xref: /freebsd/sys/powerpc/include/atomic.h (revision 6e778a7efdc0e804471750157f6bacd1ef7d1580)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2008 Marcel Moolenaar
5  * Copyright (c) 2001 Benno Rice
6  * Copyright (c) 2001 David E. O'Brien
7  * Copyright (c) 1998 Doug Rabson
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  * $FreeBSD$
32  */
33 
34 #ifndef _MACHINE_ATOMIC_H_
35 #define	_MACHINE_ATOMIC_H_
36 
37 #ifndef _SYS_CDEFS_H_
38 #error this file needs sys/cdefs.h as a prerequisite
39 #endif
40 
41 /*
42  * The __ATOMIC_REL/ACQ() macros provide memory barriers only in conjunction
43  * with the atomic lXarx/stXcx. sequences below. They are not exposed outside
44  * of this file. See also Appendix B.2 of Book II of the architecture manual.
45  *
46  * Note that not all Book-E processors accept the light-weight sync variant.
47  * In particular, early models of E500 cores are known to wedge. Bank on all
48  * 64-bit capable CPUs to accept lwsync properly and pressimize 32-bit CPUs
49  * to use the heavier-weight sync.
50  */
51 
52 #ifdef __powerpc64__
53 #define mb()		__asm __volatile("sync" : : : "memory")
54 #define rmb()		__asm __volatile("lwsync" : : : "memory")
55 #define wmb()		__asm __volatile("lwsync" : : : "memory")
56 #define __ATOMIC_REL()	__asm __volatile("lwsync" : : : "memory")
57 #define __ATOMIC_ACQ()	__asm __volatile("isync" : : : "memory")
58 #else
59 #define mb()		__asm __volatile("sync" : : : "memory")
60 #define rmb()		__asm __volatile("sync" : : : "memory")
61 #define wmb()		__asm __volatile("sync" : : : "memory")
62 #define __ATOMIC_REL()	__asm __volatile("sync" : : : "memory")
63 #define __ATOMIC_ACQ()	__asm __volatile("isync" : : : "memory")
64 #endif
65 
66 static __inline void
67 powerpc_lwsync(void)
68 {
69 
70 #ifdef __powerpc64__
71 	__asm __volatile("lwsync" : : : "memory");
72 #else
73 	__asm __volatile("sync" : : : "memory");
74 #endif
75 }
76 
77 /*
78  * atomic_add(p, v)
79  * { *p += v; }
80  */
81 
82 #define __atomic_add_int(p, v, t)				\
83     __asm __volatile(						\
84 	"1:	lwarx	%0, 0, %2\n"				\
85 	"	add	%0, %3, %0\n"				\
86 	"	stwcx.	%0, 0, %2\n"				\
87 	"	bne-	1b\n"					\
88 	: "=&r" (t), "=m" (*p)					\
89 	: "r" (p), "r" (v), "m" (*p)				\
90 	: "cr0", "memory")					\
91     /* __atomic_add_int */
92 
93 #ifdef __powerpc64__
94 #define __atomic_add_long(p, v, t)				\
95     __asm __volatile(						\
96 	"1:	ldarx	%0, 0, %2\n"				\
97 	"	add	%0, %3, %0\n"				\
98 	"	stdcx.	%0, 0, %2\n"				\
99 	"	bne-	1b\n"					\
100 	: "=&r" (t), "=m" (*p)					\
101 	: "r" (p), "r" (v), "m" (*p)				\
102 	: "cr0", "memory")					\
103     /* __atomic_add_long */
104 #else
105 #define	__atomic_add_long(p, v, t)				\
106     __asm __volatile(						\
107 	"1:	lwarx	%0, 0, %2\n"				\
108 	"	add	%0, %3, %0\n"				\
109 	"	stwcx.	%0, 0, %2\n"				\
110 	"	bne-	1b\n"					\
111 	: "=&r" (t), "=m" (*p)					\
112 	: "r" (p), "r" (v), "m" (*p)				\
113 	: "cr0", "memory")					\
114     /* __atomic_add_long */
115 #endif
116 
117 #define	_ATOMIC_ADD(type)					\
118     static __inline void					\
119     atomic_add_##type(volatile u_##type *p, u_##type v) {	\
120 	u_##type t;						\
121 	__atomic_add_##type(p, v, t);				\
122     }								\
123 								\
124     static __inline void					\
125     atomic_add_acq_##type(volatile u_##type *p, u_##type v) {	\
126 	u_##type t;						\
127 	__atomic_add_##type(p, v, t);				\
128 	__ATOMIC_ACQ();						\
129     }								\
130 								\
131     static __inline void					\
132     atomic_add_rel_##type(volatile u_##type *p, u_##type v) {	\
133 	u_##type t;						\
134 	__ATOMIC_REL();						\
135 	__atomic_add_##type(p, v, t);				\
136     }								\
137     /* _ATOMIC_ADD */
138 
139 _ATOMIC_ADD(int)
140 _ATOMIC_ADD(long)
141 
142 #define	atomic_add_32		atomic_add_int
143 #define	atomic_add_acq_32	atomic_add_acq_int
144 #define	atomic_add_rel_32	atomic_add_rel_int
145 
146 #ifdef __powerpc64__
147 #define	atomic_add_64		atomic_add_long
148 #define	atomic_add_acq_64	atomic_add_acq_long
149 #define	atomic_add_rel_64	atomic_add_rel_long
150 
151 #define	atomic_add_ptr		atomic_add_long
152 #define	atomic_add_acq_ptr	atomic_add_acq_long
153 #define	atomic_add_rel_ptr	atomic_add_rel_long
154 #else
155 #define	atomic_add_ptr		atomic_add_int
156 #define	atomic_add_acq_ptr	atomic_add_acq_int
157 #define	atomic_add_rel_ptr	atomic_add_rel_int
158 #endif
159 #undef _ATOMIC_ADD
160 #undef __atomic_add_long
161 #undef __atomic_add_int
162 
163 /*
164  * atomic_clear(p, v)
165  * { *p &= ~v; }
166  */
167 
168 #define __atomic_clear_int(p, v, t)				\
169     __asm __volatile(						\
170 	"1:	lwarx	%0, 0, %2\n"				\
171 	"	andc	%0, %0, %3\n"				\
172 	"	stwcx.	%0, 0, %2\n"				\
173 	"	bne-	1b\n"					\
174 	: "=&r" (t), "=m" (*p)					\
175 	: "r" (p), "r" (v), "m" (*p)				\
176 	: "cr0", "memory")					\
177     /* __atomic_clear_int */
178 
179 #ifdef __powerpc64__
180 #define __atomic_clear_long(p, v, t)				\
181     __asm __volatile(						\
182 	"1:	ldarx	%0, 0, %2\n"				\
183 	"	andc	%0, %0, %3\n"				\
184 	"	stdcx.	%0, 0, %2\n"				\
185 	"	bne-	1b\n"					\
186 	: "=&r" (t), "=m" (*p)					\
187 	: "r" (p), "r" (v), "m" (*p)				\
188 	: "cr0", "memory")					\
189     /* __atomic_clear_long */
190 #else
191 #define	__atomic_clear_long(p, v, t)				\
192     __asm __volatile(						\
193 	"1:	lwarx	%0, 0, %2\n"				\
194 	"	andc	%0, %0, %3\n"				\
195 	"	stwcx.	%0, 0, %2\n"				\
196 	"	bne-	1b\n"					\
197 	: "=&r" (t), "=m" (*p)					\
198 	: "r" (p), "r" (v), "m" (*p)				\
199 	: "cr0", "memory")					\
200     /* __atomic_clear_long */
201 #endif
202 
203 #define	_ATOMIC_CLEAR(type)					\
204     static __inline void					\
205     atomic_clear_##type(volatile u_##type *p, u_##type v) {	\
206 	u_##type t;						\
207 	__atomic_clear_##type(p, v, t);				\
208     }								\
209 								\
210     static __inline void					\
211     atomic_clear_acq_##type(volatile u_##type *p, u_##type v) {	\
212 	u_##type t;						\
213 	__atomic_clear_##type(p, v, t);				\
214 	__ATOMIC_ACQ();						\
215     }								\
216 								\
217     static __inline void					\
218     atomic_clear_rel_##type(volatile u_##type *p, u_##type v) {	\
219 	u_##type t;						\
220 	__ATOMIC_REL();						\
221 	__atomic_clear_##type(p, v, t);				\
222     }								\
223     /* _ATOMIC_CLEAR */
224 
225 
226 _ATOMIC_CLEAR(int)
227 _ATOMIC_CLEAR(long)
228 
229 #define	atomic_clear_32		atomic_clear_int
230 #define	atomic_clear_acq_32	atomic_clear_acq_int
231 #define	atomic_clear_rel_32	atomic_clear_rel_int
232 
233 #ifdef __powerpc64__
234 #define	atomic_clear_64		atomic_clear_long
235 #define	atomic_clear_acq_64	atomic_clear_acq_long
236 #define	atomic_clear_rel_64	atomic_clear_rel_long
237 
238 #define	atomic_clear_ptr	atomic_clear_long
239 #define	atomic_clear_acq_ptr	atomic_clear_acq_long
240 #define	atomic_clear_rel_ptr	atomic_clear_rel_long
241 #else
242 #define	atomic_clear_ptr	atomic_clear_int
243 #define	atomic_clear_acq_ptr	atomic_clear_acq_int
244 #define	atomic_clear_rel_ptr	atomic_clear_rel_int
245 #endif
246 #undef _ATOMIC_CLEAR
247 #undef __atomic_clear_long
248 #undef __atomic_clear_int
249 
250 /*
251  * atomic_cmpset(p, o, n)
252  */
253 /* TODO -- see below */
254 
255 /*
256  * atomic_load_acq(p)
257  */
258 /* TODO -- see below */
259 
260 /*
261  * atomic_readandclear(p)
262  */
263 /* TODO -- see below */
264 
265 /*
266  * atomic_set(p, v)
267  * { *p |= v; }
268  */
269 
270 #define __atomic_set_int(p, v, t)				\
271     __asm __volatile(						\
272 	"1:	lwarx	%0, 0, %2\n"				\
273 	"	or	%0, %3, %0\n"				\
274 	"	stwcx.	%0, 0, %2\n"				\
275 	"	bne-	1b\n"					\
276 	: "=&r" (t), "=m" (*p)					\
277 	: "r" (p), "r" (v), "m" (*p)				\
278 	: "cr0", "memory")					\
279     /* __atomic_set_int */
280 
281 #ifdef __powerpc64__
282 #define __atomic_set_long(p, v, t)				\
283     __asm __volatile(						\
284 	"1:	ldarx	%0, 0, %2\n"				\
285 	"	or	%0, %3, %0\n"				\
286 	"	stdcx.	%0, 0, %2\n"				\
287 	"	bne-	1b\n"					\
288 	: "=&r" (t), "=m" (*p)					\
289 	: "r" (p), "r" (v), "m" (*p)				\
290 	: "cr0", "memory")					\
291     /* __atomic_set_long */
292 #else
293 #define	__atomic_set_long(p, v, t)				\
294     __asm __volatile(						\
295 	"1:	lwarx	%0, 0, %2\n"				\
296 	"	or	%0, %3, %0\n"				\
297 	"	stwcx.	%0, 0, %2\n"				\
298 	"	bne-	1b\n"					\
299 	: "=&r" (t), "=m" (*p)					\
300 	: "r" (p), "r" (v), "m" (*p)				\
301 	: "cr0", "memory")					\
302     /* __atomic_set_long */
303 #endif
304 
305 #define	_ATOMIC_SET(type)					\
306     static __inline void					\
307     atomic_set_##type(volatile u_##type *p, u_##type v) {	\
308 	u_##type t;						\
309 	__atomic_set_##type(p, v, t);				\
310     }								\
311 								\
312     static __inline void					\
313     atomic_set_acq_##type(volatile u_##type *p, u_##type v) {	\
314 	u_##type t;						\
315 	__atomic_set_##type(p, v, t);				\
316 	__ATOMIC_ACQ();						\
317     }								\
318 								\
319     static __inline void					\
320     atomic_set_rel_##type(volatile u_##type *p, u_##type v) {	\
321 	u_##type t;						\
322 	__ATOMIC_REL();						\
323 	__atomic_set_##type(p, v, t);				\
324     }								\
325     /* _ATOMIC_SET */
326 
327 _ATOMIC_SET(int)
328 _ATOMIC_SET(long)
329 
330 #define	atomic_set_32		atomic_set_int
331 #define	atomic_set_acq_32	atomic_set_acq_int
332 #define	atomic_set_rel_32	atomic_set_rel_int
333 
334 #ifdef __powerpc64__
335 #define	atomic_set_64		atomic_set_long
336 #define	atomic_set_acq_64	atomic_set_acq_long
337 #define	atomic_set_rel_64	atomic_set_rel_long
338 
339 #define	atomic_set_ptr		atomic_set_long
340 #define	atomic_set_acq_ptr	atomic_set_acq_long
341 #define	atomic_set_rel_ptr	atomic_set_rel_long
342 #else
343 #define	atomic_set_ptr		atomic_set_int
344 #define	atomic_set_acq_ptr	atomic_set_acq_int
345 #define	atomic_set_rel_ptr	atomic_set_rel_int
346 #endif
347 #undef _ATOMIC_SET
348 #undef __atomic_set_long
349 #undef __atomic_set_int
350 
351 /*
352  * atomic_subtract(p, v)
353  * { *p -= v; }
354  */
355 
356 #define __atomic_subtract_int(p, v, t)				\
357     __asm __volatile(						\
358 	"1:	lwarx	%0, 0, %2\n"				\
359 	"	subf	%0, %3, %0\n"				\
360 	"	stwcx.	%0, 0, %2\n"				\
361 	"	bne-	1b\n"					\
362 	: "=&r" (t), "=m" (*p)					\
363 	: "r" (p), "r" (v), "m" (*p)				\
364 	: "cr0", "memory")					\
365     /* __atomic_subtract_int */
366 
367 #ifdef __powerpc64__
368 #define __atomic_subtract_long(p, v, t)				\
369     __asm __volatile(						\
370 	"1:	ldarx	%0, 0, %2\n"				\
371 	"	subf	%0, %3, %0\n"				\
372 	"	stdcx.	%0, 0, %2\n"				\
373 	"	bne-	1b\n"					\
374 	: "=&r" (t), "=m" (*p)					\
375 	: "r" (p), "r" (v), "m" (*p)				\
376 	: "cr0", "memory")					\
377     /* __atomic_subtract_long */
378 #else
379 #define	__atomic_subtract_long(p, v, t)				\
380     __asm __volatile(						\
381 	"1:	lwarx	%0, 0, %2\n"				\
382 	"	subf	%0, %3, %0\n"				\
383 	"	stwcx.	%0, 0, %2\n"				\
384 	"	bne-	1b\n"					\
385 	: "=&r" (t), "=m" (*p)					\
386 	: "r" (p), "r" (v), "m" (*p)				\
387 	: "cr0", "memory")					\
388     /* __atomic_subtract_long */
389 #endif
390 
391 #define	_ATOMIC_SUBTRACT(type)						\
392     static __inline void						\
393     atomic_subtract_##type(volatile u_##type *p, u_##type v) {		\
394 	u_##type t;							\
395 	__atomic_subtract_##type(p, v, t);				\
396     }									\
397 									\
398     static __inline void						\
399     atomic_subtract_acq_##type(volatile u_##type *p, u_##type v) {	\
400 	u_##type t;							\
401 	__atomic_subtract_##type(p, v, t);				\
402 	__ATOMIC_ACQ();							\
403     }									\
404 									\
405     static __inline void						\
406     atomic_subtract_rel_##type(volatile u_##type *p, u_##type v) {	\
407 	u_##type t;							\
408 	__ATOMIC_REL();							\
409 	__atomic_subtract_##type(p, v, t);				\
410     }									\
411     /* _ATOMIC_SUBTRACT */
412 
413 _ATOMIC_SUBTRACT(int)
414 _ATOMIC_SUBTRACT(long)
415 
416 #define	atomic_subtract_32	atomic_subtract_int
417 #define	atomic_subtract_acq_32	atomic_subtract_acq_int
418 #define	atomic_subtract_rel_32	atomic_subtract_rel_int
419 
420 #ifdef __powerpc64__
421 #define	atomic_subtract_64	atomic_subtract_long
422 #define	atomic_subtract_acq_64	atomic_subract_acq_long
423 #define	atomic_subtract_rel_64	atomic_subtract_rel_long
424 
425 #define	atomic_subtract_ptr	atomic_subtract_long
426 #define	atomic_subtract_acq_ptr	atomic_subtract_acq_long
427 #define	atomic_subtract_rel_ptr	atomic_subtract_rel_long
428 #else
429 #define	atomic_subtract_ptr	atomic_subtract_int
430 #define	atomic_subtract_acq_ptr	atomic_subtract_acq_int
431 #define	atomic_subtract_rel_ptr	atomic_subtract_rel_int
432 #endif
433 #undef _ATOMIC_SUBTRACT
434 #undef __atomic_subtract_long
435 #undef __atomic_subtract_int
436 
437 /*
438  * atomic_store_rel(p, v)
439  */
440 /* TODO -- see below */
441 
442 /*
443  * Old/original implementations that still need revisiting.
444  */
445 
446 static __inline u_int
447 atomic_readandclear_int(volatile u_int *addr)
448 {
449 	u_int result,temp;
450 
451 	__asm __volatile (
452 		"\tsync\n"			/* drain writes */
453 		"1:\tlwarx %0, 0, %3\n\t"	/* load old value */
454 		"li %1, 0\n\t"			/* load new value */
455 		"stwcx. %1, 0, %3\n\t"      	/* attempt to store */
456 		"bne- 1b\n\t"			/* spin if failed */
457 		: "=&r"(result), "=&r"(temp), "=m" (*addr)
458 		: "r" (addr), "m" (*addr)
459 		: "cr0", "memory");
460 
461 	return (result);
462 }
463 
464 #ifdef __powerpc64__
465 static __inline u_long
466 atomic_readandclear_long(volatile u_long *addr)
467 {
468 	u_long result,temp;
469 
470 	__asm __volatile (
471 		"\tsync\n"			/* drain writes */
472 		"1:\tldarx %0, 0, %3\n\t"	/* load old value */
473 		"li %1, 0\n\t"			/* load new value */
474 		"stdcx. %1, 0, %3\n\t"      	/* attempt to store */
475 		"bne- 1b\n\t"			/* spin if failed */
476 		: "=&r"(result), "=&r"(temp), "=m" (*addr)
477 		: "r" (addr), "m" (*addr)
478 		: "cr0", "memory");
479 
480 	return (result);
481 }
482 #endif
483 
484 #define	atomic_readandclear_32		atomic_readandclear_int
485 
486 #ifdef __powerpc64__
487 #define	atomic_readandclear_64		atomic_readandclear_long
488 
489 #define	atomic_readandclear_ptr		atomic_readandclear_long
490 #else
491 static __inline u_long
492 atomic_readandclear_long(volatile u_long *addr)
493 {
494 
495 	return ((u_long)atomic_readandclear_int((volatile u_int *)addr));
496 }
497 
498 #define	atomic_readandclear_ptr		atomic_readandclear_int
499 #endif
500 
501 /*
502  * We assume that a = b will do atomic loads and stores.
503  */
504 #define	ATOMIC_STORE_LOAD(TYPE)					\
505 static __inline u_##TYPE					\
506 atomic_load_acq_##TYPE(volatile u_##TYPE *p)			\
507 {								\
508 	u_##TYPE v;						\
509 								\
510 	v = *p;							\
511 	mb();							\
512 	return (v);						\
513 }								\
514 								\
515 static __inline void						\
516 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)	\
517 {								\
518 								\
519 	powerpc_lwsync();					\
520 	*p = v;							\
521 }
522 
523 ATOMIC_STORE_LOAD(int)
524 
525 #define	atomic_load_acq_32	atomic_load_acq_int
526 #define	atomic_store_rel_32	atomic_store_rel_int
527 
528 #ifdef __powerpc64__
529 ATOMIC_STORE_LOAD(long)
530 
531 #define	atomic_load_acq_64	atomic_load_acq_long
532 #define	atomic_store_rel_64	atomic_store_rel_long
533 
534 #define	atomic_load_acq_ptr	atomic_load_acq_long
535 #define	atomic_store_rel_ptr	atomic_store_rel_long
536 #else
537 static __inline u_long
538 atomic_load_acq_long(volatile u_long *addr)
539 {
540 
541 	return ((u_long)atomic_load_acq_int((volatile u_int *)addr));
542 }
543 
544 static __inline void
545 atomic_store_rel_long(volatile u_long *addr, u_long val)
546 {
547 
548 	atomic_store_rel_int((volatile u_int *)addr, (u_int)val);
549 }
550 
551 #define	atomic_load_acq_ptr	atomic_load_acq_int
552 #define	atomic_store_rel_ptr	atomic_store_rel_int
553 #endif
554 #undef ATOMIC_STORE_LOAD
555 
556 /*
557  * Atomically compare the value stored at *p with cmpval and if the
558  * two values are equal, update the value of *p with newval. Returns
559  * zero if the compare failed, nonzero otherwise.
560  */
561 static __inline int
562 atomic_cmpset_int(volatile u_int* p, u_int cmpval, u_int newval)
563 {
564 	int	ret;
565 
566 	__asm __volatile (
567 		"1:\tlwarx %0, 0, %2\n\t"	/* load old value */
568 		"cmplw %3, %0\n\t"		/* compare */
569 		"bne 2f\n\t"			/* exit if not equal */
570 		"stwcx. %4, 0, %2\n\t"      	/* attempt to store */
571 		"bne- 1b\n\t"			/* spin if failed */
572 		"li %0, 1\n\t"			/* success - retval = 1 */
573 		"b 3f\n\t"			/* we've succeeded */
574 		"2:\n\t"
575 		"stwcx. %0, 0, %2\n\t"       	/* clear reservation (74xx) */
576 		"li %0, 0\n\t"			/* failure - retval = 0 */
577 		"3:\n\t"
578 		: "=&r" (ret), "=m" (*p)
579 		: "r" (p), "r" (cmpval), "r" (newval), "m" (*p)
580 		: "cr0", "memory");
581 
582 	return (ret);
583 }
584 static __inline int
585 atomic_cmpset_long(volatile u_long* p, u_long cmpval, u_long newval)
586 {
587 	int ret;
588 
589 	__asm __volatile (
590 	    #ifdef __powerpc64__
591 		"1:\tldarx %0, 0, %2\n\t"	/* load old value */
592 		"cmpld %3, %0\n\t"		/* compare */
593 		"bne 2f\n\t"			/* exit if not equal */
594 		"stdcx. %4, 0, %2\n\t"		/* attempt to store */
595 	    #else
596 		"1:\tlwarx %0, 0, %2\n\t"	/* load old value */
597 		"cmplw %3, %0\n\t"		/* compare */
598 		"bne 2f\n\t"			/* exit if not equal */
599 		"stwcx. %4, 0, %2\n\t"		/* attempt to store */
600 	    #endif
601 		"bne- 1b\n\t"			/* spin if failed */
602 		"li %0, 1\n\t"			/* success - retval = 1 */
603 		"b 3f\n\t"			/* we've succeeded */
604 		"2:\n\t"
605 	    #ifdef __powerpc64__
606 		"stdcx. %0, 0, %2\n\t"		/* clear reservation (74xx) */
607 	    #else
608 		"stwcx. %0, 0, %2\n\t"		/* clear reservation (74xx) */
609 	    #endif
610 		"li %0, 0\n\t"			/* failure - retval = 0 */
611 		"3:\n\t"
612 		: "=&r" (ret), "=m" (*p)
613 		: "r" (p), "r" (cmpval), "r" (newval), "m" (*p)
614 		: "cr0", "memory");
615 
616 	return (ret);
617 }
618 
619 static __inline int
620 atomic_cmpset_acq_int(volatile u_int *p, u_int cmpval, u_int newval)
621 {
622 	int retval;
623 
624 	retval = atomic_cmpset_int(p, cmpval, newval);
625 	__ATOMIC_ACQ();
626 	return (retval);
627 }
628 
629 static __inline int
630 atomic_cmpset_rel_int(volatile u_int *p, u_int cmpval, u_int newval)
631 {
632 	__ATOMIC_REL();
633 	return (atomic_cmpset_int(p, cmpval, newval));
634 }
635 
636 static __inline int
637 atomic_cmpset_acq_long(volatile u_long *p, u_long cmpval, u_long newval)
638 {
639 	u_long retval;
640 
641 	retval = atomic_cmpset_long(p, cmpval, newval);
642 	__ATOMIC_ACQ();
643 	return (retval);
644 }
645 
646 static __inline int
647 atomic_cmpset_rel_long(volatile u_long *p, u_long cmpval, u_long newval)
648 {
649 	__ATOMIC_REL();
650 	return (atomic_cmpset_long(p, cmpval, newval));
651 }
652 
653 #define	atomic_cmpset_32	atomic_cmpset_int
654 #define	atomic_cmpset_acq_32	atomic_cmpset_acq_int
655 #define	atomic_cmpset_rel_32	atomic_cmpset_rel_int
656 
657 #ifdef __powerpc64__
658 #define	atomic_cmpset_64	atomic_cmpset_long
659 #define	atomic_cmpset_acq_64	atomic_cmpset_acq_long
660 #define	atomic_cmpset_rel_64	atomic_cmpset_rel_long
661 
662 #define	atomic_cmpset_ptr	atomic_cmpset_long
663 #define	atomic_cmpset_acq_ptr	atomic_cmpset_acq_long
664 #define	atomic_cmpset_rel_ptr	atomic_cmpset_rel_long
665 #else
666 #define	atomic_cmpset_ptr	atomic_cmpset_int
667 #define	atomic_cmpset_acq_ptr	atomic_cmpset_acq_int
668 #define	atomic_cmpset_rel_ptr	atomic_cmpset_rel_int
669 #endif
670 
671 /*
672  * Atomically compare the value stored at *p with *cmpval and if the
673  * two values are equal, update the value of *p with newval. Returns
674  * zero if the compare failed and sets *cmpval to the read value from *p,
675  * nonzero otherwise.
676  */
677 static __inline int
678 atomic_fcmpset_int(volatile u_int *p, u_int *cmpval, u_int newval)
679 {
680 	int	ret;
681 
682 	__asm __volatile (
683 		"lwarx %0, 0, %3\n\t"	/* load old value */
684 		"cmplw %4, %0\n\t"		/* compare */
685 		"bne 1f\n\t"			/* exit if not equal */
686 		"stwcx. %5, 0, %3\n\t"      	/* attempt to store */
687 		"bne- 1f\n\t"			/* exit if failed */
688 		"li %0, 1\n\t"			/* success - retval = 1 */
689 		"b 2f\n\t"			/* we've succeeded */
690 		"1:\n\t"
691 		"stwcx. %0, 0, %3\n\t"       	/* clear reservation (74xx) */
692 		"stwx %0, 0, %7\n\t"
693 		"li %0, 0\n\t"			/* failure - retval = 0 */
694 		"2:\n\t"
695 		: "=&r" (ret), "=m" (*p), "=m" (*cmpval)
696 		: "r" (p), "r" (*cmpval), "r" (newval), "m" (*p), "r"(cmpval)
697 		: "cr0", "memory");
698 
699 	return (ret);
700 }
701 static __inline int
702 atomic_fcmpset_long(volatile u_long *p, u_long *cmpval, u_long newval)
703 {
704 	int ret;
705 
706 	__asm __volatile (
707 	    #ifdef __powerpc64__
708 		"ldarx %0, 0, %3\n\t"	/* load old value */
709 		"cmpld %4, %0\n\t"		/* compare */
710 		"bne 1f\n\t"			/* exit if not equal */
711 		"stdcx. %5, 0, %3\n\t"		/* attempt to store */
712 	    #else
713 		"lwarx %0, 0, %3\n\t"	/* load old value */
714 		"cmplw %4, %0\n\t"		/* compare */
715 		"bne 1f\n\t"			/* exit if not equal */
716 		"stwcx. %5, 0, %3\n\t"		/* attempt to store */
717 	    #endif
718 		"bne- 1f\n\t"			/* exit if failed */
719 		"li %0, 1\n\t"			/* success - retval = 1 */
720 		"b 2f\n\t"			/* we've succeeded */
721 		"1:\n\t"
722 	    #ifdef __powerpc64__
723 		"stdcx. %0, 0, %3\n\t"		/* clear reservation (74xx) */
724 		"stdx %0, 0, %7\n\t"
725 	    #else
726 		"stwcx. %0, 0, %3\n\t"		/* clear reservation (74xx) */
727 		"stwx %0, 0, %7\n\t"
728 	    #endif
729 		"li %0, 0\n\t"			/* failure - retval = 0 */
730 		"2:\n\t"
731 		: "=&r" (ret), "=m" (*p), "=m" (*cmpval)
732 		: "r" (p), "r" (*cmpval), "r" (newval), "m" (*p), "r"(cmpval)
733 		: "cr0", "memory");
734 
735 	return (ret);
736 }
737 
738 static __inline int
739 atomic_fcmpset_acq_int(volatile u_int *p, u_int *cmpval, u_int newval)
740 {
741 	int retval;
742 
743 	retval = atomic_fcmpset_int(p, cmpval, newval);
744 	__ATOMIC_ACQ();
745 	return (retval);
746 }
747 
748 static __inline int
749 atomic_fcmpset_rel_int(volatile u_int *p, u_int *cmpval, u_int newval)
750 {
751 	__ATOMIC_REL();
752 	return (atomic_fcmpset_int(p, cmpval, newval));
753 }
754 
755 static __inline int
756 atomic_fcmpset_acq_long(volatile u_long *p, u_long *cmpval, u_long newval)
757 {
758 	u_long retval;
759 
760 	retval = atomic_fcmpset_long(p, cmpval, newval);
761 	__ATOMIC_ACQ();
762 	return (retval);
763 }
764 
765 static __inline int
766 atomic_fcmpset_rel_long(volatile u_long *p, u_long *cmpval, u_long newval)
767 {
768 	__ATOMIC_REL();
769 	return (atomic_fcmpset_long(p, cmpval, newval));
770 }
771 
772 #define	atomic_fcmpset_32	atomic_fcmpset_int
773 #define	atomic_fcmpset_acq_32	atomic_fcmpset_acq_int
774 #define	atomic_fcmpset_rel_32	atomic_fcmpset_rel_int
775 
776 #ifdef __powerpc64__
777 #define	atomic_fcmpset_64	atomic_fcmpset_long
778 #define	atomic_fcmpset_acq_64	atomic_fcmpset_acq_long
779 #define	atomic_fcmpset_rel_64	atomic_fcmpset_rel_long
780 
781 #define	atomic_fcmpset_ptr	atomic_fcmpset_long
782 #define	atomic_fcmpset_acq_ptr	atomic_fcmpset_acq_long
783 #define	atomic_fcmpset_rel_ptr	atomic_fcmpset_rel_long
784 #else
785 #define	atomic_fcmpset_ptr	atomic_fcmpset_int
786 #define	atomic_fcmpset_acq_ptr	atomic_fcmpset_acq_int
787 #define	atomic_fcmpset_rel_ptr	atomic_fcmpset_rel_int
788 #endif
789 
790 static __inline u_int
791 atomic_fetchadd_int(volatile u_int *p, u_int v)
792 {
793 	u_int value;
794 
795 	do {
796 		value = *p;
797 	} while (!atomic_cmpset_int(p, value, value + v));
798 	return (value);
799 }
800 
801 static __inline u_long
802 atomic_fetchadd_long(volatile u_long *p, u_long v)
803 {
804 	u_long value;
805 
806 	do {
807 		value = *p;
808 	} while (!atomic_cmpset_long(p, value, value + v));
809 	return (value);
810 }
811 
812 static __inline u_int
813 atomic_swap_32(volatile u_int *p, u_int v)
814 {
815 	u_int prev;
816 
817 	__asm __volatile(
818 	"1:	lwarx	%0,0,%2\n"
819 	"	stwcx.	%3,0,%2\n"
820 	"	bne-	1b\n"
821 	: "=&r" (prev), "+m" (*(volatile u_int *)p)
822 	: "r" (p), "r" (v)
823 	: "cr0", "memory");
824 
825 	return (prev);
826 }
827 
828 #ifdef __powerpc64__
829 static __inline u_long
830 atomic_swap_64(volatile u_long *p, u_long v)
831 {
832 	u_long prev;
833 
834 	__asm __volatile(
835 	"1:	ldarx	%0,0,%2\n"
836 	"	stdcx.	%3,0,%2\n"
837 	"	bne-	1b\n"
838 	: "=&r" (prev), "+m" (*(volatile u_long *)p)
839 	: "r" (p), "r" (v)
840 	: "cr0", "memory");
841 
842 	return (prev);
843 }
844 #endif
845 
846 #define	atomic_fetchadd_32	atomic_fetchadd_int
847 #define	atomic_swap_int		atomic_swap_32
848 
849 #ifdef __powerpc64__
850 #define	atomic_fetchadd_64	atomic_fetchadd_long
851 #define	atomic_swap_long	atomic_swap_64
852 #define	atomic_swap_ptr		atomic_swap_64
853 #endif
854 
855 #undef __ATOMIC_REL
856 #undef __ATOMIC_ACQ
857 
858 static __inline void
859 atomic_thread_fence_acq(void)
860 {
861 
862 	powerpc_lwsync();
863 }
864 
865 static __inline void
866 atomic_thread_fence_rel(void)
867 {
868 
869 	powerpc_lwsync();
870 }
871 
872 static __inline void
873 atomic_thread_fence_acq_rel(void)
874 {
875 
876 	powerpc_lwsync();
877 }
878 
879 static __inline void
880 atomic_thread_fence_seq_cst(void)
881 {
882 
883 	__asm __volatile("sync" : : : "memory");
884 }
885 
886 #endif /* ! _MACHINE_ATOMIC_H_ */
887