xref: /freebsd/sys/i386/include/atomic.h (revision d06955f9bdb1416d9196043ed781f9b36dae9adc)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1998 Doug Rabson
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD$
29  */
30 #ifndef _MACHINE_ATOMIC_H_
31 #define	_MACHINE_ATOMIC_H_
32 
33 #ifndef _SYS_CDEFS_H_
34 #error this file needs sys/cdefs.h as a prerequisite
35 #endif
36 
37 #include <sys/atomic_common.h>
38 
39 #ifdef _KERNEL
40 #include <machine/md_var.h>
41 #include <machine/specialreg.h>
42 #endif
43 
44 #ifndef __OFFSETOF_MONITORBUF
45 /*
46  * __OFFSETOF_MONITORBUF == __pcpu_offset(pc_monitorbuf).
47  *
48  * The open-coded number is used instead of the symbolic expression to
49  * avoid a dependency on sys/pcpu.h in machine/atomic.h consumers.
50  * An assertion in i386/vm_machdep.c ensures that the value is correct.
51  */
52 #define	__OFFSETOF_MONITORBUF	0x80
53 
54 static __inline void
55 __mbk(void)
56 {
57 
58 	__asm __volatile("lock; addl $0,%%fs:%0"
59 	    : "+m" (*(u_int *)__OFFSETOF_MONITORBUF) : : "memory", "cc");
60 }
61 
62 static __inline void
63 __mbu(void)
64 {
65 
66 	__asm __volatile("lock; addl $0,(%%esp)" : : : "memory", "cc");
67 }
68 #endif
69 
70 /*
71  * Various simple operations on memory, each of which is atomic in the
72  * presence of interrupts and multiple processors.
73  *
74  * atomic_set_char(P, V)	(*(u_char *)(P) |= (V))
75  * atomic_clear_char(P, V)	(*(u_char *)(P) &= ~(V))
76  * atomic_add_char(P, V)	(*(u_char *)(P) += (V))
77  * atomic_subtract_char(P, V)	(*(u_char *)(P) -= (V))
78  *
79  * atomic_set_short(P, V)	(*(u_short *)(P) |= (V))
80  * atomic_clear_short(P, V)	(*(u_short *)(P) &= ~(V))
81  * atomic_add_short(P, V)	(*(u_short *)(P) += (V))
82  * atomic_subtract_short(P, V)	(*(u_short *)(P) -= (V))
83  *
84  * atomic_set_int(P, V)		(*(u_int *)(P) |= (V))
85  * atomic_clear_int(P, V)	(*(u_int *)(P) &= ~(V))
86  * atomic_add_int(P, V)		(*(u_int *)(P) += (V))
87  * atomic_subtract_int(P, V)	(*(u_int *)(P) -= (V))
88  * atomic_swap_int(P, V)	(return (*(u_int *)(P)); *(u_int *)(P) = (V);)
89  * atomic_readandclear_int(P)	(return (*(u_int *)(P)); *(u_int *)(P) = 0;)
90  *
91  * atomic_set_long(P, V)	(*(u_long *)(P) |= (V))
92  * atomic_clear_long(P, V)	(*(u_long *)(P) &= ~(V))
93  * atomic_add_long(P, V)	(*(u_long *)(P) += (V))
94  * atomic_subtract_long(P, V)	(*(u_long *)(P) -= (V))
95  * atomic_swap_long(P, V)	(return (*(u_long *)(P)); *(u_long *)(P) = (V);)
96  * atomic_readandclear_long(P)	(return (*(u_long *)(P)); *(u_long *)(P) = 0;)
97  */
98 
99 /*
100  * The above functions are expanded inline in the statically-linked
101  * kernel.  Lock prefixes are generated if an SMP kernel is being
102  * built.
103  *
104  * Kernel modules call real functions which are built into the kernel.
105  * This allows kernel modules to be portable between UP and SMP systems.
106  */
107 #if defined(KLD_MODULE) || !defined(__GNUCLIKE_ASM)
108 #define	ATOMIC_ASM(NAME, TYPE, OP, CONS, V)			\
109 void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v);	\
110 void atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
111 
112 int	atomic_cmpset_char(volatile u_char *dst, u_char expect, u_char src);
113 int	atomic_cmpset_short(volatile u_short *dst, u_short expect, u_short src);
114 int	atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src);
115 int	atomic_fcmpset_char(volatile u_char *dst, u_char *expect, u_char src);
116 int	atomic_fcmpset_short(volatile u_short *dst, u_short *expect,
117 	    u_short src);
118 int	atomic_fcmpset_int(volatile u_int *dst, u_int *expect, u_int src);
119 u_int	atomic_fetchadd_int(volatile u_int *p, u_int v);
120 int	atomic_testandset_int(volatile u_int *p, u_int v);
121 int	atomic_testandclear_int(volatile u_int *p, u_int v);
122 void	atomic_thread_fence_acq(void);
123 void	atomic_thread_fence_acq_rel(void);
124 void	atomic_thread_fence_rel(void);
125 void	atomic_thread_fence_seq_cst(void);
126 
127 #define	ATOMIC_LOAD(TYPE)					\
128 u_##TYPE	atomic_load_acq_##TYPE(volatile u_##TYPE *p)
129 #define	ATOMIC_STORE(TYPE)					\
130 void		atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
131 
132 int		atomic_cmpset_64(volatile uint64_t *, uint64_t, uint64_t);
133 uint64_t	atomic_load_acq_64(volatile uint64_t *);
134 void		atomic_store_rel_64(volatile uint64_t *, uint64_t);
135 uint64_t	atomic_swap_64(volatile uint64_t *, uint64_t);
136 uint64_t	atomic_fetchadd_64(volatile uint64_t *, uint64_t);
137 
138 #else /* !KLD_MODULE && __GNUCLIKE_ASM */
139 
140 /*
141  * For userland, always use lock prefixes so that the binaries will run
142  * on both SMP and !SMP systems.
143  */
144 #if defined(SMP) || !defined(_KERNEL)
145 #define	MPLOCKED	"lock ; "
146 #else
147 #define	MPLOCKED
148 #endif
149 
150 /*
151  * The assembly is volatilized to avoid code chunk removal by the compiler.
152  * GCC aggressively reorders operations and memory clobbering is necessary
153  * in order to avoid that for memory barriers.
154  */
155 #define	ATOMIC_ASM(NAME, TYPE, OP, CONS, V)		\
156 static __inline void					\
157 atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
158 {							\
159 	__asm __volatile(MPLOCKED OP			\
160 	: "+m" (*p)					\
161 	: CONS (V)					\
162 	: "cc");					\
163 }							\
164 							\
165 static __inline void					\
166 atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
167 {							\
168 	__asm __volatile(MPLOCKED OP			\
169 	: "+m" (*p)					\
170 	: CONS (V)					\
171 	: "memory", "cc");				\
172 }							\
173 struct __hack
174 
175 /*
176  * Atomic compare and set, used by the mutex functions.
177  *
178  * cmpset:
179  *	if (*dst == expect)
180  *		*dst = src
181  *
182  * fcmpset:
183  *	if (*dst == *expect)
184  *		*dst = src
185  *	else
186  *		*expect = *dst
187  *
188  * Returns 0 on failure, non-zero on success.
189  */
190 #define	ATOMIC_CMPSET(TYPE, CONS)			\
191 static __inline int					\
192 atomic_cmpset_##TYPE(volatile u_##TYPE *dst, u_##TYPE expect, u_##TYPE src) \
193 {							\
194 	u_char res;					\
195 							\
196 	__asm __volatile(				\
197 	"	" MPLOCKED "		"		\
198 	"	cmpxchg	%3,%1 ;		"		\
199 	"	sete	%0 ;		"		\
200 	"# atomic_cmpset_" #TYPE "	"		\
201 	: "=q" (res),			/* 0 */		\
202 	  "+m" (*dst),			/* 1 */		\
203 	  "+a" (expect)			/* 2 */		\
204 	: CONS (src)			/* 3 */		\
205 	: "memory", "cc");				\
206 	return (res);					\
207 }							\
208 							\
209 static __inline int					\
210 atomic_fcmpset_##TYPE(volatile u_##TYPE *dst, u_##TYPE *expect, u_##TYPE src) \
211 {							\
212 	u_char res;					\
213 							\
214 	__asm __volatile(				\
215 	"	" MPLOCKED "		"		\
216 	"	cmpxchg	%3,%1 ;		"		\
217 	"	sete	%0 ;		"		\
218 	"# atomic_fcmpset_" #TYPE "	"		\
219 	: "=q" (res),			/* 0 */		\
220 	  "+m" (*dst),			/* 1 */		\
221 	  "+a" (*expect)		/* 2 */		\
222 	: CONS (src)			/* 3 */		\
223 	: "memory", "cc");				\
224 	return (res);					\
225 }
226 
227 ATOMIC_CMPSET(char, "q");
228 ATOMIC_CMPSET(short, "r");
229 ATOMIC_CMPSET(int, "r");
230 
231 /*
232  * Atomically add the value of v to the integer pointed to by p and return
233  * the previous value of *p.
234  */
235 static __inline u_int
236 atomic_fetchadd_int(volatile u_int *p, u_int v)
237 {
238 
239 	__asm __volatile(
240 	"	" MPLOCKED "		"
241 	"	xaddl	%0,%1 ;		"
242 	"# atomic_fetchadd_int"
243 	: "+r" (v),			/* 0 */
244 	  "+m" (*p)			/* 1 */
245 	: : "cc");
246 	return (v);
247 }
248 
249 static __inline int
250 atomic_testandset_int(volatile u_int *p, u_int v)
251 {
252 	u_char res;
253 
254 	__asm __volatile(
255 	"	" MPLOCKED "		"
256 	"	btsl	%2,%1 ;		"
257 	"	setc	%0 ;		"
258 	"# atomic_testandset_int"
259 	: "=q" (res),			/* 0 */
260 	  "+m" (*p)			/* 1 */
261 	: "Ir" (v & 0x1f)		/* 2 */
262 	: "cc");
263 	return (res);
264 }
265 
266 static __inline int
267 atomic_testandclear_int(volatile u_int *p, u_int v)
268 {
269 	u_char res;
270 
271 	__asm __volatile(
272 	"	" MPLOCKED "		"
273 	"	btrl	%2,%1 ;		"
274 	"	setc	%0 ;		"
275 	"# atomic_testandclear_int"
276 	: "=q" (res),			/* 0 */
277 	  "+m" (*p)			/* 1 */
278 	: "Ir" (v & 0x1f)		/* 2 */
279 	: "cc");
280 	return (res);
281 }
282 
283 /*
284  * We assume that a = b will do atomic loads and stores.  Due to the
285  * IA32 memory model, a simple store guarantees release semantics.
286  *
287  * However, a load may pass a store if they are performed on distinct
288  * addresses, so we need Store/Load barrier for sequentially
289  * consistent fences in SMP kernels.  We use "lock addl $0,mem" for a
290  * Store/Load barrier, as recommended by the AMD Software Optimization
291  * Guide, and not mfence.  In the kernel, we use a private per-cpu
292  * cache line for "mem", to avoid introducing false data
293  * dependencies.  In user space, we use the word at the top of the
294  * stack.
295  *
296  * For UP kernels, however, the memory of the single processor is
297  * always consistent, so we only need to stop the compiler from
298  * reordering accesses in a way that violates the semantics of acquire
299  * and release.
300  */
301 
302 #if defined(_KERNEL)
303 #if defined(SMP)
304 #define	__storeload_barrier()	__mbk()
305 #else /* _KERNEL && UP */
306 #define	__storeload_barrier()	__compiler_membar()
307 #endif /* SMP */
308 #else /* !_KERNEL */
309 #define	__storeload_barrier()	__mbu()
310 #endif /* _KERNEL*/
311 
312 #define	ATOMIC_LOAD(TYPE)					\
313 static __inline u_##TYPE					\
314 atomic_load_acq_##TYPE(volatile u_##TYPE *p)			\
315 {								\
316 	u_##TYPE res;						\
317 								\
318 	res = *p;						\
319 	__compiler_membar();					\
320 	return (res);						\
321 }								\
322 struct __hack
323 
324 #define	ATOMIC_STORE(TYPE)					\
325 static __inline void						\
326 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)	\
327 {								\
328 								\
329 	__compiler_membar();					\
330 	*p = v;							\
331 }								\
332 struct __hack
333 
334 static __inline void
335 atomic_thread_fence_acq(void)
336 {
337 
338 	__compiler_membar();
339 }
340 
341 static __inline void
342 atomic_thread_fence_rel(void)
343 {
344 
345 	__compiler_membar();
346 }
347 
348 static __inline void
349 atomic_thread_fence_acq_rel(void)
350 {
351 
352 	__compiler_membar();
353 }
354 
355 static __inline void
356 atomic_thread_fence_seq_cst(void)
357 {
358 
359 	__storeload_barrier();
360 }
361 
362 #ifdef _KERNEL
363 
364 #ifdef WANT_FUNCTIONS
365 int		atomic_cmpset_64_i386(volatile uint64_t *, uint64_t, uint64_t);
366 int		atomic_cmpset_64_i586(volatile uint64_t *, uint64_t, uint64_t);
367 uint64_t	atomic_load_acq_64_i386(volatile uint64_t *);
368 uint64_t	atomic_load_acq_64_i586(volatile uint64_t *);
369 void		atomic_store_rel_64_i386(volatile uint64_t *, uint64_t);
370 void		atomic_store_rel_64_i586(volatile uint64_t *, uint64_t);
371 uint64_t	atomic_swap_64_i386(volatile uint64_t *, uint64_t);
372 uint64_t	atomic_swap_64_i586(volatile uint64_t *, uint64_t);
373 #endif
374 
375 /* I486 does not support SMP or CMPXCHG8B. */
376 static __inline int
377 atomic_cmpset_64_i386(volatile uint64_t *dst, uint64_t expect, uint64_t src)
378 {
379 	volatile uint32_t *p;
380 	u_char res;
381 
382 	p = (volatile uint32_t *)dst;
383 	__asm __volatile(
384 	"	pushfl ;		"
385 	"	cli ;			"
386 	"	xorl	%1,%%eax ;	"
387 	"	xorl	%2,%%edx ;	"
388 	"	orl	%%edx,%%eax ;	"
389 	"	jne	1f ;		"
390 	"	movl	%4,%1 ;		"
391 	"	movl	%5,%2 ;		"
392 	"1:				"
393 	"	sete	%3 ;		"
394 	"	popfl"
395 	: "+A" (expect),		/* 0 */
396 	  "+m" (*p),			/* 1 */
397 	  "+m" (*(p + 1)),		/* 2 */
398 	  "=q" (res)			/* 3 */
399 	: "r" ((uint32_t)src),		/* 4 */
400 	  "r" ((uint32_t)(src >> 32))	/* 5 */
401 	: "memory", "cc");
402 	return (res);
403 }
404 
405 static __inline uint64_t
406 atomic_load_acq_64_i386(volatile uint64_t *p)
407 {
408 	volatile uint32_t *q;
409 	uint64_t res;
410 
411 	q = (volatile uint32_t *)p;
412 	__asm __volatile(
413 	"	pushfl ;		"
414 	"	cli ;			"
415 	"	movl	%1,%%eax ;	"
416 	"	movl	%2,%%edx ;	"
417 	"	popfl"
418 	: "=&A" (res)			/* 0 */
419 	: "m" (*q),			/* 1 */
420 	  "m" (*(q + 1))		/* 2 */
421 	: "memory");
422 	return (res);
423 }
424 
425 static __inline void
426 atomic_store_rel_64_i386(volatile uint64_t *p, uint64_t v)
427 {
428 	volatile uint32_t *q;
429 
430 	q = (volatile uint32_t *)p;
431 	__asm __volatile(
432 	"	pushfl ;		"
433 	"	cli ;			"
434 	"	movl	%%eax,%0 ;	"
435 	"	movl	%%edx,%1 ;	"
436 	"	popfl"
437 	: "=m" (*q),			/* 0 */
438 	  "=m" (*(q + 1))		/* 1 */
439 	: "A" (v)			/* 2 */
440 	: "memory");
441 }
442 
443 static __inline uint64_t
444 atomic_swap_64_i386(volatile uint64_t *p, uint64_t v)
445 {
446 	volatile uint32_t *q;
447 	uint64_t res;
448 
449 	q = (volatile uint32_t *)p;
450 	__asm __volatile(
451 	"	pushfl ;		"
452 	"	cli ;			"
453 	"	movl	%1,%%eax ;	"
454 	"	movl	%2,%%edx ;	"
455 	"	movl	%4,%2 ;		"
456 	"	movl	%3,%1 ;		"
457 	"	popfl"
458 	: "=&A" (res),			/* 0 */
459 	  "+m" (*q),			/* 1 */
460 	  "+m" (*(q + 1))		/* 2 */
461 	: "r" ((uint32_t)v),		/* 3 */
462 	  "r" ((uint32_t)(v >> 32)));	/* 4 */
463 	return (res);
464 }
465 
466 static __inline int
467 atomic_cmpset_64_i586(volatile uint64_t *dst, uint64_t expect, uint64_t src)
468 {
469 	u_char res;
470 
471 	__asm __volatile(
472 	"	" MPLOCKED "		"
473 	"	cmpxchg8b %1 ;		"
474 	"	sete	%0"
475 	: "=q" (res),			/* 0 */
476 	  "+m" (*dst),			/* 1 */
477 	  "+A" (expect)			/* 2 */
478 	: "b" ((uint32_t)src),		/* 3 */
479 	  "c" ((uint32_t)(src >> 32))	/* 4 */
480 	: "memory", "cc");
481 	return (res);
482 }
483 
484 static __inline uint64_t
485 atomic_load_acq_64_i586(volatile uint64_t *p)
486 {
487 	uint64_t res;
488 
489 	__asm __volatile(
490 	"	movl	%%ebx,%%eax ;	"
491 	"	movl	%%ecx,%%edx ;	"
492 	"	" MPLOCKED "		"
493 	"	cmpxchg8b %1"
494 	: "=&A" (res),			/* 0 */
495 	  "+m" (*p)			/* 1 */
496 	: : "memory", "cc");
497 	return (res);
498 }
499 
500 static __inline void
501 atomic_store_rel_64_i586(volatile uint64_t *p, uint64_t v)
502 {
503 
504 	__asm __volatile(
505 	"	movl	%%eax,%%ebx ;	"
506 	"	movl	%%edx,%%ecx ;	"
507 	"1:				"
508 	"	" MPLOCKED "		"
509 	"	cmpxchg8b %0 ;		"
510 	"	jne	1b"
511 	: "+m" (*p),			/* 0 */
512 	  "+A" (v)			/* 1 */
513 	: : "ebx", "ecx", "memory", "cc");
514 }
515 
516 static __inline uint64_t
517 atomic_swap_64_i586(volatile uint64_t *p, uint64_t v)
518 {
519 
520 	__asm __volatile(
521 	"	movl	%%eax,%%ebx ;	"
522 	"	movl	%%edx,%%ecx ;	"
523 	"1:				"
524 	"	" MPLOCKED "		"
525 	"	cmpxchg8b %0 ;		"
526 	"	jne	1b"
527 	: "+m" (*p),			/* 0 */
528 	  "+A" (v)			/* 1 */
529 	: : "ebx", "ecx", "memory", "cc");
530 	return (v);
531 }
532 
533 static __inline int
534 atomic_cmpset_64(volatile uint64_t *dst, uint64_t expect, uint64_t src)
535 {
536 
537 	if ((cpu_feature & CPUID_CX8) == 0)
538 		return (atomic_cmpset_64_i386(dst, expect, src));
539 	else
540 		return (atomic_cmpset_64_i586(dst, expect, src));
541 }
542 
543 static __inline uint64_t
544 atomic_load_acq_64(volatile uint64_t *p)
545 {
546 
547 	if ((cpu_feature & CPUID_CX8) == 0)
548 		return (atomic_load_acq_64_i386(p));
549 	else
550 		return (atomic_load_acq_64_i586(p));
551 }
552 
553 static __inline void
554 atomic_store_rel_64(volatile uint64_t *p, uint64_t v)
555 {
556 
557 	if ((cpu_feature & CPUID_CX8) == 0)
558 		atomic_store_rel_64_i386(p, v);
559 	else
560 		atomic_store_rel_64_i586(p, v);
561 }
562 
563 static __inline uint64_t
564 atomic_swap_64(volatile uint64_t *p, uint64_t v)
565 {
566 
567 	if ((cpu_feature & CPUID_CX8) == 0)
568 		return (atomic_swap_64_i386(p, v));
569 	else
570 		return (atomic_swap_64_i586(p, v));
571 }
572 
573 static __inline uint64_t
574 atomic_fetchadd_64(volatile uint64_t *p, uint64_t v)
575 {
576 
577 	for (;;) {
578 		uint64_t t = *p;
579 		if (atomic_cmpset_64(p, t, t + v))
580 			return (t);
581 	}
582 }
583 
584 #endif /* _KERNEL */
585 
586 #endif /* KLD_MODULE || !__GNUCLIKE_ASM */
587 
588 ATOMIC_ASM(set,	     char,  "orb %b1,%0",  "iq",  v);
589 ATOMIC_ASM(clear,    char,  "andb %b1,%0", "iq", ~v);
590 ATOMIC_ASM(add,	     char,  "addb %b1,%0", "iq",  v);
591 ATOMIC_ASM(subtract, char,  "subb %b1,%0", "iq",  v);
592 
593 ATOMIC_ASM(set,	     short, "orw %w1,%0",  "ir",  v);
594 ATOMIC_ASM(clear,    short, "andw %w1,%0", "ir", ~v);
595 ATOMIC_ASM(add,	     short, "addw %w1,%0", "ir",  v);
596 ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir",  v);
597 
598 ATOMIC_ASM(set,	     int,   "orl %1,%0",   "ir",  v);
599 ATOMIC_ASM(clear,    int,   "andl %1,%0",  "ir", ~v);
600 ATOMIC_ASM(add,	     int,   "addl %1,%0",  "ir",  v);
601 ATOMIC_ASM(subtract, int,   "subl %1,%0",  "ir",  v);
602 
603 ATOMIC_ASM(set,	     long,  "orl %1,%0",   "ir",  v);
604 ATOMIC_ASM(clear,    long,  "andl %1,%0",  "ir", ~v);
605 ATOMIC_ASM(add,	     long,  "addl %1,%0",  "ir",  v);
606 ATOMIC_ASM(subtract, long,  "subl %1,%0",  "ir",  v);
607 
608 #define	ATOMIC_LOADSTORE(TYPE)				\
609 	ATOMIC_LOAD(TYPE);				\
610 	ATOMIC_STORE(TYPE)
611 
612 ATOMIC_LOADSTORE(char);
613 ATOMIC_LOADSTORE(short);
614 ATOMIC_LOADSTORE(int);
615 ATOMIC_LOADSTORE(long);
616 
617 #undef ATOMIC_ASM
618 #undef ATOMIC_LOAD
619 #undef ATOMIC_STORE
620 #undef ATOMIC_LOADSTORE
621 
622 #ifndef WANT_FUNCTIONS
623 
624 static __inline int
625 atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src)
626 {
627 
628 	return (atomic_cmpset_int((volatile u_int *)dst, (u_int)expect,
629 	    (u_int)src));
630 }
631 
632 static __inline u_long
633 atomic_fetchadd_long(volatile u_long *p, u_long v)
634 {
635 
636 	return (atomic_fetchadd_int((volatile u_int *)p, (u_int)v));
637 }
638 
639 static __inline int
640 atomic_testandset_long(volatile u_long *p, u_int v)
641 {
642 
643 	return (atomic_testandset_int((volatile u_int *)p, v));
644 }
645 
646 static __inline int
647 atomic_testandclear_long(volatile u_long *p, u_int v)
648 {
649 
650 	return (atomic_testandclear_int((volatile u_int *)p, v));
651 }
652 
653 /* Read the current value and store a new value in the destination. */
654 #ifdef __GNUCLIKE_ASM
655 
656 static __inline u_int
657 atomic_swap_int(volatile u_int *p, u_int v)
658 {
659 
660 	__asm __volatile(
661 	"	xchgl	%1,%0 ;		"
662 	"# atomic_swap_int"
663 	: "+r" (v),			/* 0 */
664 	  "+m" (*p));			/* 1 */
665 	return (v);
666 }
667 
668 static __inline u_long
669 atomic_swap_long(volatile u_long *p, u_long v)
670 {
671 
672 	return (atomic_swap_int((volatile u_int *)p, (u_int)v));
673 }
674 
675 #else /* !__GNUCLIKE_ASM */
676 
677 u_int	atomic_swap_int(volatile u_int *p, u_int v);
678 u_long	atomic_swap_long(volatile u_long *p, u_long v);
679 
680 #endif /* __GNUCLIKE_ASM */
681 
682 #define	atomic_set_acq_char		atomic_set_barr_char
683 #define	atomic_set_rel_char		atomic_set_barr_char
684 #define	atomic_clear_acq_char		atomic_clear_barr_char
685 #define	atomic_clear_rel_char		atomic_clear_barr_char
686 #define	atomic_add_acq_char		atomic_add_barr_char
687 #define	atomic_add_rel_char		atomic_add_barr_char
688 #define	atomic_subtract_acq_char	atomic_subtract_barr_char
689 #define	atomic_subtract_rel_char	atomic_subtract_barr_char
690 #define	atomic_cmpset_acq_char		atomic_cmpset_char
691 #define	atomic_cmpset_rel_char		atomic_cmpset_char
692 #define	atomic_fcmpset_acq_char		atomic_fcmpset_char
693 #define	atomic_fcmpset_rel_char		atomic_fcmpset_char
694 
695 #define	atomic_set_acq_short		atomic_set_barr_short
696 #define	atomic_set_rel_short		atomic_set_barr_short
697 #define	atomic_clear_acq_short		atomic_clear_barr_short
698 #define	atomic_clear_rel_short		atomic_clear_barr_short
699 #define	atomic_add_acq_short		atomic_add_barr_short
700 #define	atomic_add_rel_short		atomic_add_barr_short
701 #define	atomic_subtract_acq_short	atomic_subtract_barr_short
702 #define	atomic_subtract_rel_short	atomic_subtract_barr_short
703 #define	atomic_cmpset_acq_short		atomic_cmpset_short
704 #define	atomic_cmpset_rel_short		atomic_cmpset_short
705 #define	atomic_fcmpset_acq_short	atomic_fcmpset_short
706 #define	atomic_fcmpset_rel_short	atomic_fcmpset_short
707 
708 #define	atomic_set_acq_int		atomic_set_barr_int
709 #define	atomic_set_rel_int		atomic_set_barr_int
710 #define	atomic_clear_acq_int		atomic_clear_barr_int
711 #define	atomic_clear_rel_int		atomic_clear_barr_int
712 #define	atomic_add_acq_int		atomic_add_barr_int
713 #define	atomic_add_rel_int		atomic_add_barr_int
714 #define	atomic_subtract_acq_int		atomic_subtract_barr_int
715 #define	atomic_subtract_rel_int		atomic_subtract_barr_int
716 #define	atomic_cmpset_acq_int		atomic_cmpset_int
717 #define	atomic_cmpset_rel_int		atomic_cmpset_int
718 #define	atomic_fcmpset_acq_int		atomic_fcmpset_int
719 #define	atomic_fcmpset_rel_int		atomic_fcmpset_int
720 
721 #define	atomic_set_acq_long		atomic_set_barr_long
722 #define	atomic_set_rel_long		atomic_set_barr_long
723 #define	atomic_clear_acq_long		atomic_clear_barr_long
724 #define	atomic_clear_rel_long		atomic_clear_barr_long
725 #define	atomic_add_acq_long		atomic_add_barr_long
726 #define	atomic_add_rel_long		atomic_add_barr_long
727 #define	atomic_subtract_acq_long	atomic_subtract_barr_long
728 #define	atomic_subtract_rel_long	atomic_subtract_barr_long
729 #define	atomic_cmpset_acq_long		atomic_cmpset_long
730 #define	atomic_cmpset_rel_long		atomic_cmpset_long
731 #define	atomic_fcmpset_acq_long		atomic_fcmpset_long
732 #define	atomic_fcmpset_rel_long		atomic_fcmpset_long
733 
734 #define	atomic_readandclear_int(p)	atomic_swap_int(p, 0)
735 #define	atomic_readandclear_long(p)	atomic_swap_long(p, 0)
736 
737 /* Operations on 8-bit bytes. */
738 #define	atomic_set_8		atomic_set_char
739 #define	atomic_set_acq_8	atomic_set_acq_char
740 #define	atomic_set_rel_8	atomic_set_rel_char
741 #define	atomic_clear_8		atomic_clear_char
742 #define	atomic_clear_acq_8	atomic_clear_acq_char
743 #define	atomic_clear_rel_8	atomic_clear_rel_char
744 #define	atomic_add_8		atomic_add_char
745 #define	atomic_add_acq_8	atomic_add_acq_char
746 #define	atomic_add_rel_8	atomic_add_rel_char
747 #define	atomic_subtract_8	atomic_subtract_char
748 #define	atomic_subtract_acq_8	atomic_subtract_acq_char
749 #define	atomic_subtract_rel_8	atomic_subtract_rel_char
750 #define	atomic_load_acq_8	atomic_load_acq_char
751 #define	atomic_store_rel_8	atomic_store_rel_char
752 #define	atomic_cmpset_8		atomic_cmpset_char
753 #define	atomic_cmpset_acq_8	atomic_cmpset_acq_char
754 #define	atomic_cmpset_rel_8	atomic_cmpset_rel_char
755 #define	atomic_fcmpset_8	atomic_fcmpset_char
756 #define	atomic_fcmpset_acq_8	atomic_fcmpset_acq_char
757 #define	atomic_fcmpset_rel_8	atomic_fcmpset_rel_char
758 
759 /* Operations on 16-bit words. */
760 #define	atomic_set_16		atomic_set_short
761 #define	atomic_set_acq_16	atomic_set_acq_short
762 #define	atomic_set_rel_16	atomic_set_rel_short
763 #define	atomic_clear_16		atomic_clear_short
764 #define	atomic_clear_acq_16	atomic_clear_acq_short
765 #define	atomic_clear_rel_16	atomic_clear_rel_short
766 #define	atomic_add_16		atomic_add_short
767 #define	atomic_add_acq_16	atomic_add_acq_short
768 #define	atomic_add_rel_16	atomic_add_rel_short
769 #define	atomic_subtract_16	atomic_subtract_short
770 #define	atomic_subtract_acq_16	atomic_subtract_acq_short
771 #define	atomic_subtract_rel_16	atomic_subtract_rel_short
772 #define	atomic_load_acq_16	atomic_load_acq_short
773 #define	atomic_store_rel_16	atomic_store_rel_short
774 #define	atomic_cmpset_16	atomic_cmpset_short
775 #define	atomic_cmpset_acq_16	atomic_cmpset_acq_short
776 #define	atomic_cmpset_rel_16	atomic_cmpset_rel_short
777 #define	atomic_fcmpset_16	atomic_fcmpset_short
778 #define	atomic_fcmpset_acq_16	atomic_fcmpset_acq_short
779 #define	atomic_fcmpset_rel_16	atomic_fcmpset_rel_short
780 
781 /* Operations on 32-bit double words. */
782 #define	atomic_set_32		atomic_set_int
783 #define	atomic_set_acq_32	atomic_set_acq_int
784 #define	atomic_set_rel_32	atomic_set_rel_int
785 #define	atomic_clear_32		atomic_clear_int
786 #define	atomic_clear_acq_32	atomic_clear_acq_int
787 #define	atomic_clear_rel_32	atomic_clear_rel_int
788 #define	atomic_add_32		atomic_add_int
789 #define	atomic_add_acq_32	atomic_add_acq_int
790 #define	atomic_add_rel_32	atomic_add_rel_int
791 #define	atomic_subtract_32	atomic_subtract_int
792 #define	atomic_subtract_acq_32	atomic_subtract_acq_int
793 #define	atomic_subtract_rel_32	atomic_subtract_rel_int
794 #define	atomic_load_acq_32	atomic_load_acq_int
795 #define	atomic_store_rel_32	atomic_store_rel_int
796 #define	atomic_cmpset_32	atomic_cmpset_int
797 #define	atomic_cmpset_acq_32	atomic_cmpset_acq_int
798 #define	atomic_cmpset_rel_32	atomic_cmpset_rel_int
799 #define	atomic_fcmpset_32	atomic_fcmpset_int
800 #define	atomic_fcmpset_acq_32	atomic_fcmpset_acq_int
801 #define	atomic_fcmpset_rel_32	atomic_fcmpset_rel_int
802 #define	atomic_swap_32		atomic_swap_int
803 #define	atomic_readandclear_32	atomic_readandclear_int
804 #define	atomic_fetchadd_32	atomic_fetchadd_int
805 #define	atomic_testandset_32	atomic_testandset_int
806 #define	atomic_testandclear_32	atomic_testandclear_int
807 
808 /* Operations on pointers. */
809 #define	atomic_set_ptr(p, v) \
810 	atomic_set_int((volatile u_int *)(p), (u_int)(v))
811 #define	atomic_set_acq_ptr(p, v) \
812 	atomic_set_acq_int((volatile u_int *)(p), (u_int)(v))
813 #define	atomic_set_rel_ptr(p, v) \
814 	atomic_set_rel_int((volatile u_int *)(p), (u_int)(v))
815 #define	atomic_clear_ptr(p, v) \
816 	atomic_clear_int((volatile u_int *)(p), (u_int)(v))
817 #define	atomic_clear_acq_ptr(p, v) \
818 	atomic_clear_acq_int((volatile u_int *)(p), (u_int)(v))
819 #define	atomic_clear_rel_ptr(p, v) \
820 	atomic_clear_rel_int((volatile u_int *)(p), (u_int)(v))
821 #define	atomic_add_ptr(p, v) \
822 	atomic_add_int((volatile u_int *)(p), (u_int)(v))
823 #define	atomic_add_acq_ptr(p, v) \
824 	atomic_add_acq_int((volatile u_int *)(p), (u_int)(v))
825 #define	atomic_add_rel_ptr(p, v) \
826 	atomic_add_rel_int((volatile u_int *)(p), (u_int)(v))
827 #define	atomic_subtract_ptr(p, v) \
828 	atomic_subtract_int((volatile u_int *)(p), (u_int)(v))
829 #define	atomic_subtract_acq_ptr(p, v) \
830 	atomic_subtract_acq_int((volatile u_int *)(p), (u_int)(v))
831 #define	atomic_subtract_rel_ptr(p, v) \
832 	atomic_subtract_rel_int((volatile u_int *)(p), (u_int)(v))
833 #define	atomic_load_acq_ptr(p) \
834 	atomic_load_acq_int((volatile u_int *)(p))
835 #define	atomic_store_rel_ptr(p, v) \
836 	atomic_store_rel_int((volatile u_int *)(p), (v))
837 #define	atomic_cmpset_ptr(dst, old, new) \
838 	atomic_cmpset_int((volatile u_int *)(dst), (u_int)(old), (u_int)(new))
839 #define	atomic_cmpset_acq_ptr(dst, old, new) \
840 	atomic_cmpset_acq_int((volatile u_int *)(dst), (u_int)(old), \
841 	    (u_int)(new))
842 #define	atomic_cmpset_rel_ptr(dst, old, new) \
843 	atomic_cmpset_rel_int((volatile u_int *)(dst), (u_int)(old), \
844 	    (u_int)(new))
845 #define	atomic_fcmpset_ptr(dst, old, new) \
846 	atomic_fcmpset_int((volatile u_int *)(dst), (u_int *)(old), (u_int)(new))
847 #define	atomic_fcmpset_acq_ptr(dst, old, new) \
848 	atomic_fcmpset_acq_int((volatile u_int *)(dst), (u_int *)(old), \
849 	    (u_int)(new))
850 #define	atomic_fcmpset_rel_ptr(dst, old, new) \
851 	atomic_fcmpset_rel_int((volatile u_int *)(dst), (u_int *)(old), \
852 	    (u_int)(new))
853 #define	atomic_swap_ptr(p, v) \
854 	atomic_swap_int((volatile u_int *)(p), (u_int)(v))
855 #define	atomic_readandclear_ptr(p) \
856 	atomic_readandclear_int((volatile u_int *)(p))
857 
858 #endif /* !WANT_FUNCTIONS */
859 
860 #if defined(_KERNEL)
861 #define	mb()	__mbk()
862 #define	wmb()	__mbk()
863 #define	rmb()	__mbk()
864 #else
865 #define	mb()	__mbu()
866 #define	wmb()	__mbu()
867 #define	rmb()	__mbu()
868 #endif
869 
870 #endif /* !_MACHINE_ATOMIC_H_ */
871