8
26 #else
27 #define __SIZEOF_LONG__ 4
28 #endif
29 #endif
30
31 #ifndef BITS_PER_LONG
32 #define BITS_PER_LONG (__SIZEOF_LONG__ * 8)
33 #endif
34
35 #define ILLEGAL_INSTR ".long 0xd00d00"
36
37 #ifndef _INCLUDE_API_H
38
39 /*
40 * Using a isync as second barrier for exchange to provide acquire semantic.
41 * According to atomic_ops/sysdeps/gcc/powerpc.h, the documentation is "fairly
42 * explicit that this also has acquire semantics."
43 * Derived from AO_compare_and_swap(), but removed the comparison.
44 */
45
46 /* xchg */
47
48 static __attribute__((always_inline))
49 unsigned long _atomic_exchange(volatile void *addr, unsigned long val, int len)
50 {
51 switch (len) {
52 case 4:
53 {
54 unsigned int result;
55
56 __asm__ __volatile__(
57 "lwsync\n"
58 "1:\t" "lwarx %0,0,%1\n" /* load and reserve */
59 "stwcx. %2,0,%1\n" /* else store conditional */
60 "bne- 1b\n" /* retry if lost reservation */
61 "isync\n"
62 : "=&r"(result)
63 : "r"(addr), "r"(val)
64 : "memory", "cc");
65
66 return result;
67 }
68 #if (BITS_PER_LONG == 64)
69 case 8:
70 {
71 unsigned long result;
72
73 __asm__ __volatile__(
74 "lwsync\n"
75 "1:\t" "ldarx %0,0,%1\n" /* load and reserve */
76 "stdcx. %2,0,%1\n" /* else store conditional */
77 "bne- 1b\n" /* retry if lost reservation */
78 "isync\n"
79 : "=&r"(result)
80 : "r"(addr), "r"(val)
81 : "memory", "cc");
82
83 return result;
84 }
85 #endif
86 }
87 /* generate an illegal instruction. Cannot catch this with linker tricks
88 * when optimizations are disabled. */
89 __asm__ __volatile__(ILLEGAL_INSTR);
90 return 0;
91 }
92
93 #define xchg(addr, v) (__typeof__(*(addr))) _atomic_exchange((addr), (v), \
94 sizeof(*(addr)))
95
96 /* cmpxchg */
97
98 static __attribute__((always_inline))
99 unsigned long _atomic_cmpxchg(volatile void *addr, unsigned long old,
100 unsigned long _new, int len)
101 {
102 switch (len) {
103 case 4:
104 {
105 unsigned int old_val;
106
107 __asm__ __volatile__(
108 "lwsync\n"
109 "1:\t" "lwarx %0,0,%1\n" /* load and reserve */
110 "cmpd %0,%3\n" /* if load is not equal to */
111 "bne 2f\n" /* old, fail */
112 "stwcx. %2,0,%1\n" /* else store conditional */
113 "bne- 1b\n" /* retry if lost reservation */
114 "isync\n"
115 "2:\n"
116 : "=&r"(old_val),
117 : "r"(addr), "r"((unsigned int)_new),
118 "r"((unsigned int)old)
119 : "memory", "cc");
120
121 return old_val;
122 }
123 #if (BITS_PER_LONG == 64)
124 case 8:
125 {
126 unsigned long old_val;
127
128 __asm__ __volatile__(
129 "lwsync\n"
130 "1:\t" "ldarx %0,0,%1\n" /* load and reserve */
131 "cmpd %0,%3\n" /* if load is not equal to */
132 "bne 2f\n" /* old, fail */
133 "stdcx. %2,0,%1\n" /* else store conditional */
134 "bne- 1b\n" /* retry if lost reservation */
135 "isync\n"
136 "2:\n"
137 : "=&r"(old_val),
138 : "r"(addr), "r"((unsigned long)_new),
139 "r"((unsigned long)old)
140 : "memory", "cc");
141
142 return old_val;
143 }
144 #endif
145 }
146 /* generate an illegal instruction. Cannot catch this with linker tricks
147 * when optimizations are disabled. */
148 __asm__ __volatile__(ILLEGAL_INSTR);
149 return 0;
150 }
151
152 #define cmpxchg(addr, old, _new) \
153 (__typeof__(*(addr))) _atomic_cmpxchg((addr), (old), (_new), \
154 sizeof(*(addr)))
155
156 /* atomic_add_return */
157
158 static __attribute__((always_inline))
159 unsigned long _atomic_add_return(volatile void *addr, unsigned long val,
160 int len)
161 {
162 switch (len) {
163 case 4:
164 {
165 unsigned int result;
166
167 __asm__ __volatile__(
168 "lwsync\n"
169 "1:\t" "lwarx %0,0,%1\n" /* load and reserve */
170 "add %0,%2,%0\n" /* add val to value loaded */
171 "stwcx. %0,0,%1\n" /* store conditional */
172 "bne- 1b\n" /* retry if lost reservation */
173 "isync\n"
174 : "=&r"(result)
175 : "r"(addr), "r"(val)
176 : "memory", "cc");
177
178 return result;
179 }
180 #if (BITS_PER_LONG == 64)
181 case 8:
182 {
183 unsigned long result;
184
185 __asm__ __volatile__(
186 "lwsync\n"
187 "1:\t" "ldarx %0,0,%1\n" /* load and reserve */
188 "add %0,%2,%0\n" /* add val to value loaded */
189 "stdcx. %0,0,%1\n" /* store conditional */
190 "bne- 1b\n" /* retry if lost reservation */
191 "isync\n"
192 : "=&r"(result)
193 : "r"(addr), "r"(val)
194 : "memory", "cc");
195
196 return result;
197 }
198 #endif
199 }
200 /* generate an illegal instruction. Cannot catch this with linker tricks
201 * when optimizations are disabled. */
202 __asm__ __volatile__(ILLEGAL_INSTR);
203 return 0;
204 }
205
206 #define atomic_add_return(addr, v) \
207 (__typeof__(*(addr))) _atomic_add((addr), (v), sizeof(*(addr)))
208
209 /* atomic_sub_return, atomic_add, atomic_sub, atomic_inc, atomic_dec */
210
211 #define atomic_sub_return(addr, v) atomic_add_return((addr), -(v))
212
213 #define atomic_add(addr, v) (void)atomic_add_return((addr), (v))
214 #define atomic_sub(addr, v) (void)atomic_sub_return((addr), (v))
215
216 #define atomic_inc(addr, v) atomic_add((addr), 1)
217 #define atomic_dec(addr, v) atomic_add((addr), -1)
218
219 #endif /* #ifndef _INCLUDE_API_H */
220
221 #endif /* ARCH_ATOMIC_PPC_H */
This page took 0.04384 seconds and 4 git commands to generate.