xref: /linux/arch/arm64/include/asm/cmpxchg.h (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Based on arch/arm/include/asm/cmpxchg.h
4  *
5  * Copyright (C) 2012 ARM Ltd.
6  */
7 #ifndef __ASM_CMPXCHG_H
8 #define __ASM_CMPXCHG_H
9 
10 #include <linux/build_bug.h>
11 #include <linux/compiler.h>
12 
13 #include <asm/barrier.h>
14 #include <asm/lse.h>
15 
16 /*
17  * We need separate acquire parameters for ll/sc and lse, since the full
18  * barrier case is generated as release+dmb for the former and
19  * acquire+release for the latter.
20  */
21 #define __XCHG_CASE(w, sfx, name, sz, mb, nop_lse, acq, acq_lse, rel, cl)	\
22 static inline u##sz __xchg_case_##name##sz(u##sz x, volatile void *ptr)		\
23 {										\
24 	u##sz ret;								\
25 	unsigned long tmp;							\
26 										\
27 	asm volatile(ARM64_LSE_ATOMIC_INSN(					\
28 	/* LL/SC */								\
29 	"	prfm	pstl1strm, %2\n"					\
30 	"1:	ld" #acq "xr" #sfx "\t%" #w "0, %2\n"				\
31 	"	st" #rel "xr" #sfx "\t%w1, %" #w "3, %2\n"			\
32 	"	cbnz	%w1, 1b\n"						\
33 	"	" #mb,								\
34 	/* LSE atomics */							\
35 	"	swp" #acq_lse #rel #sfx "\t%" #w "3, %" #w "0, %2\n"		\
36 		__nops(3)							\
37 	"	" #nop_lse)							\
38 	: "=&r" (ret), "=&r" (tmp), "+Q" (*(u##sz *)ptr)			\
39 	: "r" (x)								\
40 	: cl);									\
41 										\
42 	return ret;								\
43 }
44 
45 __XCHG_CASE(w, b,     ,  8,        ,    ,  ,  ,  ,         )
46 __XCHG_CASE(w, h,     , 16,        ,    ,  ,  ,  ,         )
47 __XCHG_CASE(w,  ,     , 32,        ,    ,  ,  ,  ,         )
48 __XCHG_CASE( ,  ,     , 64,        ,    ,  ,  ,  ,         )
49 __XCHG_CASE(w, b, acq_,  8,        ,    , a, a,  , "memory")
50 __XCHG_CASE(w, h, acq_, 16,        ,    , a, a,  , "memory")
51 __XCHG_CASE(w,  , acq_, 32,        ,    , a, a,  , "memory")
52 __XCHG_CASE( ,  , acq_, 64,        ,    , a, a,  , "memory")
53 __XCHG_CASE(w, b, rel_,  8,        ,    ,  ,  , l, "memory")
54 __XCHG_CASE(w, h, rel_, 16,        ,    ,  ,  , l, "memory")
55 __XCHG_CASE(w,  , rel_, 32,        ,    ,  ,  , l, "memory")
56 __XCHG_CASE( ,  , rel_, 64,        ,    ,  ,  , l, "memory")
57 __XCHG_CASE(w, b,  mb_,  8, dmb ish, nop,  , a, l, "memory")
58 __XCHG_CASE(w, h,  mb_, 16, dmb ish, nop,  , a, l, "memory")
59 __XCHG_CASE(w,  ,  mb_, 32, dmb ish, nop,  , a, l, "memory")
60 __XCHG_CASE( ,  ,  mb_, 64, dmb ish, nop,  , a, l, "memory")
61 
62 #undef __XCHG_CASE
63 
64 #define __XCHG_GEN(sfx)							\
65 static __always_inline unsigned long					\
66 __arch_xchg##sfx(unsigned long x, volatile void *ptr, int size)		\
67 {									\
68 	switch (size) {							\
69 	case 1:								\
70 		return __xchg_case##sfx##_8(x, ptr);			\
71 	case 2:								\
72 		return __xchg_case##sfx##_16(x, ptr);			\
73 	case 4:								\
74 		return __xchg_case##sfx##_32(x, ptr);			\
75 	case 8:								\
76 		return __xchg_case##sfx##_64(x, ptr);			\
77 	default:							\
78 		BUILD_BUG();						\
79 	}								\
80 									\
81 	unreachable();							\
82 }
83 
84 __XCHG_GEN()
85 __XCHG_GEN(_acq)
86 __XCHG_GEN(_rel)
87 __XCHG_GEN(_mb)
88 
89 #undef __XCHG_GEN
90 
91 #define __xchg_wrapper(sfx, ptr, x)					\
92 ({									\
93 	__typeof__(*(ptr)) __ret;					\
94 	__ret = (__typeof__(*(ptr)))					\
95 		__arch_xchg##sfx((unsigned long)(x), (ptr), sizeof(*(ptr))); \
96 	__ret;								\
97 })
98 
99 /* xchg */
100 #define arch_xchg_relaxed(...)	__xchg_wrapper(    , __VA_ARGS__)
101 #define arch_xchg_acquire(...)	__xchg_wrapper(_acq, __VA_ARGS__)
102 #define arch_xchg_release(...)	__xchg_wrapper(_rel, __VA_ARGS__)
103 #define arch_xchg(...)		__xchg_wrapper( _mb, __VA_ARGS__)
104 
105 #define __CMPXCHG_CASE(name, sz)			\
106 static inline u##sz __cmpxchg_case_##name##sz(volatile void *ptr,	\
107 					      u##sz old,		\
108 					      u##sz new)		\
109 {									\
110 	return __lse_ll_sc_body(_cmpxchg_case_##name##sz,		\
111 				ptr, old, new);				\
112 }
113 
114 __CMPXCHG_CASE(    ,  8)
115 __CMPXCHG_CASE(    , 16)
116 __CMPXCHG_CASE(    , 32)
117 __CMPXCHG_CASE(    , 64)
118 __CMPXCHG_CASE(acq_,  8)
119 __CMPXCHG_CASE(acq_, 16)
120 __CMPXCHG_CASE(acq_, 32)
121 __CMPXCHG_CASE(acq_, 64)
122 __CMPXCHG_CASE(rel_,  8)
123 __CMPXCHG_CASE(rel_, 16)
124 __CMPXCHG_CASE(rel_, 32)
125 __CMPXCHG_CASE(rel_, 64)
126 __CMPXCHG_CASE(mb_,  8)
127 __CMPXCHG_CASE(mb_, 16)
128 __CMPXCHG_CASE(mb_, 32)
129 __CMPXCHG_CASE(mb_, 64)
130 
131 #undef __CMPXCHG_CASE
132 
133 #define __CMPXCHG128(name)						\
134 static inline u128 __cmpxchg128##name(volatile u128 *ptr,		\
135 				      u128 old, u128 new)		\
136 {									\
137 	return __lse_ll_sc_body(_cmpxchg128##name,			\
138 				ptr, old, new);				\
139 }
140 
141 __CMPXCHG128(   )
142 __CMPXCHG128(_mb)
143 
144 #undef __CMPXCHG128
145 
146 #define __CMPXCHG_GEN(sfx)						\
147 static __always_inline unsigned long __cmpxchg##sfx(volatile void *ptr,	\
148 					   unsigned long old,		\
149 					   unsigned long new,		\
150 					   int size)			\
151 {									\
152 	switch (size) {							\
153 	case 1:								\
154 		return __cmpxchg_case##sfx##_8(ptr, old, new);		\
155 	case 2:								\
156 		return __cmpxchg_case##sfx##_16(ptr, old, new);		\
157 	case 4:								\
158 		return __cmpxchg_case##sfx##_32(ptr, old, new);		\
159 	case 8:								\
160 		return __cmpxchg_case##sfx##_64(ptr, old, new);		\
161 	default:							\
162 		BUILD_BUG();						\
163 	}								\
164 									\
165 	unreachable();							\
166 }
167 
168 __CMPXCHG_GEN()
169 __CMPXCHG_GEN(_acq)
170 __CMPXCHG_GEN(_rel)
171 __CMPXCHG_GEN(_mb)
172 
173 #undef __CMPXCHG_GEN
174 
175 #define __cmpxchg_wrapper(sfx, ptr, o, n)				\
176 ({									\
177 	__typeof__(*(ptr)) __ret;					\
178 	__ret = (__typeof__(*(ptr)))					\
179 		__cmpxchg##sfx((ptr), (unsigned long)(o),		\
180 				(unsigned long)(n), sizeof(*(ptr)));	\
181 	__ret;								\
182 })
183 
184 /* cmpxchg */
185 #define arch_cmpxchg_relaxed(...)	__cmpxchg_wrapper(    , __VA_ARGS__)
186 #define arch_cmpxchg_acquire(...)	__cmpxchg_wrapper(_acq, __VA_ARGS__)
187 #define arch_cmpxchg_release(...)	__cmpxchg_wrapper(_rel, __VA_ARGS__)
188 #define arch_cmpxchg(...)		__cmpxchg_wrapper( _mb, __VA_ARGS__)
189 #define arch_cmpxchg_local		arch_cmpxchg_relaxed
190 
191 /* cmpxchg64 */
192 #define arch_cmpxchg64_relaxed		arch_cmpxchg_relaxed
193 #define arch_cmpxchg64_acquire		arch_cmpxchg_acquire
194 #define arch_cmpxchg64_release		arch_cmpxchg_release
195 #define arch_cmpxchg64			arch_cmpxchg
196 #define arch_cmpxchg64_local		arch_cmpxchg_local
197 
198 /* cmpxchg128 */
199 #define system_has_cmpxchg128()		1
200 
201 #define arch_cmpxchg128(ptr, o, n)						\
202 ({										\
203 	__cmpxchg128_mb((ptr), (o), (n));					\
204 })
205 
206 #define arch_cmpxchg128_local(ptr, o, n)					\
207 ({										\
208 	__cmpxchg128((ptr), (o), (n));						\
209 })
210 
211 #define __CMPWAIT_CASE(w, sfx, sz)					\
212 static inline void __cmpwait_case_##sz(volatile void *ptr,		\
213 				       unsigned long val)		\
214 {									\
215 	unsigned long tmp;						\
216 									\
217 	asm volatile(							\
218 	"	sevl\n"							\
219 	"	wfe\n"							\
220 	"	ldxr" #sfx "\t%" #w "[tmp], %[v]\n"			\
221 	"	eor	%" #w "[tmp], %" #w "[tmp], %" #w "[val]\n"	\
222 	"	cbnz	%" #w "[tmp], 1f\n"				\
223 	"	wfe\n"							\
224 	"1:"								\
225 	: [tmp] "=&r" (tmp), [v] "+Q" (*(u##sz *)ptr)			\
226 	: [val] "r" (val));						\
227 }
228 
229 __CMPWAIT_CASE(w, b, 8);
230 __CMPWAIT_CASE(w, h, 16);
231 __CMPWAIT_CASE(w,  , 32);
232 __CMPWAIT_CASE( ,  , 64);
233 
234 #undef __CMPWAIT_CASE
235 
236 #define __CMPWAIT_GEN(sfx)						\
237 static __always_inline void __cmpwait##sfx(volatile void *ptr,		\
238 				  unsigned long val,			\
239 				  int size)				\
240 {									\
241 	switch (size) {							\
242 	case 1:								\
243 		return __cmpwait_case##sfx##_8(ptr, (u8)val);		\
244 	case 2:								\
245 		return __cmpwait_case##sfx##_16(ptr, (u16)val);		\
246 	case 4:								\
247 		return __cmpwait_case##sfx##_32(ptr, val);		\
248 	case 8:								\
249 		return __cmpwait_case##sfx##_64(ptr, val);		\
250 	default:							\
251 		BUILD_BUG();						\
252 	}								\
253 									\
254 	unreachable();							\
255 }
256 
257 __CMPWAIT_GEN()
258 
259 #undef __CMPWAIT_GEN
260 
261 #define __cmpwait_relaxed(ptr, val) \
262 	__cmpwait((ptr), (unsigned long)(val), sizeof(*(ptr)))
263 
264 #endif	/* __ASM_CMPXCHG_H */
265