xref: /openbmc/linux/arch/powerpc/include/asm/atomic.h (revision 63705da3)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_POWERPC_ATOMIC_H_
3 #define _ASM_POWERPC_ATOMIC_H_
4 
5 /*
6  * PowerPC atomic operations
7  */
8 
9 #ifdef __KERNEL__
10 #include <linux/types.h>
11 #include <asm/cmpxchg.h>
12 #include <asm/barrier.h>
13 #include <asm/asm-const.h>
14 
15 /*
16  * Since *_return_relaxed and {cmp}xchg_relaxed are implemented with
17  * a "bne-" instruction at the end, so an isync is enough as a acquire barrier
18  * on the platform without lwsync.
19  */
20 #define __atomic_acquire_fence()					\
21 	__asm__ __volatile__(PPC_ACQUIRE_BARRIER "" : : : "memory")
22 
23 #define __atomic_release_fence()					\
24 	__asm__ __volatile__(PPC_RELEASE_BARRIER "" : : : "memory")
25 
26 static __inline__ int arch_atomic_read(const atomic_t *v)
27 {
28 	int t;
29 
30 	__asm__ __volatile__("lwz%U1%X1 %0,%1" : "=r"(t) : "m<>"(v->counter));
31 
32 	return t;
33 }
34 
35 static __inline__ void arch_atomic_set(atomic_t *v, int i)
36 {
37 	__asm__ __volatile__("stw%U0%X0 %1,%0" : "=m<>"(v->counter) : "r"(i));
38 }
39 
40 #define ATOMIC_OP(op, asm_op)						\
41 static __inline__ void arch_atomic_##op(int a, atomic_t *v)		\
42 {									\
43 	int t;								\
44 									\
45 	__asm__ __volatile__(						\
46 "1:	lwarx	%0,0,%3		# atomic_" #op "\n"			\
47 	#asm_op " %0,%2,%0\n"						\
48 "	stwcx.	%0,0,%3 \n"						\
49 "	bne-	1b\n"							\
50 	: "=&r" (t), "+m" (v->counter)					\
51 	: "r" (a), "r" (&v->counter)					\
52 	: "cc");							\
53 }									\
54 
55 #define ATOMIC_OP_RETURN_RELAXED(op, asm_op)				\
56 static inline int arch_atomic_##op##_return_relaxed(int a, atomic_t *v)	\
57 {									\
58 	int t;								\
59 									\
60 	__asm__ __volatile__(						\
61 "1:	lwarx	%0,0,%3		# atomic_" #op "_return_relaxed\n"	\
62 	#asm_op " %0,%2,%0\n"						\
63 "	stwcx.	%0,0,%3\n"						\
64 "	bne-	1b\n"							\
65 	: "=&r" (t), "+m" (v->counter)					\
66 	: "r" (a), "r" (&v->counter)					\
67 	: "cc");							\
68 									\
69 	return t;							\
70 }
71 
72 #define ATOMIC_FETCH_OP_RELAXED(op, asm_op)				\
73 static inline int arch_atomic_fetch_##op##_relaxed(int a, atomic_t *v)	\
74 {									\
75 	int res, t;							\
76 									\
77 	__asm__ __volatile__(						\
78 "1:	lwarx	%0,0,%4		# atomic_fetch_" #op "_relaxed\n"	\
79 	#asm_op " %1,%3,%0\n"						\
80 "	stwcx.	%1,0,%4\n"						\
81 "	bne-	1b\n"							\
82 	: "=&r" (res), "=&r" (t), "+m" (v->counter)			\
83 	: "r" (a), "r" (&v->counter)					\
84 	: "cc");							\
85 									\
86 	return res;							\
87 }
88 
89 #define ATOMIC_OPS(op, asm_op)						\
90 	ATOMIC_OP(op, asm_op)						\
91 	ATOMIC_OP_RETURN_RELAXED(op, asm_op)				\
92 	ATOMIC_FETCH_OP_RELAXED(op, asm_op)
93 
94 ATOMIC_OPS(add, add)
95 ATOMIC_OPS(sub, subf)
96 
97 #define arch_atomic_add_return_relaxed arch_atomic_add_return_relaxed
98 #define arch_atomic_sub_return_relaxed arch_atomic_sub_return_relaxed
99 
100 #define arch_atomic_fetch_add_relaxed arch_atomic_fetch_add_relaxed
101 #define arch_atomic_fetch_sub_relaxed arch_atomic_fetch_sub_relaxed
102 
103 #undef ATOMIC_OPS
104 #define ATOMIC_OPS(op, asm_op)						\
105 	ATOMIC_OP(op, asm_op)						\
106 	ATOMIC_FETCH_OP_RELAXED(op, asm_op)
107 
108 ATOMIC_OPS(and, and)
109 ATOMIC_OPS(or, or)
110 ATOMIC_OPS(xor, xor)
111 
112 #define arch_atomic_fetch_and_relaxed arch_atomic_fetch_and_relaxed
113 #define arch_atomic_fetch_or_relaxed  arch_atomic_fetch_or_relaxed
114 #define arch_atomic_fetch_xor_relaxed arch_atomic_fetch_xor_relaxed
115 
116 #undef ATOMIC_OPS
117 #undef ATOMIC_FETCH_OP_RELAXED
118 #undef ATOMIC_OP_RETURN_RELAXED
119 #undef ATOMIC_OP
120 
121 static __inline__ void arch_atomic_inc(atomic_t *v)
122 {
123 	int t;
124 
125 	__asm__ __volatile__(
126 "1:	lwarx	%0,0,%2		# atomic_inc\n\
127 	addic	%0,%0,1\n"
128 "	stwcx.	%0,0,%2 \n\
129 	bne-	1b"
130 	: "=&r" (t), "+m" (v->counter)
131 	: "r" (&v->counter)
132 	: "cc", "xer");
133 }
134 #define arch_atomic_inc arch_atomic_inc
135 
136 static __inline__ int arch_atomic_inc_return_relaxed(atomic_t *v)
137 {
138 	int t;
139 
140 	__asm__ __volatile__(
141 "1:	lwarx	%0,0,%2		# atomic_inc_return_relaxed\n"
142 "	addic	%0,%0,1\n"
143 "	stwcx.	%0,0,%2\n"
144 "	bne-	1b"
145 	: "=&r" (t), "+m" (v->counter)
146 	: "r" (&v->counter)
147 	: "cc", "xer");
148 
149 	return t;
150 }
151 
152 static __inline__ void arch_atomic_dec(atomic_t *v)
153 {
154 	int t;
155 
156 	__asm__ __volatile__(
157 "1:	lwarx	%0,0,%2		# atomic_dec\n\
158 	addic	%0,%0,-1\n"
159 "	stwcx.	%0,0,%2\n\
160 	bne-	1b"
161 	: "=&r" (t), "+m" (v->counter)
162 	: "r" (&v->counter)
163 	: "cc", "xer");
164 }
165 #define arch_atomic_dec arch_atomic_dec
166 
167 static __inline__ int arch_atomic_dec_return_relaxed(atomic_t *v)
168 {
169 	int t;
170 
171 	__asm__ __volatile__(
172 "1:	lwarx	%0,0,%2		# atomic_dec_return_relaxed\n"
173 "	addic	%0,%0,-1\n"
174 "	stwcx.	%0,0,%2\n"
175 "	bne-	1b"
176 	: "=&r" (t), "+m" (v->counter)
177 	: "r" (&v->counter)
178 	: "cc", "xer");
179 
180 	return t;
181 }
182 
183 #define arch_atomic_inc_return_relaxed arch_atomic_inc_return_relaxed
184 #define arch_atomic_dec_return_relaxed arch_atomic_dec_return_relaxed
185 
186 #define arch_atomic_cmpxchg(v, o, n) \
187 	(arch_cmpxchg(&((v)->counter), (o), (n)))
188 #define arch_atomic_cmpxchg_relaxed(v, o, n) \
189 	arch_cmpxchg_relaxed(&((v)->counter), (o), (n))
190 #define arch_atomic_cmpxchg_acquire(v, o, n) \
191 	arch_cmpxchg_acquire(&((v)->counter), (o), (n))
192 
193 #define arch_atomic_xchg(v, new) \
194 	(arch_xchg(&((v)->counter), new))
195 #define arch_atomic_xchg_relaxed(v, new) \
196 	arch_xchg_relaxed(&((v)->counter), (new))
197 
198 /*
199  * Don't want to override the generic atomic_try_cmpxchg_acquire, because
200  * we add a lock hint to the lwarx, which may not be wanted for the
201  * _acquire case (and is not used by the other _acquire variants so it
202  * would be a surprise).
203  */
204 static __always_inline bool
205 arch_atomic_try_cmpxchg_lock(atomic_t *v, int *old, int new)
206 {
207 	int r, o = *old;
208 
209 	__asm__ __volatile__ (
210 "1:	lwarx	%0,0,%2,%5	# atomic_try_cmpxchg_acquire		\n"
211 "	cmpw	0,%0,%3							\n"
212 "	bne-	2f							\n"
213 "	stwcx.	%4,0,%2							\n"
214 "	bne-	1b							\n"
215 "\t"	PPC_ACQUIRE_BARRIER "						\n"
216 "2:									\n"
217 	: "=&r" (r), "+m" (v->counter)
218 	: "r" (&v->counter), "r" (o), "r" (new), "i" (IS_ENABLED(CONFIG_PPC64) ? 1 : 0)
219 	: "cr0", "memory");
220 
221 	if (unlikely(r != o))
222 		*old = r;
223 	return likely(r == o);
224 }
225 
226 /**
227  * atomic_fetch_add_unless - add unless the number is a given value
228  * @v: pointer of type atomic_t
229  * @a: the amount to add to v...
230  * @u: ...unless v is equal to u.
231  *
232  * Atomically adds @a to @v, so long as it was not @u.
233  * Returns the old value of @v.
234  */
235 static __inline__ int arch_atomic_fetch_add_unless(atomic_t *v, int a, int u)
236 {
237 	int t;
238 
239 	__asm__ __volatile__ (
240 	PPC_ATOMIC_ENTRY_BARRIER
241 "1:	lwarx	%0,0,%1		# atomic_fetch_add_unless\n\
242 	cmpw	0,%0,%3 \n\
243 	beq	2f \n\
244 	add	%0,%2,%0 \n"
245 "	stwcx.	%0,0,%1 \n\
246 	bne-	1b \n"
247 	PPC_ATOMIC_EXIT_BARRIER
248 "	subf	%0,%2,%0 \n\
249 2:"
250 	: "=&r" (t)
251 	: "r" (&v->counter), "r" (a), "r" (u)
252 	: "cc", "memory");
253 
254 	return t;
255 }
256 #define arch_atomic_fetch_add_unless arch_atomic_fetch_add_unless
257 
258 /**
259  * atomic_inc_not_zero - increment unless the number is zero
260  * @v: pointer of type atomic_t
261  *
262  * Atomically increments @v by 1, so long as @v is non-zero.
263  * Returns non-zero if @v was non-zero, and zero otherwise.
264  */
265 static __inline__ int arch_atomic_inc_not_zero(atomic_t *v)
266 {
267 	int t1, t2;
268 
269 	__asm__ __volatile__ (
270 	PPC_ATOMIC_ENTRY_BARRIER
271 "1:	lwarx	%0,0,%2		# atomic_inc_not_zero\n\
272 	cmpwi	0,%0,0\n\
273 	beq-	2f\n\
274 	addic	%1,%0,1\n"
275 "	stwcx.	%1,0,%2\n\
276 	bne-	1b\n"
277 	PPC_ATOMIC_EXIT_BARRIER
278 	"\n\
279 2:"
280 	: "=&r" (t1), "=&r" (t2)
281 	: "r" (&v->counter)
282 	: "cc", "xer", "memory");
283 
284 	return t1;
285 }
286 #define arch_atomic_inc_not_zero(v) arch_atomic_inc_not_zero((v))
287 
288 /*
289  * Atomically test *v and decrement if it is greater than 0.
290  * The function returns the old value of *v minus 1, even if
291  * the atomic variable, v, was not decremented.
292  */
293 static __inline__ int arch_atomic_dec_if_positive(atomic_t *v)
294 {
295 	int t;
296 
297 	__asm__ __volatile__(
298 	PPC_ATOMIC_ENTRY_BARRIER
299 "1:	lwarx	%0,0,%1		# atomic_dec_if_positive\n\
300 	cmpwi	%0,1\n\
301 	addi	%0,%0,-1\n\
302 	blt-	2f\n"
303 "	stwcx.	%0,0,%1\n\
304 	bne-	1b"
305 	PPC_ATOMIC_EXIT_BARRIER
306 	"\n\
307 2:"	: "=&b" (t)
308 	: "r" (&v->counter)
309 	: "cc", "memory");
310 
311 	return t;
312 }
313 #define arch_atomic_dec_if_positive arch_atomic_dec_if_positive
314 
315 #ifdef __powerpc64__
316 
317 #define ATOMIC64_INIT(i)	{ (i) }
318 
319 static __inline__ s64 arch_atomic64_read(const atomic64_t *v)
320 {
321 	s64 t;
322 
323 	__asm__ __volatile__("ld%U1%X1 %0,%1" : "=r"(t) : "m<>"(v->counter));
324 
325 	return t;
326 }
327 
328 static __inline__ void arch_atomic64_set(atomic64_t *v, s64 i)
329 {
330 	__asm__ __volatile__("std%U0%X0 %1,%0" : "=m<>"(v->counter) : "r"(i));
331 }
332 
333 #define ATOMIC64_OP(op, asm_op)						\
334 static __inline__ void arch_atomic64_##op(s64 a, atomic64_t *v)		\
335 {									\
336 	s64 t;								\
337 									\
338 	__asm__ __volatile__(						\
339 "1:	ldarx	%0,0,%3		# atomic64_" #op "\n"			\
340 	#asm_op " %0,%2,%0\n"						\
341 "	stdcx.	%0,0,%3 \n"						\
342 "	bne-	1b\n"							\
343 	: "=&r" (t), "+m" (v->counter)					\
344 	: "r" (a), "r" (&v->counter)					\
345 	: "cc");							\
346 }
347 
348 #define ATOMIC64_OP_RETURN_RELAXED(op, asm_op)				\
349 static inline s64							\
350 arch_atomic64_##op##_return_relaxed(s64 a, atomic64_t *v)		\
351 {									\
352 	s64 t;								\
353 									\
354 	__asm__ __volatile__(						\
355 "1:	ldarx	%0,0,%3		# atomic64_" #op "_return_relaxed\n"	\
356 	#asm_op " %0,%2,%0\n"						\
357 "	stdcx.	%0,0,%3\n"						\
358 "	bne-	1b\n"							\
359 	: "=&r" (t), "+m" (v->counter)					\
360 	: "r" (a), "r" (&v->counter)					\
361 	: "cc");							\
362 									\
363 	return t;							\
364 }
365 
366 #define ATOMIC64_FETCH_OP_RELAXED(op, asm_op)				\
367 static inline s64							\
368 arch_atomic64_fetch_##op##_relaxed(s64 a, atomic64_t *v)		\
369 {									\
370 	s64 res, t;							\
371 									\
372 	__asm__ __volatile__(						\
373 "1:	ldarx	%0,0,%4		# atomic64_fetch_" #op "_relaxed\n"	\
374 	#asm_op " %1,%3,%0\n"						\
375 "	stdcx.	%1,0,%4\n"						\
376 "	bne-	1b\n"							\
377 	: "=&r" (res), "=&r" (t), "+m" (v->counter)			\
378 	: "r" (a), "r" (&v->counter)					\
379 	: "cc");							\
380 									\
381 	return res;							\
382 }
383 
384 #define ATOMIC64_OPS(op, asm_op)					\
385 	ATOMIC64_OP(op, asm_op)						\
386 	ATOMIC64_OP_RETURN_RELAXED(op, asm_op)				\
387 	ATOMIC64_FETCH_OP_RELAXED(op, asm_op)
388 
389 ATOMIC64_OPS(add, add)
390 ATOMIC64_OPS(sub, subf)
391 
392 #define arch_atomic64_add_return_relaxed arch_atomic64_add_return_relaxed
393 #define arch_atomic64_sub_return_relaxed arch_atomic64_sub_return_relaxed
394 
395 #define arch_atomic64_fetch_add_relaxed arch_atomic64_fetch_add_relaxed
396 #define arch_atomic64_fetch_sub_relaxed arch_atomic64_fetch_sub_relaxed
397 
398 #undef ATOMIC64_OPS
399 #define ATOMIC64_OPS(op, asm_op)					\
400 	ATOMIC64_OP(op, asm_op)						\
401 	ATOMIC64_FETCH_OP_RELAXED(op, asm_op)
402 
403 ATOMIC64_OPS(and, and)
404 ATOMIC64_OPS(or, or)
405 ATOMIC64_OPS(xor, xor)
406 
407 #define arch_atomic64_fetch_and_relaxed arch_atomic64_fetch_and_relaxed
408 #define arch_atomic64_fetch_or_relaxed  arch_atomic64_fetch_or_relaxed
409 #define arch_atomic64_fetch_xor_relaxed arch_atomic64_fetch_xor_relaxed
410 
411 #undef ATOPIC64_OPS
412 #undef ATOMIC64_FETCH_OP_RELAXED
413 #undef ATOMIC64_OP_RETURN_RELAXED
414 #undef ATOMIC64_OP
415 
416 static __inline__ void arch_atomic64_inc(atomic64_t *v)
417 {
418 	s64 t;
419 
420 	__asm__ __volatile__(
421 "1:	ldarx	%0,0,%2		# atomic64_inc\n\
422 	addic	%0,%0,1\n\
423 	stdcx.	%0,0,%2 \n\
424 	bne-	1b"
425 	: "=&r" (t), "+m" (v->counter)
426 	: "r" (&v->counter)
427 	: "cc", "xer");
428 }
429 #define arch_atomic64_inc arch_atomic64_inc
430 
431 static __inline__ s64 arch_atomic64_inc_return_relaxed(atomic64_t *v)
432 {
433 	s64 t;
434 
435 	__asm__ __volatile__(
436 "1:	ldarx	%0,0,%2		# atomic64_inc_return_relaxed\n"
437 "	addic	%0,%0,1\n"
438 "	stdcx.	%0,0,%2\n"
439 "	bne-	1b"
440 	: "=&r" (t), "+m" (v->counter)
441 	: "r" (&v->counter)
442 	: "cc", "xer");
443 
444 	return t;
445 }
446 
447 static __inline__ void arch_atomic64_dec(atomic64_t *v)
448 {
449 	s64 t;
450 
451 	__asm__ __volatile__(
452 "1:	ldarx	%0,0,%2		# atomic64_dec\n\
453 	addic	%0,%0,-1\n\
454 	stdcx.	%0,0,%2\n\
455 	bne-	1b"
456 	: "=&r" (t), "+m" (v->counter)
457 	: "r" (&v->counter)
458 	: "cc", "xer");
459 }
460 #define arch_atomic64_dec arch_atomic64_dec
461 
462 static __inline__ s64 arch_atomic64_dec_return_relaxed(atomic64_t *v)
463 {
464 	s64 t;
465 
466 	__asm__ __volatile__(
467 "1:	ldarx	%0,0,%2		# atomic64_dec_return_relaxed\n"
468 "	addic	%0,%0,-1\n"
469 "	stdcx.	%0,0,%2\n"
470 "	bne-	1b"
471 	: "=&r" (t), "+m" (v->counter)
472 	: "r" (&v->counter)
473 	: "cc", "xer");
474 
475 	return t;
476 }
477 
478 #define arch_atomic64_inc_return_relaxed arch_atomic64_inc_return_relaxed
479 #define arch_atomic64_dec_return_relaxed arch_atomic64_dec_return_relaxed
480 
481 /*
482  * Atomically test *v and decrement if it is greater than 0.
483  * The function returns the old value of *v minus 1.
484  */
485 static __inline__ s64 arch_atomic64_dec_if_positive(atomic64_t *v)
486 {
487 	s64 t;
488 
489 	__asm__ __volatile__(
490 	PPC_ATOMIC_ENTRY_BARRIER
491 "1:	ldarx	%0,0,%1		# atomic64_dec_if_positive\n\
492 	addic.	%0,%0,-1\n\
493 	blt-	2f\n\
494 	stdcx.	%0,0,%1\n\
495 	bne-	1b"
496 	PPC_ATOMIC_EXIT_BARRIER
497 	"\n\
498 2:"	: "=&r" (t)
499 	: "r" (&v->counter)
500 	: "cc", "xer", "memory");
501 
502 	return t;
503 }
504 #define arch_atomic64_dec_if_positive arch_atomic64_dec_if_positive
505 
506 #define arch_atomic64_cmpxchg(v, o, n) \
507 	(arch_cmpxchg(&((v)->counter), (o), (n)))
508 #define arch_atomic64_cmpxchg_relaxed(v, o, n) \
509 	arch_cmpxchg_relaxed(&((v)->counter), (o), (n))
510 #define arch_atomic64_cmpxchg_acquire(v, o, n) \
511 	arch_cmpxchg_acquire(&((v)->counter), (o), (n))
512 
513 #define arch_atomic64_xchg(v, new) \
514 	(arch_xchg(&((v)->counter), new))
515 #define arch_atomic64_xchg_relaxed(v, new) \
516 	arch_xchg_relaxed(&((v)->counter), (new))
517 
518 /**
519  * atomic64_fetch_add_unless - add unless the number is a given value
520  * @v: pointer of type atomic64_t
521  * @a: the amount to add to v...
522  * @u: ...unless v is equal to u.
523  *
524  * Atomically adds @a to @v, so long as it was not @u.
525  * Returns the old value of @v.
526  */
527 static __inline__ s64 arch_atomic64_fetch_add_unless(atomic64_t *v, s64 a, s64 u)
528 {
529 	s64 t;
530 
531 	__asm__ __volatile__ (
532 	PPC_ATOMIC_ENTRY_BARRIER
533 "1:	ldarx	%0,0,%1		# atomic64_fetch_add_unless\n\
534 	cmpd	0,%0,%3 \n\
535 	beq	2f \n\
536 	add	%0,%2,%0 \n"
537 "	stdcx.	%0,0,%1 \n\
538 	bne-	1b \n"
539 	PPC_ATOMIC_EXIT_BARRIER
540 "	subf	%0,%2,%0 \n\
541 2:"
542 	: "=&r" (t)
543 	: "r" (&v->counter), "r" (a), "r" (u)
544 	: "cc", "memory");
545 
546 	return t;
547 }
548 #define arch_atomic64_fetch_add_unless arch_atomic64_fetch_add_unless
549 
550 /**
551  * atomic_inc64_not_zero - increment unless the number is zero
552  * @v: pointer of type atomic64_t
553  *
554  * Atomically increments @v by 1, so long as @v is non-zero.
555  * Returns non-zero if @v was non-zero, and zero otherwise.
556  */
557 static __inline__ int arch_atomic64_inc_not_zero(atomic64_t *v)
558 {
559 	s64 t1, t2;
560 
561 	__asm__ __volatile__ (
562 	PPC_ATOMIC_ENTRY_BARRIER
563 "1:	ldarx	%0,0,%2		# atomic64_inc_not_zero\n\
564 	cmpdi	0,%0,0\n\
565 	beq-	2f\n\
566 	addic	%1,%0,1\n\
567 	stdcx.	%1,0,%2\n\
568 	bne-	1b\n"
569 	PPC_ATOMIC_EXIT_BARRIER
570 	"\n\
571 2:"
572 	: "=&r" (t1), "=&r" (t2)
573 	: "r" (&v->counter)
574 	: "cc", "xer", "memory");
575 
576 	return t1 != 0;
577 }
578 #define arch_atomic64_inc_not_zero(v) arch_atomic64_inc_not_zero((v))
579 
580 #endif /* __powerpc64__ */
581 
582 #endif /* __KERNEL__ */
583 #endif /* _ASM_POWERPC_ATOMIC_H_ */
584