xref: /linux/arch/arm/include/asm/atomic.h (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  *  arch/arm/include/asm/atomic.h
4  *
5  *  Copyright (C) 1996 Russell King.
6  *  Copyright (C) 2002 Deep Blue Solutions Ltd.
7  */
8 #ifndef __ASM_ARM_ATOMIC_H
9 #define __ASM_ARM_ATOMIC_H
10 
11 #include <linux/compiler.h>
12 #include <linux/prefetch.h>
13 #include <linux/types.h>
14 #include <linux/irqflags.h>
15 #include <asm/barrier.h>
16 #include <asm/cmpxchg.h>
17 
18 #ifdef __KERNEL__
19 
20 /*
21  * On ARM, ordinary assignment (str instruction) doesn't clear the local
22  * strex/ldrex monitor on some implementations. The reason we can use it for
23  * atomic_set() is the clrex or dummy strex done on every exception return.
24  */
25 #define arch_atomic_read(v)	READ_ONCE((v)->counter)
26 #define arch_atomic_set(v,i)	WRITE_ONCE(((v)->counter), (i))
27 
28 #if __LINUX_ARM_ARCH__ >= 6
29 
30 /*
31  * ARMv6 UP and SMP safe atomic ops.  We use load exclusive and
32  * store exclusive to ensure that these are atomic.  We may loop
33  * to ensure that the update happens.
34  */
35 
36 #define ATOMIC_OP(op, c_op, asm_op)					\
37 static inline void arch_atomic_##op(int i, atomic_t *v)			\
38 {									\
39 	unsigned long tmp;						\
40 	int result;							\
41 									\
42 	prefetchw(&v->counter);						\
43 	__asm__ __volatile__("@ atomic_" #op "\n"			\
44 "1:	ldrex	%0, [%3]\n"						\
45 "	" #asm_op "	%0, %0, %4\n"					\
46 "	strex	%1, %0, [%3]\n"						\
47 "	teq	%1, #0\n"						\
48 "	bne	1b"							\
49 	: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter)		\
50 	: "r" (&v->counter), "Ir" (i)					\
51 	: "cc");							\
52 }									\
53 
54 #define ATOMIC_OP_RETURN(op, c_op, asm_op)				\
55 static inline int arch_atomic_##op##_return_relaxed(int i, atomic_t *v)	\
56 {									\
57 	unsigned long tmp;						\
58 	int result;							\
59 									\
60 	prefetchw(&v->counter);						\
61 									\
62 	__asm__ __volatile__("@ atomic_" #op "_return\n"		\
63 "1:	ldrex	%0, [%3]\n"						\
64 "	" #asm_op "	%0, %0, %4\n"					\
65 "	strex	%1, %0, [%3]\n"						\
66 "	teq	%1, #0\n"						\
67 "	bne	1b"							\
68 	: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter)		\
69 	: "r" (&v->counter), "Ir" (i)					\
70 	: "cc");							\
71 									\
72 	return result;							\
73 }
74 
75 #define ATOMIC_FETCH_OP(op, c_op, asm_op)				\
76 static inline int arch_atomic_fetch_##op##_relaxed(int i, atomic_t *v)	\
77 {									\
78 	unsigned long tmp;						\
79 	int result, val;						\
80 									\
81 	prefetchw(&v->counter);						\
82 									\
83 	__asm__ __volatile__("@ atomic_fetch_" #op "\n"			\
84 "1:	ldrex	%0, [%4]\n"						\
85 "	" #asm_op "	%1, %0, %5\n"					\
86 "	strex	%2, %1, [%4]\n"						\
87 "	teq	%2, #0\n"						\
88 "	bne	1b"							\
89 	: "=&r" (result), "=&r" (val), "=&r" (tmp), "+Qo" (v->counter)	\
90 	: "r" (&v->counter), "Ir" (i)					\
91 	: "cc");							\
92 									\
93 	return result;							\
94 }
95 
96 #define arch_atomic_add_return_relaxed		arch_atomic_add_return_relaxed
97 #define arch_atomic_sub_return_relaxed		arch_atomic_sub_return_relaxed
98 #define arch_atomic_fetch_add_relaxed		arch_atomic_fetch_add_relaxed
99 #define arch_atomic_fetch_sub_relaxed		arch_atomic_fetch_sub_relaxed
100 
101 #define arch_atomic_fetch_and_relaxed		arch_atomic_fetch_and_relaxed
102 #define arch_atomic_fetch_andnot_relaxed	arch_atomic_fetch_andnot_relaxed
103 #define arch_atomic_fetch_or_relaxed		arch_atomic_fetch_or_relaxed
104 #define arch_atomic_fetch_xor_relaxed		arch_atomic_fetch_xor_relaxed
105 
106 static inline int arch_atomic_cmpxchg_relaxed(atomic_t *ptr, int old, int new)
107 {
108 	int oldval;
109 	unsigned long res;
110 
111 	prefetchw(&ptr->counter);
112 
113 	do {
114 		__asm__ __volatile__("@ atomic_cmpxchg\n"
115 		"ldrex	%1, [%3]\n"
116 		"mov	%0, #0\n"
117 		"teq	%1, %4\n"
118 		"strexeq %0, %5, [%3]\n"
119 		    : "=&r" (res), "=&r" (oldval), "+Qo" (ptr->counter)
120 		    : "r" (&ptr->counter), "Ir" (old), "r" (new)
121 		    : "cc");
122 	} while (res);
123 
124 	return oldval;
125 }
126 #define arch_atomic_cmpxchg_relaxed		arch_atomic_cmpxchg_relaxed
127 
128 static inline int arch_atomic_fetch_add_unless(atomic_t *v, int a, int u)
129 {
130 	int oldval, newval;
131 	unsigned long tmp;
132 
133 	smp_mb();
134 	prefetchw(&v->counter);
135 
136 	__asm__ __volatile__ ("@ atomic_add_unless\n"
137 "1:	ldrex	%0, [%4]\n"
138 "	teq	%0, %5\n"
139 "	beq	2f\n"
140 "	add	%1, %0, %6\n"
141 "	strex	%2, %1, [%4]\n"
142 "	teq	%2, #0\n"
143 "	bne	1b\n"
144 "2:"
145 	: "=&r" (oldval), "=&r" (newval), "=&r" (tmp), "+Qo" (v->counter)
146 	: "r" (&v->counter), "r" (u), "r" (a)
147 	: "cc");
148 
149 	if (oldval != u)
150 		smp_mb();
151 
152 	return oldval;
153 }
154 #define arch_atomic_fetch_add_unless		arch_atomic_fetch_add_unless
155 
156 #else /* ARM_ARCH_6 */
157 
158 #ifdef CONFIG_SMP
159 #error SMP not supported on pre-ARMv6 CPUs
160 #endif
161 
162 #define ATOMIC_OP(op, c_op, asm_op)					\
163 static inline void arch_atomic_##op(int i, atomic_t *v)			\
164 {									\
165 	unsigned long flags;						\
166 									\
167 	raw_local_irq_save(flags);					\
168 	v->counter c_op i;						\
169 	raw_local_irq_restore(flags);					\
170 }									\
171 
172 #define ATOMIC_OP_RETURN(op, c_op, asm_op)				\
173 static inline int arch_atomic_##op##_return(int i, atomic_t *v)		\
174 {									\
175 	unsigned long flags;						\
176 	int val;							\
177 									\
178 	raw_local_irq_save(flags);					\
179 	v->counter c_op i;						\
180 	val = v->counter;						\
181 	raw_local_irq_restore(flags);					\
182 									\
183 	return val;							\
184 }
185 
186 #define ATOMIC_FETCH_OP(op, c_op, asm_op)				\
187 static inline int arch_atomic_fetch_##op(int i, atomic_t *v)		\
188 {									\
189 	unsigned long flags;						\
190 	int val;							\
191 									\
192 	raw_local_irq_save(flags);					\
193 	val = v->counter;						\
194 	v->counter c_op i;						\
195 	raw_local_irq_restore(flags);					\
196 									\
197 	return val;							\
198 }
199 
200 #define arch_atomic_add_return			arch_atomic_add_return
201 #define arch_atomic_sub_return			arch_atomic_sub_return
202 #define arch_atomic_fetch_add			arch_atomic_fetch_add
203 #define arch_atomic_fetch_sub			arch_atomic_fetch_sub
204 
205 #define arch_atomic_fetch_and			arch_atomic_fetch_and
206 #define arch_atomic_fetch_andnot		arch_atomic_fetch_andnot
207 #define arch_atomic_fetch_or			arch_atomic_fetch_or
208 #define arch_atomic_fetch_xor			arch_atomic_fetch_xor
209 
210 static inline int arch_atomic_cmpxchg(atomic_t *v, int old, int new)
211 {
212 	int ret;
213 	unsigned long flags;
214 
215 	raw_local_irq_save(flags);
216 	ret = v->counter;
217 	if (likely(ret == old))
218 		v->counter = new;
219 	raw_local_irq_restore(flags);
220 
221 	return ret;
222 }
223 #define arch_atomic_cmpxchg arch_atomic_cmpxchg
224 
225 #endif /* __LINUX_ARM_ARCH__ */
226 
227 #define ATOMIC_OPS(op, c_op, asm_op)					\
228 	ATOMIC_OP(op, c_op, asm_op)					\
229 	ATOMIC_OP_RETURN(op, c_op, asm_op)				\
230 	ATOMIC_FETCH_OP(op, c_op, asm_op)
231 
232 ATOMIC_OPS(add, +=, add)
233 ATOMIC_OPS(sub, -=, sub)
234 
235 #define arch_atomic_andnot arch_atomic_andnot
236 
237 #undef ATOMIC_OPS
238 #define ATOMIC_OPS(op, c_op, asm_op)					\
239 	ATOMIC_OP(op, c_op, asm_op)					\
240 	ATOMIC_FETCH_OP(op, c_op, asm_op)
241 
242 ATOMIC_OPS(and, &=, and)
243 ATOMIC_OPS(andnot, &= ~, bic)
244 ATOMIC_OPS(or,  |=, orr)
245 ATOMIC_OPS(xor, ^=, eor)
246 
247 #undef ATOMIC_OPS
248 #undef ATOMIC_FETCH_OP
249 #undef ATOMIC_OP_RETURN
250 #undef ATOMIC_OP
251 
252 #ifndef CONFIG_GENERIC_ATOMIC64
253 typedef struct {
254 	s64 counter;
255 } atomic64_t;
256 
257 #define ATOMIC64_INIT(i) { (i) }
258 
259 #ifdef CONFIG_ARM_LPAE
260 static inline s64 arch_atomic64_read(const atomic64_t *v)
261 {
262 	s64 result;
263 
264 	__asm__ __volatile__("@ atomic64_read\n"
265 "	ldrd	%0, %H0, [%1]"
266 	: "=&r" (result)
267 	: "r" (&v->counter), "Qo" (v->counter)
268 	);
269 
270 	return result;
271 }
272 
273 static inline void arch_atomic64_set(atomic64_t *v, s64 i)
274 {
275 	__asm__ __volatile__("@ atomic64_set\n"
276 "	strd	%2, %H2, [%1]"
277 	: "=Qo" (v->counter)
278 	: "r" (&v->counter), "r" (i)
279 	);
280 }
281 #else
282 static inline s64 arch_atomic64_read(const atomic64_t *v)
283 {
284 	s64 result;
285 
286 	__asm__ __volatile__("@ atomic64_read\n"
287 "	ldrexd	%0, %H0, [%1]"
288 	: "=&r" (result)
289 	: "r" (&v->counter), "Qo" (v->counter)
290 	);
291 
292 	return result;
293 }
294 
295 static inline void arch_atomic64_set(atomic64_t *v, s64 i)
296 {
297 	s64 tmp;
298 
299 	prefetchw(&v->counter);
300 	__asm__ __volatile__("@ atomic64_set\n"
301 "1:	ldrexd	%0, %H0, [%2]\n"
302 "	strexd	%0, %3, %H3, [%2]\n"
303 "	teq	%0, #0\n"
304 "	bne	1b"
305 	: "=&r" (tmp), "=Qo" (v->counter)
306 	: "r" (&v->counter), "r" (i)
307 	: "cc");
308 }
309 #endif
310 
311 #define ATOMIC64_OP(op, op1, op2)					\
312 static inline void arch_atomic64_##op(s64 i, atomic64_t *v)		\
313 {									\
314 	s64 result;							\
315 	unsigned long tmp;						\
316 									\
317 	prefetchw(&v->counter);						\
318 	__asm__ __volatile__("@ atomic64_" #op "\n"			\
319 "1:	ldrexd	%0, %H0, [%3]\n"					\
320 "	" #op1 " %Q0, %Q0, %Q4\n"					\
321 "	" #op2 " %R0, %R0, %R4\n"					\
322 "	strexd	%1, %0, %H0, [%3]\n"					\
323 "	teq	%1, #0\n"						\
324 "	bne	1b"							\
325 	: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter)		\
326 	: "r" (&v->counter), "r" (i)					\
327 	: "cc");							\
328 }									\
329 
330 #define ATOMIC64_OP_RETURN(op, op1, op2)				\
331 static inline s64							\
332 arch_atomic64_##op##_return_relaxed(s64 i, atomic64_t *v)		\
333 {									\
334 	s64 result;							\
335 	unsigned long tmp;						\
336 									\
337 	prefetchw(&v->counter);						\
338 									\
339 	__asm__ __volatile__("@ atomic64_" #op "_return\n"		\
340 "1:	ldrexd	%0, %H0, [%3]\n"					\
341 "	" #op1 " %Q0, %Q0, %Q4\n"					\
342 "	" #op2 " %R0, %R0, %R4\n"					\
343 "	strexd	%1, %0, %H0, [%3]\n"					\
344 "	teq	%1, #0\n"						\
345 "	bne	1b"							\
346 	: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter)		\
347 	: "r" (&v->counter), "r" (i)					\
348 	: "cc");							\
349 									\
350 	return result;							\
351 }
352 
353 #define ATOMIC64_FETCH_OP(op, op1, op2)					\
354 static inline s64							\
355 arch_atomic64_fetch_##op##_relaxed(s64 i, atomic64_t *v)		\
356 {									\
357 	s64 result, val;						\
358 	unsigned long tmp;						\
359 									\
360 	prefetchw(&v->counter);						\
361 									\
362 	__asm__ __volatile__("@ atomic64_fetch_" #op "\n"		\
363 "1:	ldrexd	%0, %H0, [%4]\n"					\
364 "	" #op1 " %Q1, %Q0, %Q5\n"					\
365 "	" #op2 " %R1, %R0, %R5\n"					\
366 "	strexd	%2, %1, %H1, [%4]\n"					\
367 "	teq	%2, #0\n"						\
368 "	bne	1b"							\
369 	: "=&r" (result), "=&r" (val), "=&r" (tmp), "+Qo" (v->counter)	\
370 	: "r" (&v->counter), "r" (i)					\
371 	: "cc");							\
372 									\
373 	return result;							\
374 }
375 
376 #define ATOMIC64_OPS(op, op1, op2)					\
377 	ATOMIC64_OP(op, op1, op2)					\
378 	ATOMIC64_OP_RETURN(op, op1, op2)				\
379 	ATOMIC64_FETCH_OP(op, op1, op2)
380 
381 ATOMIC64_OPS(add, adds, adc)
382 ATOMIC64_OPS(sub, subs, sbc)
383 
384 #define arch_atomic64_add_return_relaxed	arch_atomic64_add_return_relaxed
385 #define arch_atomic64_sub_return_relaxed	arch_atomic64_sub_return_relaxed
386 #define arch_atomic64_fetch_add_relaxed		arch_atomic64_fetch_add_relaxed
387 #define arch_atomic64_fetch_sub_relaxed		arch_atomic64_fetch_sub_relaxed
388 
389 #undef ATOMIC64_OPS
390 #define ATOMIC64_OPS(op, op1, op2)					\
391 	ATOMIC64_OP(op, op1, op2)					\
392 	ATOMIC64_FETCH_OP(op, op1, op2)
393 
394 #define arch_atomic64_andnot arch_atomic64_andnot
395 
396 ATOMIC64_OPS(and, and, and)
397 ATOMIC64_OPS(andnot, bic, bic)
398 ATOMIC64_OPS(or,  orr, orr)
399 ATOMIC64_OPS(xor, eor, eor)
400 
401 #define arch_atomic64_fetch_and_relaxed		arch_atomic64_fetch_and_relaxed
402 #define arch_atomic64_fetch_andnot_relaxed	arch_atomic64_fetch_andnot_relaxed
403 #define arch_atomic64_fetch_or_relaxed		arch_atomic64_fetch_or_relaxed
404 #define arch_atomic64_fetch_xor_relaxed		arch_atomic64_fetch_xor_relaxed
405 
406 #undef ATOMIC64_OPS
407 #undef ATOMIC64_FETCH_OP
408 #undef ATOMIC64_OP_RETURN
409 #undef ATOMIC64_OP
410 
411 static inline s64 arch_atomic64_cmpxchg_relaxed(atomic64_t *ptr, s64 old, s64 new)
412 {
413 	s64 oldval;
414 	unsigned long res;
415 
416 	prefetchw(&ptr->counter);
417 
418 	do {
419 		__asm__ __volatile__("@ atomic64_cmpxchg\n"
420 		"ldrexd		%1, %H1, [%3]\n"
421 		"mov		%0, #0\n"
422 		"teq		%1, %4\n"
423 		"teqeq		%H1, %H4\n"
424 		"strexdeq	%0, %5, %H5, [%3]"
425 		: "=&r" (res), "=&r" (oldval), "+Qo" (ptr->counter)
426 		: "r" (&ptr->counter), "r" (old), "r" (new)
427 		: "cc");
428 	} while (res);
429 
430 	return oldval;
431 }
432 #define arch_atomic64_cmpxchg_relaxed	arch_atomic64_cmpxchg_relaxed
433 
434 static inline s64 arch_atomic64_xchg_relaxed(atomic64_t *ptr, s64 new)
435 {
436 	s64 result;
437 	unsigned long tmp;
438 
439 	prefetchw(&ptr->counter);
440 
441 	__asm__ __volatile__("@ atomic64_xchg\n"
442 "1:	ldrexd	%0, %H0, [%3]\n"
443 "	strexd	%1, %4, %H4, [%3]\n"
444 "	teq	%1, #0\n"
445 "	bne	1b"
446 	: "=&r" (result), "=&r" (tmp), "+Qo" (ptr->counter)
447 	: "r" (&ptr->counter), "r" (new)
448 	: "cc");
449 
450 	return result;
451 }
452 #define arch_atomic64_xchg_relaxed		arch_atomic64_xchg_relaxed
453 
454 static inline s64 arch_atomic64_dec_if_positive(atomic64_t *v)
455 {
456 	s64 result;
457 	unsigned long tmp;
458 
459 	smp_mb();
460 	prefetchw(&v->counter);
461 
462 	__asm__ __volatile__("@ atomic64_dec_if_positive\n"
463 "1:	ldrexd	%0, %H0, [%3]\n"
464 "	subs	%Q0, %Q0, #1\n"
465 "	sbc	%R0, %R0, #0\n"
466 "	teq	%R0, #0\n"
467 "	bmi	2f\n"
468 "	strexd	%1, %0, %H0, [%3]\n"
469 "	teq	%1, #0\n"
470 "	bne	1b\n"
471 "2:"
472 	: "=&r" (result), "=&r" (tmp), "+Qo" (v->counter)
473 	: "r" (&v->counter)
474 	: "cc");
475 
476 	smp_mb();
477 
478 	return result;
479 }
480 #define arch_atomic64_dec_if_positive arch_atomic64_dec_if_positive
481 
482 static inline s64 arch_atomic64_fetch_add_unless(atomic64_t *v, s64 a, s64 u)
483 {
484 	s64 oldval, newval;
485 	unsigned long tmp;
486 
487 	smp_mb();
488 	prefetchw(&v->counter);
489 
490 	__asm__ __volatile__("@ atomic64_add_unless\n"
491 "1:	ldrexd	%0, %H0, [%4]\n"
492 "	teq	%0, %5\n"
493 "	teqeq	%H0, %H5\n"
494 "	beq	2f\n"
495 "	adds	%Q1, %Q0, %Q6\n"
496 "	adc	%R1, %R0, %R6\n"
497 "	strexd	%2, %1, %H1, [%4]\n"
498 "	teq	%2, #0\n"
499 "	bne	1b\n"
500 "2:"
501 	: "=&r" (oldval), "=&r" (newval), "=&r" (tmp), "+Qo" (v->counter)
502 	: "r" (&v->counter), "r" (u), "r" (a)
503 	: "cc");
504 
505 	if (oldval != u)
506 		smp_mb();
507 
508 	return oldval;
509 }
510 #define arch_atomic64_fetch_add_unless arch_atomic64_fetch_add_unless
511 
512 #endif /* !CONFIG_GENERIC_ATOMIC64 */
513 #endif
514 #endif
515