1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4  * Copyright (C) 2012 Regents of the University of California
5  * Copyright (C) 2017 SiFive
6  */
7 
8 #ifndef _ASM_RISCV_ATOMIC_H
9 #define _ASM_RISCV_ATOMIC_H
10 
11 #ifdef CONFIG_GENERIC_ATOMIC64
12 # include <asm-generic/atomic64.h>
13 #else
14 # if (__riscv_xlen < 64)
15 #  error "64-bit atomics require XLEN to be at least 64"
16 # endif
17 #endif
18 
19 #include <asm/cmpxchg.h>
20 #include <asm/barrier.h>
21 
22 #define __atomic_acquire_fence()					\
23 	__asm__ __volatile__(RISCV_ACQUIRE_BARRIER "" ::: "memory")
24 
25 #define __atomic_release_fence()					\
26 	__asm__ __volatile__(RISCV_RELEASE_BARRIER "" ::: "memory");
27 
arch_atomic_read(const atomic_t * v)28 static __always_inline int arch_atomic_read(const atomic_t *v)
29 {
30 	return READ_ONCE(v->counter);
31 }
arch_atomic_set(atomic_t * v,int i)32 static __always_inline void arch_atomic_set(atomic_t *v, int i)
33 {
34 	WRITE_ONCE(v->counter, i);
35 }
36 
37 #ifndef CONFIG_GENERIC_ATOMIC64
38 #define ATOMIC64_INIT(i) { (i) }
arch_atomic64_read(const atomic64_t * v)39 static __always_inline s64 arch_atomic64_read(const atomic64_t *v)
40 {
41 	return READ_ONCE(v->counter);
42 }
arch_atomic64_set(atomic64_t * v,s64 i)43 static __always_inline void arch_atomic64_set(atomic64_t *v, s64 i)
44 {
45 	WRITE_ONCE(v->counter, i);
46 }
47 #endif
48 
49 /*
50  * First, the atomic ops that have no ordering constraints and therefor don't
51  * have the AQ or RL bits set.  These don't return anything, so there's only
52  * one version to worry about.
53  */
54 #define ATOMIC_OP(op, asm_op, I, asm_type, c_type, prefix)		\
55 static __always_inline							\
56 void arch_atomic##prefix##_##op(c_type i, atomic##prefix##_t *v)	\
57 {									\
58 	__asm__ __volatile__ (						\
59 		"	amo" #asm_op "." #asm_type " zero, %1, %0"	\
60 		: "+A" (v->counter)					\
61 		: "r" (I)						\
62 		: "memory");						\
63 }									\
64 
65 #ifdef CONFIG_GENERIC_ATOMIC64
66 #define ATOMIC_OPS(op, asm_op, I)					\
67         ATOMIC_OP (op, asm_op, I, w, int,   )
68 #else
69 #define ATOMIC_OPS(op, asm_op, I)					\
70         ATOMIC_OP (op, asm_op, I, w, int,   )				\
71         ATOMIC_OP (op, asm_op, I, d, s64, 64)
72 #endif
73 
ATOMIC_OPS(add,add,i)74 ATOMIC_OPS(add, add,  i)
75 ATOMIC_OPS(sub, add, -i)
76 ATOMIC_OPS(and, and,  i)
77 ATOMIC_OPS( or,  or,  i)
78 ATOMIC_OPS(xor, xor,  i)
79 
80 #undef ATOMIC_OP
81 #undef ATOMIC_OPS
82 
83 /*
84  * Atomic ops that have ordered, relaxed, acquire, and release variants.
85  * There's two flavors of these: the arithmatic ops have both fetch and return
86  * versions, while the logical ops only have fetch versions.
87  */
88 #define ATOMIC_FETCH_OP(op, asm_op, I, asm_type, c_type, prefix)	\
89 static __always_inline							\
90 c_type arch_atomic##prefix##_fetch_##op##_relaxed(c_type i,		\
91 					     atomic##prefix##_t *v)	\
92 {									\
93 	register c_type ret;						\
94 	__asm__ __volatile__ (						\
95 		"	amo" #asm_op "." #asm_type " %1, %2, %0"	\
96 		: "+A" (v->counter), "=r" (ret)				\
97 		: "r" (I)						\
98 		: "memory");						\
99 	return ret;							\
100 }									\
101 static __always_inline							\
102 c_type arch_atomic##prefix##_fetch_##op(c_type i, atomic##prefix##_t *v)	\
103 {									\
104 	register c_type ret;						\
105 	__asm__ __volatile__ (						\
106 		"	amo" #asm_op "." #asm_type ".aqrl  %1, %2, %0"	\
107 		: "+A" (v->counter), "=r" (ret)				\
108 		: "r" (I)						\
109 		: "memory");						\
110 	return ret;							\
111 }
112 
113 #define ATOMIC_OP_RETURN(op, asm_op, c_op, I, asm_type, c_type, prefix)	\
114 static __always_inline							\
115 c_type arch_atomic##prefix##_##op##_return_relaxed(c_type i,		\
116 					      atomic##prefix##_t *v)	\
117 {									\
118         return arch_atomic##prefix##_fetch_##op##_relaxed(i, v) c_op I;	\
119 }									\
120 static __always_inline							\
121 c_type arch_atomic##prefix##_##op##_return(c_type i, atomic##prefix##_t *v)	\
122 {									\
123         return arch_atomic##prefix##_fetch_##op(i, v) c_op I;		\
124 }
125 
126 #ifdef CONFIG_GENERIC_ATOMIC64
127 #define ATOMIC_OPS(op, asm_op, c_op, I)					\
128         ATOMIC_FETCH_OP( op, asm_op,       I, w, int,   )		\
129         ATOMIC_OP_RETURN(op, asm_op, c_op, I, w, int,   )
130 #else
131 #define ATOMIC_OPS(op, asm_op, c_op, I)					\
132         ATOMIC_FETCH_OP( op, asm_op,       I, w, int,   )		\
133         ATOMIC_OP_RETURN(op, asm_op, c_op, I, w, int,   )		\
134         ATOMIC_FETCH_OP( op, asm_op,       I, d, s64, 64)		\
135         ATOMIC_OP_RETURN(op, asm_op, c_op, I, d, s64, 64)
136 #endif
137 
138 ATOMIC_OPS(add, add, +,  i)
139 ATOMIC_OPS(sub, add, +, -i)
140 
141 #define arch_atomic_add_return_relaxed	arch_atomic_add_return_relaxed
142 #define arch_atomic_sub_return_relaxed	arch_atomic_sub_return_relaxed
143 #define arch_atomic_add_return		arch_atomic_add_return
144 #define arch_atomic_sub_return		arch_atomic_sub_return
145 
146 #define arch_atomic_fetch_add_relaxed	arch_atomic_fetch_add_relaxed
147 #define arch_atomic_fetch_sub_relaxed	arch_atomic_fetch_sub_relaxed
148 #define arch_atomic_fetch_add		arch_atomic_fetch_add
149 #define arch_atomic_fetch_sub		arch_atomic_fetch_sub
150 
151 #ifndef CONFIG_GENERIC_ATOMIC64
152 #define arch_atomic64_add_return_relaxed	arch_atomic64_add_return_relaxed
153 #define arch_atomic64_sub_return_relaxed	arch_atomic64_sub_return_relaxed
154 #define arch_atomic64_add_return		arch_atomic64_add_return
155 #define arch_atomic64_sub_return		arch_atomic64_sub_return
156 
157 #define arch_atomic64_fetch_add_relaxed	arch_atomic64_fetch_add_relaxed
158 #define arch_atomic64_fetch_sub_relaxed	arch_atomic64_fetch_sub_relaxed
159 #define arch_atomic64_fetch_add		arch_atomic64_fetch_add
160 #define arch_atomic64_fetch_sub		arch_atomic64_fetch_sub
161 #endif
162 
163 #undef ATOMIC_OPS
164 
165 #ifdef CONFIG_GENERIC_ATOMIC64
166 #define ATOMIC_OPS(op, asm_op, I)					\
167         ATOMIC_FETCH_OP(op, asm_op, I, w, int,   )
168 #else
169 #define ATOMIC_OPS(op, asm_op, I)					\
170         ATOMIC_FETCH_OP(op, asm_op, I, w, int,   )			\
171         ATOMIC_FETCH_OP(op, asm_op, I, d, s64, 64)
172 #endif
173 
174 ATOMIC_OPS(and, and, i)
175 ATOMIC_OPS( or,  or, i)
176 ATOMIC_OPS(xor, xor, i)
177 
178 #define arch_atomic_fetch_and_relaxed	arch_atomic_fetch_and_relaxed
179 #define arch_atomic_fetch_or_relaxed	arch_atomic_fetch_or_relaxed
180 #define arch_atomic_fetch_xor_relaxed	arch_atomic_fetch_xor_relaxed
181 #define arch_atomic_fetch_and		arch_atomic_fetch_and
182 #define arch_atomic_fetch_or		arch_atomic_fetch_or
183 #define arch_atomic_fetch_xor		arch_atomic_fetch_xor
184 
185 #ifndef CONFIG_GENERIC_ATOMIC64
186 #define arch_atomic64_fetch_and_relaxed	arch_atomic64_fetch_and_relaxed
187 #define arch_atomic64_fetch_or_relaxed	arch_atomic64_fetch_or_relaxed
188 #define arch_atomic64_fetch_xor_relaxed	arch_atomic64_fetch_xor_relaxed
189 #define arch_atomic64_fetch_and		arch_atomic64_fetch_and
190 #define arch_atomic64_fetch_or		arch_atomic64_fetch_or
191 #define arch_atomic64_fetch_xor		arch_atomic64_fetch_xor
192 #endif
193 
194 #undef ATOMIC_OPS
195 
196 #undef ATOMIC_FETCH_OP
197 #undef ATOMIC_OP_RETURN
198 
199 /* This is required to provide a full barrier on success. */
200 static __always_inline int arch_atomic_fetch_add_unless(atomic_t *v, int a, int u)
201 {
202        int prev, rc;
203 
204 	__asm__ __volatile__ (
205 		"0:	lr.w     %[p],  %[c]\n"
206 		"	beq      %[p],  %[u], 1f\n"
207 		"	add      %[rc], %[p], %[a]\n"
208 		"	sc.w.rl  %[rc], %[rc], %[c]\n"
209 		"	bnez     %[rc], 0b\n"
210 		"	fence    rw, rw\n"
211 		"1:\n"
212 		: [p]"=&r" (prev), [rc]"=&r" (rc), [c]"+A" (v->counter)
213 		: [a]"r" (a), [u]"r" (u)
214 		: "memory");
215 	return prev;
216 }
217 #define arch_atomic_fetch_add_unless arch_atomic_fetch_add_unless
218 
219 #ifndef CONFIG_GENERIC_ATOMIC64
arch_atomic64_fetch_add_unless(atomic64_t * v,s64 a,s64 u)220 static __always_inline s64 arch_atomic64_fetch_add_unless(atomic64_t *v, s64 a, s64 u)
221 {
222        s64 prev;
223        long rc;
224 
225 	__asm__ __volatile__ (
226 		"0:	lr.d     %[p],  %[c]\n"
227 		"	beq      %[p],  %[u], 1f\n"
228 		"	add      %[rc], %[p], %[a]\n"
229 		"	sc.d.rl  %[rc], %[rc], %[c]\n"
230 		"	bnez     %[rc], 0b\n"
231 		"	fence    rw, rw\n"
232 		"1:\n"
233 		: [p]"=&r" (prev), [rc]"=&r" (rc), [c]"+A" (v->counter)
234 		: [a]"r" (a), [u]"r" (u)
235 		: "memory");
236 	return prev;
237 }
238 #define arch_atomic64_fetch_add_unless arch_atomic64_fetch_add_unless
239 #endif
240 
241 /*
242  * atomic_{cmp,}xchg is required to have exactly the same ordering semantics as
243  * {cmp,}xchg and the operations that return, so they need a full barrier.
244  */
245 #define ATOMIC_OP(c_t, prefix, size)					\
246 static __always_inline							\
247 c_t arch_atomic##prefix##_xchg_relaxed(atomic##prefix##_t *v, c_t n)	\
248 {									\
249 	return __xchg_relaxed(&(v->counter), n, size);			\
250 }									\
251 static __always_inline							\
252 c_t arch_atomic##prefix##_xchg_acquire(atomic##prefix##_t *v, c_t n)	\
253 {									\
254 	return __xchg_acquire(&(v->counter), n, size);			\
255 }									\
256 static __always_inline							\
257 c_t arch_atomic##prefix##_xchg_release(atomic##prefix##_t *v, c_t n)	\
258 {									\
259 	return __xchg_release(&(v->counter), n, size);			\
260 }									\
261 static __always_inline							\
262 c_t arch_atomic##prefix##_xchg(atomic##prefix##_t *v, c_t n)		\
263 {									\
264 	return __xchg(&(v->counter), n, size);				\
265 }									\
266 static __always_inline							\
267 c_t arch_atomic##prefix##_cmpxchg_relaxed(atomic##prefix##_t *v,	\
268 				     c_t o, c_t n)			\
269 {									\
270 	return __cmpxchg_relaxed(&(v->counter), o, n, size);		\
271 }									\
272 static __always_inline							\
273 c_t arch_atomic##prefix##_cmpxchg_acquire(atomic##prefix##_t *v,	\
274 				     c_t o, c_t n)			\
275 {									\
276 	return __cmpxchg_acquire(&(v->counter), o, n, size);		\
277 }									\
278 static __always_inline							\
279 c_t arch_atomic##prefix##_cmpxchg_release(atomic##prefix##_t *v,	\
280 				     c_t o, c_t n)			\
281 {									\
282 	return __cmpxchg_release(&(v->counter), o, n, size);		\
283 }									\
284 static __always_inline							\
285 c_t arch_atomic##prefix##_cmpxchg(atomic##prefix##_t *v, c_t o, c_t n)	\
286 {									\
287 	return __cmpxchg(&(v->counter), o, n, size);			\
288 }
289 
290 #ifdef CONFIG_GENERIC_ATOMIC64
291 #define ATOMIC_OPS()							\
292 	ATOMIC_OP(int,   , 4)
293 #else
294 #define ATOMIC_OPS()							\
295 	ATOMIC_OP(int,   , 4)						\
296 	ATOMIC_OP(s64, 64, 8)
297 #endif
298 
ATOMIC_OPS()299 ATOMIC_OPS()
300 
301 #define arch_atomic_xchg_relaxed	arch_atomic_xchg_relaxed
302 #define arch_atomic_xchg_acquire	arch_atomic_xchg_acquire
303 #define arch_atomic_xchg_release	arch_atomic_xchg_release
304 #define arch_atomic_xchg		arch_atomic_xchg
305 #define arch_atomic_cmpxchg_relaxed	arch_atomic_cmpxchg_relaxed
306 #define arch_atomic_cmpxchg_acquire	arch_atomic_cmpxchg_acquire
307 #define arch_atomic_cmpxchg_release	arch_atomic_cmpxchg_release
308 #define arch_atomic_cmpxchg		arch_atomic_cmpxchg
309 
310 #undef ATOMIC_OPS
311 #undef ATOMIC_OP
312 
313 static __always_inline bool arch_atomic_inc_unless_negative(atomic_t *v)
314 {
315 	int prev, rc;
316 
317 	__asm__ __volatile__ (
318 		"0:	lr.w      %[p],  %[c]\n"
319 		"	bltz      %[p],  1f\n"
320 		"	addi      %[rc], %[p], 1\n"
321 		"	sc.w.rl   %[rc], %[rc], %[c]\n"
322 		"	bnez      %[rc], 0b\n"
323 		"	fence     rw, rw\n"
324 		"1:\n"
325 		: [p]"=&r" (prev), [rc]"=&r" (rc), [c]"+A" (v->counter)
326 		:
327 		: "memory");
328 	return !(prev < 0);
329 }
330 
331 #define arch_atomic_inc_unless_negative arch_atomic_inc_unless_negative
332 
arch_atomic_dec_unless_positive(atomic_t * v)333 static __always_inline bool arch_atomic_dec_unless_positive(atomic_t *v)
334 {
335 	int prev, rc;
336 
337 	__asm__ __volatile__ (
338 		"0:	lr.w      %[p],  %[c]\n"
339 		"	bgtz      %[p],  1f\n"
340 		"	addi      %[rc], %[p], -1\n"
341 		"	sc.w.rl   %[rc], %[rc], %[c]\n"
342 		"	bnez      %[rc], 0b\n"
343 		"	fence     rw, rw\n"
344 		"1:\n"
345 		: [p]"=&r" (prev), [rc]"=&r" (rc), [c]"+A" (v->counter)
346 		:
347 		: "memory");
348 	return !(prev > 0);
349 }
350 
351 #define arch_atomic_dec_unless_positive arch_atomic_dec_unless_positive
352 
arch_atomic_dec_if_positive(atomic_t * v)353 static __always_inline int arch_atomic_dec_if_positive(atomic_t *v)
354 {
355        int prev, rc;
356 
357 	__asm__ __volatile__ (
358 		"0:	lr.w     %[p],  %[c]\n"
359 		"	addi     %[rc], %[p], -1\n"
360 		"	bltz     %[rc], 1f\n"
361 		"	sc.w.rl  %[rc], %[rc], %[c]\n"
362 		"	bnez     %[rc], 0b\n"
363 		"	fence    rw, rw\n"
364 		"1:\n"
365 		: [p]"=&r" (prev), [rc]"=&r" (rc), [c]"+A" (v->counter)
366 		:
367 		: "memory");
368 	return prev - 1;
369 }
370 
371 #define arch_atomic_dec_if_positive arch_atomic_dec_if_positive
372 
373 #ifndef CONFIG_GENERIC_ATOMIC64
arch_atomic64_inc_unless_negative(atomic64_t * v)374 static __always_inline bool arch_atomic64_inc_unless_negative(atomic64_t *v)
375 {
376 	s64 prev;
377 	long rc;
378 
379 	__asm__ __volatile__ (
380 		"0:	lr.d      %[p],  %[c]\n"
381 		"	bltz      %[p],  1f\n"
382 		"	addi      %[rc], %[p], 1\n"
383 		"	sc.d.rl   %[rc], %[rc], %[c]\n"
384 		"	bnez      %[rc], 0b\n"
385 		"	fence     rw, rw\n"
386 		"1:\n"
387 		: [p]"=&r" (prev), [rc]"=&r" (rc), [c]"+A" (v->counter)
388 		:
389 		: "memory");
390 	return !(prev < 0);
391 }
392 
393 #define arch_atomic64_inc_unless_negative arch_atomic64_inc_unless_negative
394 
arch_atomic64_dec_unless_positive(atomic64_t * v)395 static __always_inline bool arch_atomic64_dec_unless_positive(atomic64_t *v)
396 {
397 	s64 prev;
398 	long rc;
399 
400 	__asm__ __volatile__ (
401 		"0:	lr.d      %[p],  %[c]\n"
402 		"	bgtz      %[p],  1f\n"
403 		"	addi      %[rc], %[p], -1\n"
404 		"	sc.d.rl   %[rc], %[rc], %[c]\n"
405 		"	bnez      %[rc], 0b\n"
406 		"	fence     rw, rw\n"
407 		"1:\n"
408 		: [p]"=&r" (prev), [rc]"=&r" (rc), [c]"+A" (v->counter)
409 		:
410 		: "memory");
411 	return !(prev > 0);
412 }
413 
414 #define arch_atomic64_dec_unless_positive arch_atomic64_dec_unless_positive
415 
arch_atomic64_dec_if_positive(atomic64_t * v)416 static __always_inline s64 arch_atomic64_dec_if_positive(atomic64_t *v)
417 {
418        s64 prev;
419        long rc;
420 
421 	__asm__ __volatile__ (
422 		"0:	lr.d     %[p],  %[c]\n"
423 		"	addi      %[rc], %[p], -1\n"
424 		"	bltz     %[rc], 1f\n"
425 		"	sc.d.rl  %[rc], %[rc], %[c]\n"
426 		"	bnez     %[rc], 0b\n"
427 		"	fence    rw, rw\n"
428 		"1:\n"
429 		: [p]"=&r" (prev), [rc]"=&r" (rc), [c]"+A" (v->counter)
430 		:
431 		: "memory");
432 	return prev - 1;
433 }
434 
435 #define arch_atomic64_dec_if_positive	arch_atomic64_dec_if_positive
436 #endif
437 
438 #endif /* _ASM_RISCV_ATOMIC_H */
439