xref: /freebsd/sys/i386/include/atomic.h (revision 93a065e7496dfbfbd0a5b0208ef763f37ea975c7)
1 /*-
2  * Copyright (c) 1998 Doug Rabson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 #ifndef _MACHINE_ATOMIC_H_
29 #define	_MACHINE_ATOMIC_H_
30 
31 #ifndef _SYS_CDEFS_H_
32 #error this file needs sys/cdefs.h as a prerequisite
33 #endif
34 
35 #ifdef _KERNEL
36 #include <machine/md_var.h>
37 #include <machine/specialreg.h>
38 #endif
39 
40 #ifndef __OFFSETOF_MONITORBUF
41 /*
42  * __OFFSETOF_MONITORBUF == __pcpu_offset(pc_monitorbuf).
43  *
44  * The open-coded number is used instead of the symbolic expression to
45  * avoid a dependency on sys/pcpu.h in machine/atomic.h consumers.
46  * An assertion in i386/vm_machdep.c ensures that the value is correct.
47  */
48 #define	__OFFSETOF_MONITORBUF	0x180
49 
50 static __inline void
51 __mbk(void)
52 {
53 
54 	__asm __volatile("lock; addl $0,%%fs:%0"
55 	    : "+m" (*(u_int *)__OFFSETOF_MONITORBUF) : : "memory", "cc");
56 }
57 
58 static __inline void
59 __mbu(void)
60 {
61 
62 	__asm __volatile("lock; addl $0,(%%esp)" : : : "memory", "cc");
63 }
64 #endif
65 
66 /*
67  * Various simple operations on memory, each of which is atomic in the
68  * presence of interrupts and multiple processors.
69  *
70  * atomic_set_char(P, V)	(*(u_char *)(P) |= (V))
71  * atomic_clear_char(P, V)	(*(u_char *)(P) &= ~(V))
72  * atomic_add_char(P, V)	(*(u_char *)(P) += (V))
73  * atomic_subtract_char(P, V)	(*(u_char *)(P) -= (V))
74  *
75  * atomic_set_short(P, V)	(*(u_short *)(P) |= (V))
76  * atomic_clear_short(P, V)	(*(u_short *)(P) &= ~(V))
77  * atomic_add_short(P, V)	(*(u_short *)(P) += (V))
78  * atomic_subtract_short(P, V)	(*(u_short *)(P) -= (V))
79  *
80  * atomic_set_int(P, V)		(*(u_int *)(P) |= (V))
81  * atomic_clear_int(P, V)	(*(u_int *)(P) &= ~(V))
82  * atomic_add_int(P, V)		(*(u_int *)(P) += (V))
83  * atomic_subtract_int(P, V)	(*(u_int *)(P) -= (V))
84  * atomic_swap_int(P, V)	(return (*(u_int *)(P)); *(u_int *)(P) = (V);)
85  * atomic_readandclear_int(P)	(return (*(u_int *)(P)); *(u_int *)(P) = 0;)
86  *
87  * atomic_set_long(P, V)	(*(u_long *)(P) |= (V))
88  * atomic_clear_long(P, V)	(*(u_long *)(P) &= ~(V))
89  * atomic_add_long(P, V)	(*(u_long *)(P) += (V))
90  * atomic_subtract_long(P, V)	(*(u_long *)(P) -= (V))
91  * atomic_swap_long(P, V)	(return (*(u_long *)(P)); *(u_long *)(P) = (V);)
92  * atomic_readandclear_long(P)	(return (*(u_long *)(P)); *(u_long *)(P) = 0;)
93  */
94 
95 /*
96  * The above functions are expanded inline in the statically-linked
97  * kernel.  Lock prefixes are generated if an SMP kernel is being
98  * built.
99  *
100  * Kernel modules call real functions which are built into the kernel.
101  * This allows kernel modules to be portable between UP and SMP systems.
102  */
103 #if defined(KLD_MODULE) || !defined(__GNUCLIKE_ASM)
104 #define	ATOMIC_ASM(NAME, TYPE, OP, CONS, V)			\
105 void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v);	\
106 void atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
107 
108 int	atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src);
109 int	atomic_fcmpset_int(volatile u_int *dst, u_int *expect, u_int src);
110 u_int	atomic_fetchadd_int(volatile u_int *p, u_int v);
111 int	atomic_testandset_int(volatile u_int *p, u_int v);
112 int	atomic_testandclear_int(volatile u_int *p, u_int v);
113 void	atomic_thread_fence_acq(void);
114 void	atomic_thread_fence_acq_rel(void);
115 void	atomic_thread_fence_rel(void);
116 void	atomic_thread_fence_seq_cst(void);
117 
118 #define	ATOMIC_LOAD(TYPE)					\
119 u_##TYPE	atomic_load_acq_##TYPE(volatile u_##TYPE *p)
120 #define	ATOMIC_STORE(TYPE)					\
121 void		atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
122 
123 int		atomic_cmpset_64(volatile uint64_t *, uint64_t, uint64_t);
124 uint64_t	atomic_load_acq_64(volatile uint64_t *);
125 void		atomic_store_rel_64(volatile uint64_t *, uint64_t);
126 uint64_t	atomic_swap_64(volatile uint64_t *, uint64_t);
127 
128 #else /* !KLD_MODULE && __GNUCLIKE_ASM */
129 
130 /*
131  * For userland, always use lock prefixes so that the binaries will run
132  * on both SMP and !SMP systems.
133  */
134 #if defined(SMP) || !defined(_KERNEL)
135 #define	MPLOCKED	"lock ; "
136 #else
137 #define	MPLOCKED
138 #endif
139 
140 /*
141  * The assembly is volatilized to avoid code chunk removal by the compiler.
142  * GCC aggressively reorders operations and memory clobbering is necessary
143  * in order to avoid that for memory barriers.
144  */
145 #define	ATOMIC_ASM(NAME, TYPE, OP, CONS, V)		\
146 static __inline void					\
147 atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
148 {							\
149 	__asm __volatile(MPLOCKED OP			\
150 	: "+m" (*p)					\
151 	: CONS (V)					\
152 	: "cc");					\
153 }							\
154 							\
155 static __inline void					\
156 atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
157 {							\
158 	__asm __volatile(MPLOCKED OP			\
159 	: "+m" (*p)					\
160 	: CONS (V)					\
161 	: "memory", "cc");				\
162 }							\
163 struct __hack
164 
165 /*
166  * Atomic compare and set, used by the mutex functions
167  *
168  * if (*dst == expect) *dst = src (all 32 bit words)
169  *
170  * Returns 0 on failure, non-zero on success
171  */
172 
173 #ifdef CPU_DISABLE_CMPXCHG
174 
175 static __inline int
176 atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src)
177 {
178 	u_char res;
179 
180 	__asm __volatile(
181 	"	pushfl ;		"
182 	"	cli ;			"
183 	"	cmpl	%3,%1 ;		"
184 	"	jne	1f ;		"
185 	"	movl	%2,%1 ;		"
186 	"1:				"
187 	"       sete	%0 ;		"
188 	"	popfl ;			"
189 	"# atomic_cmpset_int"
190 	: "=q" (res),			/* 0 */
191 	  "+m" (*dst)			/* 1 */
192 	: "r" (src),			/* 2 */
193 	  "r" (expect)			/* 3 */
194 	: "memory");
195 	return (res);
196 }
197 
198 #else /* !CPU_DISABLE_CMPXCHG */
199 
200 static __inline int
201 atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src)
202 {
203 	u_char res;
204 
205 	__asm __volatile(
206 	"	" MPLOCKED "		"
207 	"	cmpxchgl %3,%1 ;	"
208 	"       sete	%0 ;		"
209 	"# atomic_cmpset_int"
210 	: "=q" (res),			/* 0 */
211 	  "+m" (*dst),			/* 1 */
212 	  "+a" (expect)			/* 2 */
213 	: "r" (src)			/* 3 */
214 	: "memory", "cc");
215 	return (res);
216 }
217 
218 static __inline int
219 atomic_fcmpset_int(volatile u_int *dst, u_int *expect, u_int src)
220 {
221 	u_char res;
222 
223 	__asm __volatile(
224 	"	" MPLOCKED "		"
225 	"	cmpxchgl %3,%1 ;	"
226 	"       sete	%0 ;		"
227 	"# atomic_cmpset_int"
228 	: "=q" (res),			/* 0 */
229 	  "+m" (*dst),			/* 1 */
230 	  "+a" (*expect)		/* 2 */
231 	: "r" (src)			/* 3 */
232 	: "memory", "cc");
233 	return (res);
234 }
235 
236 #endif /* CPU_DISABLE_CMPXCHG */
237 
238 /*
239  * Atomically add the value of v to the integer pointed to by p and return
240  * the previous value of *p.
241  */
242 static __inline u_int
243 atomic_fetchadd_int(volatile u_int *p, u_int v)
244 {
245 
246 	__asm __volatile(
247 	"	" MPLOCKED "		"
248 	"	xaddl	%0,%1 ;		"
249 	"# atomic_fetchadd_int"
250 	: "+r" (v),			/* 0 */
251 	  "+m" (*p)			/* 1 */
252 	: : "cc");
253 	return (v);
254 }
255 
256 static __inline int
257 atomic_testandset_int(volatile u_int *p, u_int v)
258 {
259 	u_char res;
260 
261 	__asm __volatile(
262 	"	" MPLOCKED "		"
263 	"	btsl	%2,%1 ;		"
264 	"	setc	%0 ;		"
265 	"# atomic_testandset_int"
266 	: "=q" (res),			/* 0 */
267 	  "+m" (*p)			/* 1 */
268 	: "Ir" (v & 0x1f)		/* 2 */
269 	: "cc");
270 	return (res);
271 }
272 
273 static __inline int
274 atomic_testandclear_int(volatile u_int *p, u_int v)
275 {
276 	u_char res;
277 
278 	__asm __volatile(
279 	"	" MPLOCKED "		"
280 	"	btrl	%2,%1 ;		"
281 	"	setc	%0 ;		"
282 	"# atomic_testandclear_int"
283 	: "=q" (res),			/* 0 */
284 	  "+m" (*p)			/* 1 */
285 	: "Ir" (v & 0x1f)		/* 2 */
286 	: "cc");
287 	return (res);
288 }
289 
290 /*
291  * We assume that a = b will do atomic loads and stores.  Due to the
292  * IA32 memory model, a simple store guarantees release semantics.
293  *
294  * However, a load may pass a store if they are performed on distinct
295  * addresses, so we need Store/Load barrier for sequentially
296  * consistent fences in SMP kernels.  We use "lock addl $0,mem" for a
297  * Store/Load barrier, as recommended by the AMD Software Optimization
298  * Guide, and not mfence.  In the kernel, we use a private per-cpu
299  * cache line for "mem", to avoid introducing false data
300  * dependencies.  In user space, we use the word at the top of the
301  * stack.
302  *
303  * For UP kernels, however, the memory of the single processor is
304  * always consistent, so we only need to stop the compiler from
305  * reordering accesses in a way that violates the semantics of acquire
306  * and release.
307  */
308 
309 #if defined(_KERNEL)
310 #if defined(SMP)
311 #define	__storeload_barrier()	__mbk()
312 #else /* _KERNEL && UP */
313 #define	__storeload_barrier()	__compiler_membar()
314 #endif /* SMP */
315 #else /* !_KERNEL */
316 #define	__storeload_barrier()	__mbu()
317 #endif /* _KERNEL*/
318 
319 #define	ATOMIC_LOAD(TYPE)					\
320 static __inline u_##TYPE					\
321 atomic_load_acq_##TYPE(volatile u_##TYPE *p)			\
322 {								\
323 	u_##TYPE res;						\
324 								\
325 	res = *p;						\
326 	__compiler_membar();					\
327 	return (res);						\
328 }								\
329 struct __hack
330 
331 #define	ATOMIC_STORE(TYPE)					\
332 static __inline void						\
333 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)	\
334 {								\
335 								\
336 	__compiler_membar();					\
337 	*p = v;							\
338 }								\
339 struct __hack
340 
341 static __inline void
342 atomic_thread_fence_acq(void)
343 {
344 
345 	__compiler_membar();
346 }
347 
348 static __inline void
349 atomic_thread_fence_rel(void)
350 {
351 
352 	__compiler_membar();
353 }
354 
355 static __inline void
356 atomic_thread_fence_acq_rel(void)
357 {
358 
359 	__compiler_membar();
360 }
361 
362 static __inline void
363 atomic_thread_fence_seq_cst(void)
364 {
365 
366 	__storeload_barrier();
367 }
368 
369 #ifdef _KERNEL
370 
371 #ifdef WANT_FUNCTIONS
372 int		atomic_cmpset_64_i386(volatile uint64_t *, uint64_t, uint64_t);
373 int		atomic_cmpset_64_i586(volatile uint64_t *, uint64_t, uint64_t);
374 uint64_t	atomic_load_acq_64_i386(volatile uint64_t *);
375 uint64_t	atomic_load_acq_64_i586(volatile uint64_t *);
376 void		atomic_store_rel_64_i386(volatile uint64_t *, uint64_t);
377 void		atomic_store_rel_64_i586(volatile uint64_t *, uint64_t);
378 uint64_t	atomic_swap_64_i386(volatile uint64_t *, uint64_t);
379 uint64_t	atomic_swap_64_i586(volatile uint64_t *, uint64_t);
380 #endif
381 
382 /* I486 does not support SMP or CMPXCHG8B. */
383 static __inline int
384 atomic_cmpset_64_i386(volatile uint64_t *dst, uint64_t expect, uint64_t src)
385 {
386 	volatile uint32_t *p;
387 	u_char res;
388 
389 	p = (volatile uint32_t *)dst;
390 	__asm __volatile(
391 	"	pushfl ;		"
392 	"	cli ;			"
393 	"	xorl	%1,%%eax ;	"
394 	"	xorl	%2,%%edx ;	"
395 	"	orl	%%edx,%%eax ;	"
396 	"	jne	1f ;		"
397 	"	movl	%4,%1 ;		"
398 	"	movl	%5,%2 ;		"
399 	"1:				"
400 	"	sete	%3 ;		"
401 	"	popfl"
402 	: "+A" (expect),		/* 0 */
403 	  "+m" (*p),			/* 1 */
404 	  "+m" (*(p + 1)),		/* 2 */
405 	  "=q" (res)			/* 3 */
406 	: "r" ((uint32_t)src),		/* 4 */
407 	  "r" ((uint32_t)(src >> 32))	/* 5 */
408 	: "memory", "cc");
409 	return (res);
410 }
411 
412 static __inline uint64_t
413 atomic_load_acq_64_i386(volatile uint64_t *p)
414 {
415 	volatile uint32_t *q;
416 	uint64_t res;
417 
418 	q = (volatile uint32_t *)p;
419 	__asm __volatile(
420 	"	pushfl ;		"
421 	"	cli ;			"
422 	"	movl	%1,%%eax ;	"
423 	"	movl	%2,%%edx ;	"
424 	"	popfl"
425 	: "=&A" (res)			/* 0 */
426 	: "m" (*q),			/* 1 */
427 	  "m" (*(q + 1))		/* 2 */
428 	: "memory");
429 	return (res);
430 }
431 
432 static __inline void
433 atomic_store_rel_64_i386(volatile uint64_t *p, uint64_t v)
434 {
435 	volatile uint32_t *q;
436 
437 	q = (volatile uint32_t *)p;
438 	__asm __volatile(
439 	"	pushfl ;		"
440 	"	cli ;			"
441 	"	movl	%%eax,%0 ;	"
442 	"	movl	%%edx,%1 ;	"
443 	"	popfl"
444 	: "=m" (*q),			/* 0 */
445 	  "=m" (*(q + 1))		/* 1 */
446 	: "A" (v)			/* 2 */
447 	: "memory");
448 }
449 
450 static __inline uint64_t
451 atomic_swap_64_i386(volatile uint64_t *p, uint64_t v)
452 {
453 	volatile uint32_t *q;
454 	uint64_t res;
455 
456 	q = (volatile uint32_t *)p;
457 	__asm __volatile(
458 	"	pushfl ;		"
459 	"	cli ;			"
460 	"	movl	%1,%%eax ;	"
461 	"	movl	%2,%%edx ;	"
462 	"	movl	%4,%2 ;		"
463 	"	movl	%3,%1 ;		"
464 	"	popfl"
465 	: "=&A" (res),			/* 0 */
466 	  "+m" (*q),			/* 1 */
467 	  "+m" (*(q + 1))		/* 2 */
468 	: "r" ((uint32_t)v),		/* 3 */
469 	  "r" ((uint32_t)(v >> 32)));	/* 4 */
470 	return (res);
471 }
472 
473 static __inline int
474 atomic_cmpset_64_i586(volatile uint64_t *dst, uint64_t expect, uint64_t src)
475 {
476 	u_char res;
477 
478 	__asm __volatile(
479 	"	" MPLOCKED "		"
480 	"	cmpxchg8b %1 ;		"
481 	"	sete	%0"
482 	: "=q" (res),			/* 0 */
483 	  "+m" (*dst),			/* 1 */
484 	  "+A" (expect)			/* 2 */
485 	: "b" ((uint32_t)src),		/* 3 */
486 	  "c" ((uint32_t)(src >> 32))	/* 4 */
487 	: "memory", "cc");
488 	return (res);
489 }
490 
491 static __inline uint64_t
492 atomic_load_acq_64_i586(volatile uint64_t *p)
493 {
494 	uint64_t res;
495 
496 	__asm __volatile(
497 	"	movl	%%ebx,%%eax ;	"
498 	"	movl	%%ecx,%%edx ;	"
499 	"	" MPLOCKED "		"
500 	"	cmpxchg8b %1"
501 	: "=&A" (res),			/* 0 */
502 	  "+m" (*p)			/* 1 */
503 	: : "memory", "cc");
504 	return (res);
505 }
506 
507 static __inline void
508 atomic_store_rel_64_i586(volatile uint64_t *p, uint64_t v)
509 {
510 
511 	__asm __volatile(
512 	"	movl	%%eax,%%ebx ;	"
513 	"	movl	%%edx,%%ecx ;	"
514 	"1:				"
515 	"	" MPLOCKED "		"
516 	"	cmpxchg8b %0 ;		"
517 	"	jne	1b"
518 	: "+m" (*p),			/* 0 */
519 	  "+A" (v)			/* 1 */
520 	: : "ebx", "ecx", "memory", "cc");
521 }
522 
523 static __inline uint64_t
524 atomic_swap_64_i586(volatile uint64_t *p, uint64_t v)
525 {
526 
527 	__asm __volatile(
528 	"	movl	%%eax,%%ebx ;	"
529 	"	movl	%%edx,%%ecx ;	"
530 	"1:				"
531 	"	" MPLOCKED "		"
532 	"	cmpxchg8b %0 ;		"
533 	"	jne	1b"
534 	: "+m" (*p),			/* 0 */
535 	  "+A" (v)			/* 1 */
536 	: : "ebx", "ecx", "memory", "cc");
537 	return (v);
538 }
539 
540 static __inline int
541 atomic_cmpset_64(volatile uint64_t *dst, uint64_t expect, uint64_t src)
542 {
543 
544 	if ((cpu_feature & CPUID_CX8) == 0)
545 		return (atomic_cmpset_64_i386(dst, expect, src));
546 	else
547 		return (atomic_cmpset_64_i586(dst, expect, src));
548 }
549 
550 static __inline uint64_t
551 atomic_load_acq_64(volatile uint64_t *p)
552 {
553 
554 	if ((cpu_feature & CPUID_CX8) == 0)
555 		return (atomic_load_acq_64_i386(p));
556 	else
557 		return (atomic_load_acq_64_i586(p));
558 }
559 
560 static __inline void
561 atomic_store_rel_64(volatile uint64_t *p, uint64_t v)
562 {
563 
564 	if ((cpu_feature & CPUID_CX8) == 0)
565 		atomic_store_rel_64_i386(p, v);
566 	else
567 		atomic_store_rel_64_i586(p, v);
568 }
569 
570 static __inline uint64_t
571 atomic_swap_64(volatile uint64_t *p, uint64_t v)
572 {
573 
574 	if ((cpu_feature & CPUID_CX8) == 0)
575 		return (atomic_swap_64_i386(p, v));
576 	else
577 		return (atomic_swap_64_i586(p, v));
578 }
579 
580 #endif /* _KERNEL */
581 
582 #endif /* KLD_MODULE || !__GNUCLIKE_ASM */
583 
584 ATOMIC_ASM(set,	     char,  "orb %b1,%0",  "iq",  v);
585 ATOMIC_ASM(clear,    char,  "andb %b1,%0", "iq", ~v);
586 ATOMIC_ASM(add,	     char,  "addb %b1,%0", "iq",  v);
587 ATOMIC_ASM(subtract, char,  "subb %b1,%0", "iq",  v);
588 
589 ATOMIC_ASM(set,	     short, "orw %w1,%0",  "ir",  v);
590 ATOMIC_ASM(clear,    short, "andw %w1,%0", "ir", ~v);
591 ATOMIC_ASM(add,	     short, "addw %w1,%0", "ir",  v);
592 ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir",  v);
593 
594 ATOMIC_ASM(set,	     int,   "orl %1,%0",   "ir",  v);
595 ATOMIC_ASM(clear,    int,   "andl %1,%0",  "ir", ~v);
596 ATOMIC_ASM(add,	     int,   "addl %1,%0",  "ir",  v);
597 ATOMIC_ASM(subtract, int,   "subl %1,%0",  "ir",  v);
598 
599 ATOMIC_ASM(set,	     long,  "orl %1,%0",   "ir",  v);
600 ATOMIC_ASM(clear,    long,  "andl %1,%0",  "ir", ~v);
601 ATOMIC_ASM(add,	     long,  "addl %1,%0",  "ir",  v);
602 ATOMIC_ASM(subtract, long,  "subl %1,%0",  "ir",  v);
603 
604 #define	ATOMIC_LOADSTORE(TYPE)				\
605 	ATOMIC_LOAD(TYPE);				\
606 	ATOMIC_STORE(TYPE)
607 
608 ATOMIC_LOADSTORE(char);
609 ATOMIC_LOADSTORE(short);
610 ATOMIC_LOADSTORE(int);
611 ATOMIC_LOADSTORE(long);
612 
613 #undef ATOMIC_ASM
614 #undef ATOMIC_LOAD
615 #undef ATOMIC_STORE
616 #undef ATOMIC_LOADSTORE
617 
618 #ifndef WANT_FUNCTIONS
619 
620 static __inline int
621 atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src)
622 {
623 
624 	return (atomic_cmpset_int((volatile u_int *)dst, (u_int)expect,
625 	    (u_int)src));
626 }
627 
628 static __inline u_long
629 atomic_fetchadd_long(volatile u_long *p, u_long v)
630 {
631 
632 	return (atomic_fetchadd_int((volatile u_int *)p, (u_int)v));
633 }
634 
635 static __inline int
636 atomic_testandset_long(volatile u_long *p, u_int v)
637 {
638 
639 	return (atomic_testandset_int((volatile u_int *)p, v));
640 }
641 
642 static __inline int
643 atomic_testandclear_long(volatile u_long *p, u_int v)
644 {
645 
646 	return (atomic_testandclear_int((volatile u_int *)p, v));
647 }
648 
649 /* Read the current value and store a new value in the destination. */
650 #ifdef __GNUCLIKE_ASM
651 
652 static __inline u_int
653 atomic_swap_int(volatile u_int *p, u_int v)
654 {
655 
656 	__asm __volatile(
657 	"	xchgl	%1,%0 ;		"
658 	"# atomic_swap_int"
659 	: "+r" (v),			/* 0 */
660 	  "+m" (*p));			/* 1 */
661 	return (v);
662 }
663 
664 static __inline u_long
665 atomic_swap_long(volatile u_long *p, u_long v)
666 {
667 
668 	return (atomic_swap_int((volatile u_int *)p, (u_int)v));
669 }
670 
671 #else /* !__GNUCLIKE_ASM */
672 
673 u_int	atomic_swap_int(volatile u_int *p, u_int v);
674 u_long	atomic_swap_long(volatile u_long *p, u_long v);
675 
676 #endif /* __GNUCLIKE_ASM */
677 
678 #define	atomic_set_acq_char		atomic_set_barr_char
679 #define	atomic_set_rel_char		atomic_set_barr_char
680 #define	atomic_clear_acq_char		atomic_clear_barr_char
681 #define	atomic_clear_rel_char		atomic_clear_barr_char
682 #define	atomic_add_acq_char		atomic_add_barr_char
683 #define	atomic_add_rel_char		atomic_add_barr_char
684 #define	atomic_subtract_acq_char	atomic_subtract_barr_char
685 #define	atomic_subtract_rel_char	atomic_subtract_barr_char
686 
687 #define	atomic_set_acq_short		atomic_set_barr_short
688 #define	atomic_set_rel_short		atomic_set_barr_short
689 #define	atomic_clear_acq_short		atomic_clear_barr_short
690 #define	atomic_clear_rel_short		atomic_clear_barr_short
691 #define	atomic_add_acq_short		atomic_add_barr_short
692 #define	atomic_add_rel_short		atomic_add_barr_short
693 #define	atomic_subtract_acq_short	atomic_subtract_barr_short
694 #define	atomic_subtract_rel_short	atomic_subtract_barr_short
695 
696 #define	atomic_set_acq_int		atomic_set_barr_int
697 #define	atomic_set_rel_int		atomic_set_barr_int
698 #define	atomic_clear_acq_int		atomic_clear_barr_int
699 #define	atomic_clear_rel_int		atomic_clear_barr_int
700 #define	atomic_add_acq_int		atomic_add_barr_int
701 #define	atomic_add_rel_int		atomic_add_barr_int
702 #define	atomic_subtract_acq_int		atomic_subtract_barr_int
703 #define	atomic_subtract_rel_int		atomic_subtract_barr_int
704 #define	atomic_cmpset_acq_int		atomic_cmpset_int
705 #define	atomic_cmpset_rel_int		atomic_cmpset_int
706 #define	atomic_fcmpset_acq_int		atomic_fcmpset_int
707 #define	atomic_fcmpset_rel_int		atomic_fcmpset_int
708 
709 #define	atomic_set_acq_long		atomic_set_barr_long
710 #define	atomic_set_rel_long		atomic_set_barr_long
711 #define	atomic_clear_acq_long		atomic_clear_barr_long
712 #define	atomic_clear_rel_long		atomic_clear_barr_long
713 #define	atomic_add_acq_long		atomic_add_barr_long
714 #define	atomic_add_rel_long		atomic_add_barr_long
715 #define	atomic_subtract_acq_long	atomic_subtract_barr_long
716 #define	atomic_subtract_rel_long	atomic_subtract_barr_long
717 #define	atomic_cmpset_acq_long		atomic_cmpset_long
718 #define	atomic_cmpset_rel_long		atomic_cmpset_long
719 #define	atomic_fcmpset_acq_long		atomic_fcmpset_long
720 #define	atomic_fcmpset_rel_long		atomic_fcmpset_long
721 
722 #define	atomic_readandclear_int(p)	atomic_swap_int(p, 0)
723 #define	atomic_readandclear_long(p)	atomic_swap_long(p, 0)
724 
725 /* Operations on 8-bit bytes. */
726 #define	atomic_set_8		atomic_set_char
727 #define	atomic_set_acq_8	atomic_set_acq_char
728 #define	atomic_set_rel_8	atomic_set_rel_char
729 #define	atomic_clear_8		atomic_clear_char
730 #define	atomic_clear_acq_8	atomic_clear_acq_char
731 #define	atomic_clear_rel_8	atomic_clear_rel_char
732 #define	atomic_add_8		atomic_add_char
733 #define	atomic_add_acq_8	atomic_add_acq_char
734 #define	atomic_add_rel_8	atomic_add_rel_char
735 #define	atomic_subtract_8	atomic_subtract_char
736 #define	atomic_subtract_acq_8	atomic_subtract_acq_char
737 #define	atomic_subtract_rel_8	atomic_subtract_rel_char
738 #define	atomic_load_acq_8	atomic_load_acq_char
739 #define	atomic_store_rel_8	atomic_store_rel_char
740 
741 /* Operations on 16-bit words. */
742 #define	atomic_set_16		atomic_set_short
743 #define	atomic_set_acq_16	atomic_set_acq_short
744 #define	atomic_set_rel_16	atomic_set_rel_short
745 #define	atomic_clear_16		atomic_clear_short
746 #define	atomic_clear_acq_16	atomic_clear_acq_short
747 #define	atomic_clear_rel_16	atomic_clear_rel_short
748 #define	atomic_add_16		atomic_add_short
749 #define	atomic_add_acq_16	atomic_add_acq_short
750 #define	atomic_add_rel_16	atomic_add_rel_short
751 #define	atomic_subtract_16	atomic_subtract_short
752 #define	atomic_subtract_acq_16	atomic_subtract_acq_short
753 #define	atomic_subtract_rel_16	atomic_subtract_rel_short
754 #define	atomic_load_acq_16	atomic_load_acq_short
755 #define	atomic_store_rel_16	atomic_store_rel_short
756 
757 /* Operations on 32-bit double words. */
758 #define	atomic_set_32		atomic_set_int
759 #define	atomic_set_acq_32	atomic_set_acq_int
760 #define	atomic_set_rel_32	atomic_set_rel_int
761 #define	atomic_clear_32		atomic_clear_int
762 #define	atomic_clear_acq_32	atomic_clear_acq_int
763 #define	atomic_clear_rel_32	atomic_clear_rel_int
764 #define	atomic_add_32		atomic_add_int
765 #define	atomic_add_acq_32	atomic_add_acq_int
766 #define	atomic_add_rel_32	atomic_add_rel_int
767 #define	atomic_subtract_32	atomic_subtract_int
768 #define	atomic_subtract_acq_32	atomic_subtract_acq_int
769 #define	atomic_subtract_rel_32	atomic_subtract_rel_int
770 #define	atomic_load_acq_32	atomic_load_acq_int
771 #define	atomic_store_rel_32	atomic_store_rel_int
772 #define	atomic_cmpset_32	atomic_cmpset_int
773 #define	atomic_cmpset_acq_32	atomic_cmpset_acq_int
774 #define	atomic_cmpset_rel_32	atomic_cmpset_rel_int
775 #define	atomic_fcmpset_32	atomic_fcmpset_int
776 #define	atomic_fcmpset_acq_32	atomic_fcmpset_acq_int
777 #define	atomic_fcmpset_rel_32	atomic_fcmpset_rel_int
778 #define	atomic_swap_32		atomic_swap_int
779 #define	atomic_readandclear_32	atomic_readandclear_int
780 #define	atomic_fetchadd_32	atomic_fetchadd_int
781 #define	atomic_testandset_32	atomic_testandset_int
782 #define	atomic_testandclear_32	atomic_testandclear_int
783 
784 /* Operations on pointers. */
785 #define	atomic_set_ptr(p, v) \
786 	atomic_set_int((volatile u_int *)(p), (u_int)(v))
787 #define	atomic_set_acq_ptr(p, v) \
788 	atomic_set_acq_int((volatile u_int *)(p), (u_int)(v))
789 #define	atomic_set_rel_ptr(p, v) \
790 	atomic_set_rel_int((volatile u_int *)(p), (u_int)(v))
791 #define	atomic_clear_ptr(p, v) \
792 	atomic_clear_int((volatile u_int *)(p), (u_int)(v))
793 #define	atomic_clear_acq_ptr(p, v) \
794 	atomic_clear_acq_int((volatile u_int *)(p), (u_int)(v))
795 #define	atomic_clear_rel_ptr(p, v) \
796 	atomic_clear_rel_int((volatile u_int *)(p), (u_int)(v))
797 #define	atomic_add_ptr(p, v) \
798 	atomic_add_int((volatile u_int *)(p), (u_int)(v))
799 #define	atomic_add_acq_ptr(p, v) \
800 	atomic_add_acq_int((volatile u_int *)(p), (u_int)(v))
801 #define	atomic_add_rel_ptr(p, v) \
802 	atomic_add_rel_int((volatile u_int *)(p), (u_int)(v))
803 #define	atomic_subtract_ptr(p, v) \
804 	atomic_subtract_int((volatile u_int *)(p), (u_int)(v))
805 #define	atomic_subtract_acq_ptr(p, v) \
806 	atomic_subtract_acq_int((volatile u_int *)(p), (u_int)(v))
807 #define	atomic_subtract_rel_ptr(p, v) \
808 	atomic_subtract_rel_int((volatile u_int *)(p), (u_int)(v))
809 #define	atomic_load_acq_ptr(p) \
810 	atomic_load_acq_int((volatile u_int *)(p))
811 #define	atomic_store_rel_ptr(p, v) \
812 	atomic_store_rel_int((volatile u_int *)(p), (v))
813 #define	atomic_cmpset_ptr(dst, old, new) \
814 	atomic_cmpset_int((volatile u_int *)(dst), (u_int)(old), (u_int)(new))
815 #define	atomic_cmpset_acq_ptr(dst, old, new) \
816 	atomic_cmpset_acq_int((volatile u_int *)(dst), (u_int)(old), \
817 	    (u_int)(new))
818 #define	atomic_cmpset_rel_ptr(dst, old, new) \
819 	atomic_cmpset_rel_int((volatile u_int *)(dst), (u_int)(old), \
820 	    (u_int)(new))
821 #define	atomic_fcmpset_ptr(dst, old, new) \
822 	atomic_fcmpset_int((volatile u_int *)(dst), (u_int *)(old), (u_int)(new))
823 #define	atomic_fcmpset_acq_ptr(dst, old, new) \
824 	atomic_fcmpset_acq_int((volatile u_int *)(dst), (u_int *)(old), \
825 	    (u_int)(new))
826 #define	atomic_fcmpset_rel_ptr(dst, old, new) \
827 	atomic_fcmpset_rel_int((volatile u_int *)(dst), (u_int *)(old), \
828 	    (u_int)(new))
829 #define	atomic_swap_ptr(p, v) \
830 	atomic_swap_int((volatile u_int *)(p), (u_int)(v))
831 #define	atomic_readandclear_ptr(p) \
832 	atomic_readandclear_int((volatile u_int *)(p))
833 
834 #endif /* !WANT_FUNCTIONS */
835 
836 #if defined(_KERNEL)
837 #define	mb()	__mbk()
838 #define	wmb()	__mbk()
839 #define	rmb()	__mbk()
840 #else
841 #define	mb()	__mbu()
842 #define	wmb()	__mbu()
843 #define	rmb()	__mbu()
844 #endif
845 
846 #endif /* !_MACHINE_ATOMIC_H_ */
847