xref: /openbmc/linux/arch/powerpc/include/asm/atomic.h (revision f16fe2d3)
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, suffix, sign, ...)			\
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 "%I2" suffix " %0,%0,%2\n"				\
48 "	stwcx.	%0,0,%3 \n"						\
49 "	bne-	1b\n"							\
50 	: "=&r" (t), "+m" (v->counter)					\
51 	: "r"#sign (a), "r" (&v->counter)				\
52 	: "cc", ##__VA_ARGS__);						\
53 }									\
54 
55 #define ATOMIC_OP_RETURN_RELAXED(op, asm_op, suffix, sign, ...)		\
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 "%I2" suffix " %0,%0,%2\n"				\
63 "	stwcx.	%0,0,%3\n"						\
64 "	bne-	1b\n"							\
65 	: "=&r" (t), "+m" (v->counter)					\
66 	: "r"#sign (a), "r" (&v->counter)				\
67 	: "cc", ##__VA_ARGS__);						\
68 									\
69 	return t;							\
70 }
71 
72 #define ATOMIC_FETCH_OP_RELAXED(op, asm_op, suffix, sign, ...)		\
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 "%I3" suffix " %1,%0,%3\n"				\
80 "	stwcx.	%1,0,%4\n"						\
81 "	bne-	1b\n"							\
82 	: "=&r" (res), "=&r" (t), "+m" (v->counter)			\
83 	: "r"#sign (a), "r" (&v->counter)				\
84 	: "cc", ##__VA_ARGS__);						\
85 									\
86 	return res;							\
87 }
88 
89 #define ATOMIC_OPS(op, asm_op, suffix, sign, ...)			\
90 	ATOMIC_OP(op, asm_op, suffix, sign, ##__VA_ARGS__)		\
91 	ATOMIC_OP_RETURN_RELAXED(op, asm_op, suffix, sign, ##__VA_ARGS__)\
92 	ATOMIC_FETCH_OP_RELAXED(op, asm_op, suffix, sign, ##__VA_ARGS__)
93 
94 ATOMIC_OPS(add, add, "c", I, "xer")
95 ATOMIC_OPS(sub, sub, "c", I, "xer")
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, suffix, sign)				\
105 	ATOMIC_OP(op, asm_op, suffix, sign)				\
106 	ATOMIC_FETCH_OP_RELAXED(op, asm_op, suffix, sign)
107 
108 ATOMIC_OPS(and, and, ".", K)
109 ATOMIC_OPS(or, or, "", K)
110 ATOMIC_OPS(xor, xor, "", K)
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 #define arch_atomic_cmpxchg(v, o, n) \
122 	(arch_cmpxchg(&((v)->counter), (o), (n)))
123 #define arch_atomic_cmpxchg_relaxed(v, o, n) \
124 	arch_cmpxchg_relaxed(&((v)->counter), (o), (n))
125 #define arch_atomic_cmpxchg_acquire(v, o, n) \
126 	arch_cmpxchg_acquire(&((v)->counter), (o), (n))
127 
128 #define arch_atomic_xchg(v, new) \
129 	(arch_xchg(&((v)->counter), new))
130 #define arch_atomic_xchg_relaxed(v, new) \
131 	arch_xchg_relaxed(&((v)->counter), (new))
132 
133 /*
134  * Don't want to override the generic atomic_try_cmpxchg_acquire, because
135  * we add a lock hint to the lwarx, which may not be wanted for the
136  * _acquire case (and is not used by the other _acquire variants so it
137  * would be a surprise).
138  */
139 static __always_inline bool
140 arch_atomic_try_cmpxchg_lock(atomic_t *v, int *old, int new)
141 {
142 	int r, o = *old;
143 
144 	__asm__ __volatile__ (
145 "1:	lwarx	%0,0,%2,%5	# atomic_try_cmpxchg_acquire		\n"
146 "	cmpw	0,%0,%3							\n"
147 "	bne-	2f							\n"
148 "	stwcx.	%4,0,%2							\n"
149 "	bne-	1b							\n"
150 "\t"	PPC_ACQUIRE_BARRIER "						\n"
151 "2:									\n"
152 	: "=&r" (r), "+m" (v->counter)
153 	: "r" (&v->counter), "r" (o), "r" (new), "i" (IS_ENABLED(CONFIG_PPC64) ? 1 : 0)
154 	: "cr0", "memory");
155 
156 	if (unlikely(r != o))
157 		*old = r;
158 	return likely(r == o);
159 }
160 
161 /**
162  * atomic_fetch_add_unless - add unless the number is a given value
163  * @v: pointer of type atomic_t
164  * @a: the amount to add to v...
165  * @u: ...unless v is equal to u.
166  *
167  * Atomically adds @a to @v, so long as it was not @u.
168  * Returns the old value of @v.
169  */
170 static __inline__ int arch_atomic_fetch_add_unless(atomic_t *v, int a, int u)
171 {
172 	int t;
173 
174 	__asm__ __volatile__ (
175 	PPC_ATOMIC_ENTRY_BARRIER
176 "1:	lwarx	%0,0,%1		# atomic_fetch_add_unless\n\
177 	cmpw	0,%0,%3 \n\
178 	beq	2f \n\
179 	add%I2c	%0,%0,%2 \n"
180 "	stwcx.	%0,0,%1 \n\
181 	bne-	1b \n"
182 	PPC_ATOMIC_EXIT_BARRIER
183 "	sub%I2c	%0,%0,%2 \n\
184 2:"
185 	: "=&r" (t)
186 	: "r" (&v->counter), "rI" (a), "r" (u)
187 	: "cc", "memory", "xer");
188 
189 	return t;
190 }
191 #define arch_atomic_fetch_add_unless arch_atomic_fetch_add_unless
192 
193 /*
194  * Atomically test *v and decrement if it is greater than 0.
195  * The function returns the old value of *v minus 1, even if
196  * the atomic variable, v, was not decremented.
197  */
198 static __inline__ int arch_atomic_dec_if_positive(atomic_t *v)
199 {
200 	int t;
201 
202 	__asm__ __volatile__(
203 	PPC_ATOMIC_ENTRY_BARRIER
204 "1:	lwarx	%0,0,%1		# atomic_dec_if_positive\n\
205 	cmpwi	%0,1\n\
206 	addi	%0,%0,-1\n\
207 	blt-	2f\n"
208 "	stwcx.	%0,0,%1\n\
209 	bne-	1b"
210 	PPC_ATOMIC_EXIT_BARRIER
211 	"\n\
212 2:"	: "=&b" (t)
213 	: "r" (&v->counter)
214 	: "cc", "memory");
215 
216 	return t;
217 }
218 #define arch_atomic_dec_if_positive arch_atomic_dec_if_positive
219 
220 #ifdef __powerpc64__
221 
222 #define ATOMIC64_INIT(i)	{ (i) }
223 
224 static __inline__ s64 arch_atomic64_read(const atomic64_t *v)
225 {
226 	s64 t;
227 
228 	__asm__ __volatile__("ld%U1%X1 %0,%1" : "=r"(t) : "m<>"(v->counter));
229 
230 	return t;
231 }
232 
233 static __inline__ void arch_atomic64_set(atomic64_t *v, s64 i)
234 {
235 	__asm__ __volatile__("std%U0%X0 %1,%0" : "=m<>"(v->counter) : "r"(i));
236 }
237 
238 #define ATOMIC64_OP(op, asm_op)						\
239 static __inline__ void arch_atomic64_##op(s64 a, atomic64_t *v)		\
240 {									\
241 	s64 t;								\
242 									\
243 	__asm__ __volatile__(						\
244 "1:	ldarx	%0,0,%3		# atomic64_" #op "\n"			\
245 	#asm_op " %0,%2,%0\n"						\
246 "	stdcx.	%0,0,%3 \n"						\
247 "	bne-	1b\n"							\
248 	: "=&r" (t), "+m" (v->counter)					\
249 	: "r" (a), "r" (&v->counter)					\
250 	: "cc");							\
251 }
252 
253 #define ATOMIC64_OP_RETURN_RELAXED(op, asm_op)				\
254 static inline s64							\
255 arch_atomic64_##op##_return_relaxed(s64 a, atomic64_t *v)		\
256 {									\
257 	s64 t;								\
258 									\
259 	__asm__ __volatile__(						\
260 "1:	ldarx	%0,0,%3		# atomic64_" #op "_return_relaxed\n"	\
261 	#asm_op " %0,%2,%0\n"						\
262 "	stdcx.	%0,0,%3\n"						\
263 "	bne-	1b\n"							\
264 	: "=&r" (t), "+m" (v->counter)					\
265 	: "r" (a), "r" (&v->counter)					\
266 	: "cc");							\
267 									\
268 	return t;							\
269 }
270 
271 #define ATOMIC64_FETCH_OP_RELAXED(op, asm_op)				\
272 static inline s64							\
273 arch_atomic64_fetch_##op##_relaxed(s64 a, atomic64_t *v)		\
274 {									\
275 	s64 res, t;							\
276 									\
277 	__asm__ __volatile__(						\
278 "1:	ldarx	%0,0,%4		# atomic64_fetch_" #op "_relaxed\n"	\
279 	#asm_op " %1,%3,%0\n"						\
280 "	stdcx.	%1,0,%4\n"						\
281 "	bne-	1b\n"							\
282 	: "=&r" (res), "=&r" (t), "+m" (v->counter)			\
283 	: "r" (a), "r" (&v->counter)					\
284 	: "cc");							\
285 									\
286 	return res;							\
287 }
288 
289 #define ATOMIC64_OPS(op, asm_op)					\
290 	ATOMIC64_OP(op, asm_op)						\
291 	ATOMIC64_OP_RETURN_RELAXED(op, asm_op)				\
292 	ATOMIC64_FETCH_OP_RELAXED(op, asm_op)
293 
294 ATOMIC64_OPS(add, add)
295 ATOMIC64_OPS(sub, subf)
296 
297 #define arch_atomic64_add_return_relaxed arch_atomic64_add_return_relaxed
298 #define arch_atomic64_sub_return_relaxed arch_atomic64_sub_return_relaxed
299 
300 #define arch_atomic64_fetch_add_relaxed arch_atomic64_fetch_add_relaxed
301 #define arch_atomic64_fetch_sub_relaxed arch_atomic64_fetch_sub_relaxed
302 
303 #undef ATOMIC64_OPS
304 #define ATOMIC64_OPS(op, asm_op)					\
305 	ATOMIC64_OP(op, asm_op)						\
306 	ATOMIC64_FETCH_OP_RELAXED(op, asm_op)
307 
308 ATOMIC64_OPS(and, and)
309 ATOMIC64_OPS(or, or)
310 ATOMIC64_OPS(xor, xor)
311 
312 #define arch_atomic64_fetch_and_relaxed arch_atomic64_fetch_and_relaxed
313 #define arch_atomic64_fetch_or_relaxed  arch_atomic64_fetch_or_relaxed
314 #define arch_atomic64_fetch_xor_relaxed arch_atomic64_fetch_xor_relaxed
315 
316 #undef ATOPIC64_OPS
317 #undef ATOMIC64_FETCH_OP_RELAXED
318 #undef ATOMIC64_OP_RETURN_RELAXED
319 #undef ATOMIC64_OP
320 
321 static __inline__ void arch_atomic64_inc(atomic64_t *v)
322 {
323 	s64 t;
324 
325 	__asm__ __volatile__(
326 "1:	ldarx	%0,0,%2		# atomic64_inc\n\
327 	addic	%0,%0,1\n\
328 	stdcx.	%0,0,%2 \n\
329 	bne-	1b"
330 	: "=&r" (t), "+m" (v->counter)
331 	: "r" (&v->counter)
332 	: "cc", "xer");
333 }
334 #define arch_atomic64_inc arch_atomic64_inc
335 
336 static __inline__ s64 arch_atomic64_inc_return_relaxed(atomic64_t *v)
337 {
338 	s64 t;
339 
340 	__asm__ __volatile__(
341 "1:	ldarx	%0,0,%2		# atomic64_inc_return_relaxed\n"
342 "	addic	%0,%0,1\n"
343 "	stdcx.	%0,0,%2\n"
344 "	bne-	1b"
345 	: "=&r" (t), "+m" (v->counter)
346 	: "r" (&v->counter)
347 	: "cc", "xer");
348 
349 	return t;
350 }
351 
352 static __inline__ void arch_atomic64_dec(atomic64_t *v)
353 {
354 	s64 t;
355 
356 	__asm__ __volatile__(
357 "1:	ldarx	%0,0,%2		# atomic64_dec\n\
358 	addic	%0,%0,-1\n\
359 	stdcx.	%0,0,%2\n\
360 	bne-	1b"
361 	: "=&r" (t), "+m" (v->counter)
362 	: "r" (&v->counter)
363 	: "cc", "xer");
364 }
365 #define arch_atomic64_dec arch_atomic64_dec
366 
367 static __inline__ s64 arch_atomic64_dec_return_relaxed(atomic64_t *v)
368 {
369 	s64 t;
370 
371 	__asm__ __volatile__(
372 "1:	ldarx	%0,0,%2		# atomic64_dec_return_relaxed\n"
373 "	addic	%0,%0,-1\n"
374 "	stdcx.	%0,0,%2\n"
375 "	bne-	1b"
376 	: "=&r" (t), "+m" (v->counter)
377 	: "r" (&v->counter)
378 	: "cc", "xer");
379 
380 	return t;
381 }
382 
383 #define arch_atomic64_inc_return_relaxed arch_atomic64_inc_return_relaxed
384 #define arch_atomic64_dec_return_relaxed arch_atomic64_dec_return_relaxed
385 
386 /*
387  * Atomically test *v and decrement if it is greater than 0.
388  * The function returns the old value of *v minus 1.
389  */
390 static __inline__ s64 arch_atomic64_dec_if_positive(atomic64_t *v)
391 {
392 	s64 t;
393 
394 	__asm__ __volatile__(
395 	PPC_ATOMIC_ENTRY_BARRIER
396 "1:	ldarx	%0,0,%1		# atomic64_dec_if_positive\n\
397 	addic.	%0,%0,-1\n\
398 	blt-	2f\n\
399 	stdcx.	%0,0,%1\n\
400 	bne-	1b"
401 	PPC_ATOMIC_EXIT_BARRIER
402 	"\n\
403 2:"	: "=&r" (t)
404 	: "r" (&v->counter)
405 	: "cc", "xer", "memory");
406 
407 	return t;
408 }
409 #define arch_atomic64_dec_if_positive arch_atomic64_dec_if_positive
410 
411 #define arch_atomic64_cmpxchg(v, o, n) \
412 	(arch_cmpxchg(&((v)->counter), (o), (n)))
413 #define arch_atomic64_cmpxchg_relaxed(v, o, n) \
414 	arch_cmpxchg_relaxed(&((v)->counter), (o), (n))
415 #define arch_atomic64_cmpxchg_acquire(v, o, n) \
416 	arch_cmpxchg_acquire(&((v)->counter), (o), (n))
417 
418 #define arch_atomic64_xchg(v, new) \
419 	(arch_xchg(&((v)->counter), new))
420 #define arch_atomic64_xchg_relaxed(v, new) \
421 	arch_xchg_relaxed(&((v)->counter), (new))
422 
423 /**
424  * atomic64_fetch_add_unless - add unless the number is a given value
425  * @v: pointer of type atomic64_t
426  * @a: the amount to add to v...
427  * @u: ...unless v is equal to u.
428  *
429  * Atomically adds @a to @v, so long as it was not @u.
430  * Returns the old value of @v.
431  */
432 static __inline__ s64 arch_atomic64_fetch_add_unless(atomic64_t *v, s64 a, s64 u)
433 {
434 	s64 t;
435 
436 	__asm__ __volatile__ (
437 	PPC_ATOMIC_ENTRY_BARRIER
438 "1:	ldarx	%0,0,%1		# atomic64_fetch_add_unless\n\
439 	cmpd	0,%0,%3 \n\
440 	beq	2f \n\
441 	add	%0,%2,%0 \n"
442 "	stdcx.	%0,0,%1 \n\
443 	bne-	1b \n"
444 	PPC_ATOMIC_EXIT_BARRIER
445 "	subf	%0,%2,%0 \n\
446 2:"
447 	: "=&r" (t)
448 	: "r" (&v->counter), "r" (a), "r" (u)
449 	: "cc", "memory");
450 
451 	return t;
452 }
453 #define arch_atomic64_fetch_add_unless arch_atomic64_fetch_add_unless
454 
455 /**
456  * atomic_inc64_not_zero - increment unless the number is zero
457  * @v: pointer of type atomic64_t
458  *
459  * Atomically increments @v by 1, so long as @v is non-zero.
460  * Returns non-zero if @v was non-zero, and zero otherwise.
461  */
462 static __inline__ int arch_atomic64_inc_not_zero(atomic64_t *v)
463 {
464 	s64 t1, t2;
465 
466 	__asm__ __volatile__ (
467 	PPC_ATOMIC_ENTRY_BARRIER
468 "1:	ldarx	%0,0,%2		# atomic64_inc_not_zero\n\
469 	cmpdi	0,%0,0\n\
470 	beq-	2f\n\
471 	addic	%1,%0,1\n\
472 	stdcx.	%1,0,%2\n\
473 	bne-	1b\n"
474 	PPC_ATOMIC_EXIT_BARRIER
475 	"\n\
476 2:"
477 	: "=&r" (t1), "=&r" (t2)
478 	: "r" (&v->counter)
479 	: "cc", "xer", "memory");
480 
481 	return t1 != 0;
482 }
483 #define arch_atomic64_inc_not_zero(v) arch_atomic64_inc_not_zero((v))
484 
485 #endif /* __powerpc64__ */
486 
487 #endif /* __KERNEL__ */
488 #endif /* _ASM_POWERPC_ATOMIC_H_ */
489