atomic.h revision 331722
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: stable/11/sys/i386/include/atomic.h 331722 2018-03-29 02:50:57Z eadler $
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#include <sys/atomic_common.h>
36
37#ifdef _KERNEL
38#include <machine/md_var.h>
39#include <machine/specialreg.h>
40#endif
41
42#ifndef __OFFSETOF_MONITORBUF
43/*
44 * __OFFSETOF_MONITORBUF == __pcpu_offset(pc_monitorbuf).
45 *
46 * The open-coded number is used instead of the symbolic expression to
47 * avoid a dependency on sys/pcpu.h in machine/atomic.h consumers.
48 * An assertion in i386/vm_machdep.c ensures that the value is correct.
49 */
50#define	__OFFSETOF_MONITORBUF	0x180
51
52static __inline void
53__mbk(void)
54{
55
56	__asm __volatile("lock; addl $0,%%fs:%0"
57	    : "+m" (*(u_int *)__OFFSETOF_MONITORBUF) : : "memory", "cc");
58}
59
60static __inline void
61__mbu(void)
62{
63
64	__asm __volatile("lock; addl $0,(%%esp)" : : : "memory", "cc");
65}
66#endif
67
68/*
69 * Various simple operations on memory, each of which is atomic in the
70 * presence of interrupts and multiple processors.
71 *
72 * atomic_set_char(P, V)	(*(u_char *)(P) |= (V))
73 * atomic_clear_char(P, V)	(*(u_char *)(P) &= ~(V))
74 * atomic_add_char(P, V)	(*(u_char *)(P) += (V))
75 * atomic_subtract_char(P, V)	(*(u_char *)(P) -= (V))
76 *
77 * atomic_set_short(P, V)	(*(u_short *)(P) |= (V))
78 * atomic_clear_short(P, V)	(*(u_short *)(P) &= ~(V))
79 * atomic_add_short(P, V)	(*(u_short *)(P) += (V))
80 * atomic_subtract_short(P, V)	(*(u_short *)(P) -= (V))
81 *
82 * atomic_set_int(P, V)		(*(u_int *)(P) |= (V))
83 * atomic_clear_int(P, V)	(*(u_int *)(P) &= ~(V))
84 * atomic_add_int(P, V)		(*(u_int *)(P) += (V))
85 * atomic_subtract_int(P, V)	(*(u_int *)(P) -= (V))
86 * atomic_swap_int(P, V)	(return (*(u_int *)(P)); *(u_int *)(P) = (V);)
87 * atomic_readandclear_int(P)	(return (*(u_int *)(P)); *(u_int *)(P) = 0;)
88 *
89 * atomic_set_long(P, V)	(*(u_long *)(P) |= (V))
90 * atomic_clear_long(P, V)	(*(u_long *)(P) &= ~(V))
91 * atomic_add_long(P, V)	(*(u_long *)(P) += (V))
92 * atomic_subtract_long(P, V)	(*(u_long *)(P) -= (V))
93 * atomic_swap_long(P, V)	(return (*(u_long *)(P)); *(u_long *)(P) = (V);)
94 * atomic_readandclear_long(P)	(return (*(u_long *)(P)); *(u_long *)(P) = 0;)
95 */
96
97/*
98 * The above functions are expanded inline in the statically-linked
99 * kernel.  Lock prefixes are generated if an SMP kernel is being
100 * built.
101 *
102 * Kernel modules call real functions which are built into the kernel.
103 * This allows kernel modules to be portable between UP and SMP systems.
104 */
105#if defined(KLD_MODULE) || !defined(__GNUCLIKE_ASM)
106#define	ATOMIC_ASM(NAME, TYPE, OP, CONS, V)			\
107void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v);	\
108void atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
109
110int	atomic_cmpset_char(volatile u_char *dst, u_char expect, u_char src);
111int	atomic_cmpset_short(volatile u_short *dst, u_short expect, u_short src);
112int	atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src);
113int	atomic_fcmpset_char(volatile u_char *dst, u_char *expect, u_char src);
114int	atomic_fcmpset_short(volatile u_short *dst, u_short *expect,
115	    u_short src);
116int	atomic_fcmpset_int(volatile u_int *dst, u_int *expect, u_int src);
117u_int	atomic_fetchadd_int(volatile u_int *p, u_int v);
118int	atomic_testandset_int(volatile u_int *p, u_int v);
119int	atomic_testandclear_int(volatile u_int *p, u_int v);
120void	atomic_thread_fence_acq(void);
121void	atomic_thread_fence_acq_rel(void);
122void	atomic_thread_fence_rel(void);
123void	atomic_thread_fence_seq_cst(void);
124
125#define	ATOMIC_LOAD(TYPE)					\
126u_##TYPE	atomic_load_acq_##TYPE(volatile u_##TYPE *p)
127#define	ATOMIC_STORE(TYPE)					\
128void		atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
129
130int		atomic_cmpset_64(volatile uint64_t *, uint64_t, uint64_t);
131uint64_t	atomic_load_acq_64(volatile uint64_t *);
132void		atomic_store_rel_64(volatile uint64_t *, uint64_t);
133uint64_t	atomic_swap_64(volatile uint64_t *, uint64_t);
134uint64_t	atomic_fetchadd_64(volatile uint64_t *, uint64_t);
135
136#else /* !KLD_MODULE && __GNUCLIKE_ASM */
137
138/*
139 * For userland, always use lock prefixes so that the binaries will run
140 * on both SMP and !SMP systems.
141 */
142#if defined(SMP) || !defined(_KERNEL)
143#define	MPLOCKED	"lock ; "
144#else
145#define	MPLOCKED
146#endif
147
148/*
149 * The assembly is volatilized to avoid code chunk removal by the compiler.
150 * GCC aggressively reorders operations and memory clobbering is necessary
151 * in order to avoid that for memory barriers.
152 */
153#define	ATOMIC_ASM(NAME, TYPE, OP, CONS, V)		\
154static __inline void					\
155atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
156{							\
157	__asm __volatile(MPLOCKED OP			\
158	: "+m" (*p)					\
159	: CONS (V)					\
160	: "cc");					\
161}							\
162							\
163static __inline void					\
164atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
165{							\
166	__asm __volatile(MPLOCKED OP			\
167	: "+m" (*p)					\
168	: CONS (V)					\
169	: "memory", "cc");				\
170}							\
171struct __hack
172
173/*
174 * Atomic compare and set, used by the mutex functions.
175 *
176 * cmpset:
177 *	if (*dst == expect)
178 *		*dst = src
179 *
180 * fcmpset:
181 *	if (*dst == *expect)
182 *		*dst = src
183 *	else
184 *		*expect = *dst
185 *
186 * Returns 0 on failure, non-zero on success.
187 */
188#define	ATOMIC_CMPSET(TYPE, CONS)			\
189static __inline int					\
190atomic_cmpset_##TYPE(volatile u_##TYPE *dst, u_##TYPE expect, u_##TYPE src) \
191{							\
192	u_char res;					\
193							\
194	__asm __volatile(				\
195	"	" MPLOCKED "		"		\
196	"	cmpxchg	%3,%1 ;		"		\
197	"	sete	%0 ;		"		\
198	"# atomic_cmpset_" #TYPE "	"		\
199	: "=q" (res),			/* 0 */		\
200	  "+m" (*dst),			/* 1 */		\
201	  "+a" (expect)			/* 2 */		\
202	: CONS (src)			/* 3 */		\
203	: "memory", "cc");				\
204	return (res);					\
205}							\
206							\
207static __inline int					\
208atomic_fcmpset_##TYPE(volatile u_##TYPE *dst, u_##TYPE *expect, u_##TYPE src) \
209{							\
210	u_char res;					\
211							\
212	__asm __volatile(				\
213	"	" MPLOCKED "		"		\
214	"	cmpxchg	%3,%1 ;		"		\
215	"	sete	%0 ;		"		\
216	"# atomic_fcmpset_" #TYPE "	"		\
217	: "=q" (res),			/* 0 */		\
218	  "+m" (*dst),			/* 1 */		\
219	  "+a" (*expect)		/* 2 */		\
220	: CONS (src)			/* 3 */		\
221	: "memory", "cc");				\
222	return (res);					\
223}
224
225ATOMIC_CMPSET(char, "q");
226ATOMIC_CMPSET(short, "r");
227ATOMIC_CMPSET(int, "r");
228
229/*
230 * Atomically add the value of v to the integer pointed to by p and return
231 * the previous value of *p.
232 */
233static __inline u_int
234atomic_fetchadd_int(volatile u_int *p, u_int v)
235{
236
237	__asm __volatile(
238	"	" MPLOCKED "		"
239	"	xaddl	%0,%1 ;		"
240	"# atomic_fetchadd_int"
241	: "+r" (v),			/* 0 */
242	  "+m" (*p)			/* 1 */
243	: : "cc");
244	return (v);
245}
246
247static __inline int
248atomic_testandset_int(volatile u_int *p, u_int v)
249{
250	u_char res;
251
252	__asm __volatile(
253	"	" MPLOCKED "		"
254	"	btsl	%2,%1 ;		"
255	"	setc	%0 ;		"
256	"# atomic_testandset_int"
257	: "=q" (res),			/* 0 */
258	  "+m" (*p)			/* 1 */
259	: "Ir" (v & 0x1f)		/* 2 */
260	: "cc");
261	return (res);
262}
263
264static __inline int
265atomic_testandclear_int(volatile u_int *p, u_int v)
266{
267	u_char res;
268
269	__asm __volatile(
270	"	" MPLOCKED "		"
271	"	btrl	%2,%1 ;		"
272	"	setc	%0 ;		"
273	"# atomic_testandclear_int"
274	: "=q" (res),			/* 0 */
275	  "+m" (*p)			/* 1 */
276	: "Ir" (v & 0x1f)		/* 2 */
277	: "cc");
278	return (res);
279}
280
281/*
282 * We assume that a = b will do atomic loads and stores.  Due to the
283 * IA32 memory model, a simple store guarantees release semantics.
284 *
285 * However, a load may pass a store if they are performed on distinct
286 * addresses, so we need Store/Load barrier for sequentially
287 * consistent fences in SMP kernels.  We use "lock addl $0,mem" for a
288 * Store/Load barrier, as recommended by the AMD Software Optimization
289 * Guide, and not mfence.  In the kernel, we use a private per-cpu
290 * cache line for "mem", to avoid introducing false data
291 * dependencies.  In user space, we use the word at the top of the
292 * stack.
293 *
294 * For UP kernels, however, the memory of the single processor is
295 * always consistent, so we only need to stop the compiler from
296 * reordering accesses in a way that violates the semantics of acquire
297 * and release.
298 */
299
300#if defined(_KERNEL)
301#if defined(SMP)
302#define	__storeload_barrier()	__mbk()
303#else /* _KERNEL && UP */
304#define	__storeload_barrier()	__compiler_membar()
305#endif /* SMP */
306#else /* !_KERNEL */
307#define	__storeload_barrier()	__mbu()
308#endif /* _KERNEL*/
309
310#define	ATOMIC_LOAD(TYPE)					\
311static __inline u_##TYPE					\
312atomic_load_acq_##TYPE(volatile u_##TYPE *p)			\
313{								\
314	u_##TYPE res;						\
315								\
316	res = *p;						\
317	__compiler_membar();					\
318	return (res);						\
319}								\
320struct __hack
321
322#define	ATOMIC_STORE(TYPE)					\
323static __inline void						\
324atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)	\
325{								\
326								\
327	__compiler_membar();					\
328	*p = v;							\
329}								\
330struct __hack
331
332static __inline void
333atomic_thread_fence_acq(void)
334{
335
336	__compiler_membar();
337}
338
339static __inline void
340atomic_thread_fence_rel(void)
341{
342
343	__compiler_membar();
344}
345
346static __inline void
347atomic_thread_fence_acq_rel(void)
348{
349
350	__compiler_membar();
351}
352
353static __inline void
354atomic_thread_fence_seq_cst(void)
355{
356
357	__storeload_barrier();
358}
359
360#ifdef _KERNEL
361
362#ifdef WANT_FUNCTIONS
363int		atomic_cmpset_64_i386(volatile uint64_t *, uint64_t, uint64_t);
364int		atomic_cmpset_64_i586(volatile uint64_t *, uint64_t, uint64_t);
365uint64_t	atomic_load_acq_64_i386(volatile uint64_t *);
366uint64_t	atomic_load_acq_64_i586(volatile uint64_t *);
367void		atomic_store_rel_64_i386(volatile uint64_t *, uint64_t);
368void		atomic_store_rel_64_i586(volatile uint64_t *, uint64_t);
369uint64_t	atomic_swap_64_i386(volatile uint64_t *, uint64_t);
370uint64_t	atomic_swap_64_i586(volatile uint64_t *, uint64_t);
371#endif
372
373/* I486 does not support SMP or CMPXCHG8B. */
374static __inline int
375atomic_cmpset_64_i386(volatile uint64_t *dst, uint64_t expect, uint64_t src)
376{
377	volatile uint32_t *p;
378	u_char res;
379
380	p = (volatile uint32_t *)dst;
381	__asm __volatile(
382	"	pushfl ;		"
383	"	cli ;			"
384	"	xorl	%1,%%eax ;	"
385	"	xorl	%2,%%edx ;	"
386	"	orl	%%edx,%%eax ;	"
387	"	jne	1f ;		"
388	"	movl	%4,%1 ;		"
389	"	movl	%5,%2 ;		"
390	"1:				"
391	"	sete	%3 ;		"
392	"	popfl"
393	: "+A" (expect),		/* 0 */
394	  "+m" (*p),			/* 1 */
395	  "+m" (*(p + 1)),		/* 2 */
396	  "=q" (res)			/* 3 */
397	: "r" ((uint32_t)src),		/* 4 */
398	  "r" ((uint32_t)(src >> 32))	/* 5 */
399	: "memory", "cc");
400	return (res);
401}
402
403static __inline uint64_t
404atomic_load_acq_64_i386(volatile uint64_t *p)
405{
406	volatile uint32_t *q;
407	uint64_t res;
408
409	q = (volatile uint32_t *)p;
410	__asm __volatile(
411	"	pushfl ;		"
412	"	cli ;			"
413	"	movl	%1,%%eax ;	"
414	"	movl	%2,%%edx ;	"
415	"	popfl"
416	: "=&A" (res)			/* 0 */
417	: "m" (*q),			/* 1 */
418	  "m" (*(q + 1))		/* 2 */
419	: "memory");
420	return (res);
421}
422
423static __inline void
424atomic_store_rel_64_i386(volatile uint64_t *p, uint64_t v)
425{
426	volatile uint32_t *q;
427
428	q = (volatile uint32_t *)p;
429	__asm __volatile(
430	"	pushfl ;		"
431	"	cli ;			"
432	"	movl	%%eax,%0 ;	"
433	"	movl	%%edx,%1 ;	"
434	"	popfl"
435	: "=m" (*q),			/* 0 */
436	  "=m" (*(q + 1))		/* 1 */
437	: "A" (v)			/* 2 */
438	: "memory");
439}
440
441static __inline uint64_t
442atomic_swap_64_i386(volatile uint64_t *p, uint64_t v)
443{
444	volatile uint32_t *q;
445	uint64_t res;
446
447	q = (volatile uint32_t *)p;
448	__asm __volatile(
449	"	pushfl ;		"
450	"	cli ;			"
451	"	movl	%1,%%eax ;	"
452	"	movl	%2,%%edx ;	"
453	"	movl	%4,%2 ;		"
454	"	movl	%3,%1 ;		"
455	"	popfl"
456	: "=&A" (res),			/* 0 */
457	  "+m" (*q),			/* 1 */
458	  "+m" (*(q + 1))		/* 2 */
459	: "r" ((uint32_t)v),		/* 3 */
460	  "r" ((uint32_t)(v >> 32)));	/* 4 */
461	return (res);
462}
463
464static __inline int
465atomic_cmpset_64_i586(volatile uint64_t *dst, uint64_t expect, uint64_t src)
466{
467	u_char res;
468
469	__asm __volatile(
470	"	" MPLOCKED "		"
471	"	cmpxchg8b %1 ;		"
472	"	sete	%0"
473	: "=q" (res),			/* 0 */
474	  "+m" (*dst),			/* 1 */
475	  "+A" (expect)			/* 2 */
476	: "b" ((uint32_t)src),		/* 3 */
477	  "c" ((uint32_t)(src >> 32))	/* 4 */
478	: "memory", "cc");
479	return (res);
480}
481
482static __inline uint64_t
483atomic_load_acq_64_i586(volatile uint64_t *p)
484{
485	uint64_t res;
486
487	__asm __volatile(
488	"	movl	%%ebx,%%eax ;	"
489	"	movl	%%ecx,%%edx ;	"
490	"	" MPLOCKED "		"
491	"	cmpxchg8b %1"
492	: "=&A" (res),			/* 0 */
493	  "+m" (*p)			/* 1 */
494	: : "memory", "cc");
495	return (res);
496}
497
498static __inline void
499atomic_store_rel_64_i586(volatile uint64_t *p, uint64_t v)
500{
501
502	__asm __volatile(
503	"	movl	%%eax,%%ebx ;	"
504	"	movl	%%edx,%%ecx ;	"
505	"1:				"
506	"	" MPLOCKED "		"
507	"	cmpxchg8b %0 ;		"
508	"	jne	1b"
509	: "+m" (*p),			/* 0 */
510	  "+A" (v)			/* 1 */
511	: : "ebx", "ecx", "memory", "cc");
512}
513
514static __inline uint64_t
515atomic_swap_64_i586(volatile uint64_t *p, uint64_t v)
516{
517
518	__asm __volatile(
519	"	movl	%%eax,%%ebx ;	"
520	"	movl	%%edx,%%ecx ;	"
521	"1:				"
522	"	" MPLOCKED "		"
523	"	cmpxchg8b %0 ;		"
524	"	jne	1b"
525	: "+m" (*p),			/* 0 */
526	  "+A" (v)			/* 1 */
527	: : "ebx", "ecx", "memory", "cc");
528	return (v);
529}
530
531static __inline int
532atomic_cmpset_64(volatile uint64_t *dst, uint64_t expect, uint64_t src)
533{
534
535	if ((cpu_feature & CPUID_CX8) == 0)
536		return (atomic_cmpset_64_i386(dst, expect, src));
537	else
538		return (atomic_cmpset_64_i586(dst, expect, src));
539}
540
541static __inline uint64_t
542atomic_load_acq_64(volatile uint64_t *p)
543{
544
545	if ((cpu_feature & CPUID_CX8) == 0)
546		return (atomic_load_acq_64_i386(p));
547	else
548		return (atomic_load_acq_64_i586(p));
549}
550
551static __inline void
552atomic_store_rel_64(volatile uint64_t *p, uint64_t v)
553{
554
555	if ((cpu_feature & CPUID_CX8) == 0)
556		atomic_store_rel_64_i386(p, v);
557	else
558		atomic_store_rel_64_i586(p, v);
559}
560
561static __inline uint64_t
562atomic_swap_64(volatile uint64_t *p, uint64_t v)
563{
564
565	if ((cpu_feature & CPUID_CX8) == 0)
566		return (atomic_swap_64_i386(p, v));
567	else
568		return (atomic_swap_64_i586(p, v));
569}
570
571static __inline uint64_t
572atomic_fetchadd_64(volatile uint64_t *p, uint64_t v)
573{
574
575	for (;;) {
576		uint64_t t = *p;
577		if (atomic_cmpset_64(p, t, t + v))
578			return (t);
579	}
580}
581
582#endif /* _KERNEL */
583
584#endif /* KLD_MODULE || !__GNUCLIKE_ASM */
585
586ATOMIC_ASM(set,	     char,  "orb %b1,%0",  "iq",  v);
587ATOMIC_ASM(clear,    char,  "andb %b1,%0", "iq", ~v);
588ATOMIC_ASM(add,	     char,  "addb %b1,%0", "iq",  v);
589ATOMIC_ASM(subtract, char,  "subb %b1,%0", "iq",  v);
590
591ATOMIC_ASM(set,	     short, "orw %w1,%0",  "ir",  v);
592ATOMIC_ASM(clear,    short, "andw %w1,%0", "ir", ~v);
593ATOMIC_ASM(add,	     short, "addw %w1,%0", "ir",  v);
594ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir",  v);
595
596ATOMIC_ASM(set,	     int,   "orl %1,%0",   "ir",  v);
597ATOMIC_ASM(clear,    int,   "andl %1,%0",  "ir", ~v);
598ATOMIC_ASM(add,	     int,   "addl %1,%0",  "ir",  v);
599ATOMIC_ASM(subtract, int,   "subl %1,%0",  "ir",  v);
600
601ATOMIC_ASM(set,	     long,  "orl %1,%0",   "ir",  v);
602ATOMIC_ASM(clear,    long,  "andl %1,%0",  "ir", ~v);
603ATOMIC_ASM(add,	     long,  "addl %1,%0",  "ir",  v);
604ATOMIC_ASM(subtract, long,  "subl %1,%0",  "ir",  v);
605
606#define	ATOMIC_LOADSTORE(TYPE)				\
607	ATOMIC_LOAD(TYPE);				\
608	ATOMIC_STORE(TYPE)
609
610ATOMIC_LOADSTORE(char);
611ATOMIC_LOADSTORE(short);
612ATOMIC_LOADSTORE(int);
613ATOMIC_LOADSTORE(long);
614
615#undef ATOMIC_ASM
616#undef ATOMIC_LOAD
617#undef ATOMIC_STORE
618#undef ATOMIC_LOADSTORE
619
620#ifndef WANT_FUNCTIONS
621
622static __inline int
623atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src)
624{
625
626	return (atomic_cmpset_int((volatile u_int *)dst, (u_int)expect,
627	    (u_int)src));
628}
629
630static __inline u_long
631atomic_fetchadd_long(volatile u_long *p, u_long v)
632{
633
634	return (atomic_fetchadd_int((volatile u_int *)p, (u_int)v));
635}
636
637static __inline int
638atomic_testandset_long(volatile u_long *p, u_int v)
639{
640
641	return (atomic_testandset_int((volatile u_int *)p, v));
642}
643
644static __inline int
645atomic_testandclear_long(volatile u_long *p, u_int v)
646{
647
648	return (atomic_testandclear_int((volatile u_int *)p, v));
649}
650
651/* Read the current value and store a new value in the destination. */
652#ifdef __GNUCLIKE_ASM
653
654static __inline u_int
655atomic_swap_int(volatile u_int *p, u_int v)
656{
657
658	__asm __volatile(
659	"	xchgl	%1,%0 ;		"
660	"# atomic_swap_int"
661	: "+r" (v),			/* 0 */
662	  "+m" (*p));			/* 1 */
663	return (v);
664}
665
666static __inline u_long
667atomic_swap_long(volatile u_long *p, u_long v)
668{
669
670	return (atomic_swap_int((volatile u_int *)p, (u_int)v));
671}
672
673#else /* !__GNUCLIKE_ASM */
674
675u_int	atomic_swap_int(volatile u_int *p, u_int v);
676u_long	atomic_swap_long(volatile u_long *p, u_long v);
677
678#endif /* __GNUCLIKE_ASM */
679
680#define	atomic_set_acq_char		atomic_set_barr_char
681#define	atomic_set_rel_char		atomic_set_barr_char
682#define	atomic_clear_acq_char		atomic_clear_barr_char
683#define	atomic_clear_rel_char		atomic_clear_barr_char
684#define	atomic_add_acq_char		atomic_add_barr_char
685#define	atomic_add_rel_char		atomic_add_barr_char
686#define	atomic_subtract_acq_char	atomic_subtract_barr_char
687#define	atomic_subtract_rel_char	atomic_subtract_barr_char
688#define	atomic_cmpset_acq_char		atomic_cmpset_char
689#define	atomic_cmpset_rel_char		atomic_cmpset_char
690#define	atomic_fcmpset_acq_char		atomic_fcmpset_char
691#define	atomic_fcmpset_rel_char		atomic_fcmpset_char
692
693#define	atomic_set_acq_short		atomic_set_barr_short
694#define	atomic_set_rel_short		atomic_set_barr_short
695#define	atomic_clear_acq_short		atomic_clear_barr_short
696#define	atomic_clear_rel_short		atomic_clear_barr_short
697#define	atomic_add_acq_short		atomic_add_barr_short
698#define	atomic_add_rel_short		atomic_add_barr_short
699#define	atomic_subtract_acq_short	atomic_subtract_barr_short
700#define	atomic_subtract_rel_short	atomic_subtract_barr_short
701#define	atomic_cmpset_acq_short		atomic_cmpset_short
702#define	atomic_cmpset_rel_short		atomic_cmpset_short
703#define	atomic_fcmpset_acq_short	atomic_fcmpset_short
704#define	atomic_fcmpset_rel_short	atomic_fcmpset_short
705
706#define	atomic_set_acq_int		atomic_set_barr_int
707#define	atomic_set_rel_int		atomic_set_barr_int
708#define	atomic_clear_acq_int		atomic_clear_barr_int
709#define	atomic_clear_rel_int		atomic_clear_barr_int
710#define	atomic_add_acq_int		atomic_add_barr_int
711#define	atomic_add_rel_int		atomic_add_barr_int
712#define	atomic_subtract_acq_int		atomic_subtract_barr_int
713#define	atomic_subtract_rel_int		atomic_subtract_barr_int
714#define	atomic_cmpset_acq_int		atomic_cmpset_int
715#define	atomic_cmpset_rel_int		atomic_cmpset_int
716#define	atomic_fcmpset_acq_int		atomic_fcmpset_int
717#define	atomic_fcmpset_rel_int		atomic_fcmpset_int
718
719#define	atomic_set_acq_long		atomic_set_barr_long
720#define	atomic_set_rel_long		atomic_set_barr_long
721#define	atomic_clear_acq_long		atomic_clear_barr_long
722#define	atomic_clear_rel_long		atomic_clear_barr_long
723#define	atomic_add_acq_long		atomic_add_barr_long
724#define	atomic_add_rel_long		atomic_add_barr_long
725#define	atomic_subtract_acq_long	atomic_subtract_barr_long
726#define	atomic_subtract_rel_long	atomic_subtract_barr_long
727#define	atomic_cmpset_acq_long		atomic_cmpset_long
728#define	atomic_cmpset_rel_long		atomic_cmpset_long
729#define	atomic_fcmpset_acq_long		atomic_fcmpset_long
730#define	atomic_fcmpset_rel_long		atomic_fcmpset_long
731
732#define	atomic_readandclear_int(p)	atomic_swap_int(p, 0)
733#define	atomic_readandclear_long(p)	atomic_swap_long(p, 0)
734
735/* Operations on 8-bit bytes. */
736#define	atomic_set_8		atomic_set_char
737#define	atomic_set_acq_8	atomic_set_acq_char
738#define	atomic_set_rel_8	atomic_set_rel_char
739#define	atomic_clear_8		atomic_clear_char
740#define	atomic_clear_acq_8	atomic_clear_acq_char
741#define	atomic_clear_rel_8	atomic_clear_rel_char
742#define	atomic_add_8		atomic_add_char
743#define	atomic_add_acq_8	atomic_add_acq_char
744#define	atomic_add_rel_8	atomic_add_rel_char
745#define	atomic_subtract_8	atomic_subtract_char
746#define	atomic_subtract_acq_8	atomic_subtract_acq_char
747#define	atomic_subtract_rel_8	atomic_subtract_rel_char
748#define	atomic_load_acq_8	atomic_load_acq_char
749#define	atomic_store_rel_8	atomic_store_rel_char
750#define	atomic_cmpset_8		atomic_cmpset_char
751#define	atomic_cmpset_acq_8	atomic_cmpset_acq_char
752#define	atomic_cmpset_rel_8	atomic_cmpset_rel_char
753#define	atomic_fcmpset_8	atomic_fcmpset_char
754#define	atomic_fcmpset_acq_8	atomic_fcmpset_acq_char
755#define	atomic_fcmpset_rel_8	atomic_fcmpset_rel_char
756
757/* Operations on 16-bit words. */
758#define	atomic_set_16		atomic_set_short
759#define	atomic_set_acq_16	atomic_set_acq_short
760#define	atomic_set_rel_16	atomic_set_rel_short
761#define	atomic_clear_16		atomic_clear_short
762#define	atomic_clear_acq_16	atomic_clear_acq_short
763#define	atomic_clear_rel_16	atomic_clear_rel_short
764#define	atomic_add_16		atomic_add_short
765#define	atomic_add_acq_16	atomic_add_acq_short
766#define	atomic_add_rel_16	atomic_add_rel_short
767#define	atomic_subtract_16	atomic_subtract_short
768#define	atomic_subtract_acq_16	atomic_subtract_acq_short
769#define	atomic_subtract_rel_16	atomic_subtract_rel_short
770#define	atomic_load_acq_16	atomic_load_acq_short
771#define	atomic_store_rel_16	atomic_store_rel_short
772#define	atomic_cmpset_16	atomic_cmpset_short
773#define	atomic_cmpset_acq_16	atomic_cmpset_acq_short
774#define	atomic_cmpset_rel_16	atomic_cmpset_rel_short
775#define	atomic_fcmpset_16	atomic_fcmpset_short
776#define	atomic_fcmpset_acq_16	atomic_fcmpset_acq_short
777#define	atomic_fcmpset_rel_16	atomic_fcmpset_rel_short
778
779/* Operations on 32-bit double words. */
780#define	atomic_set_32		atomic_set_int
781#define	atomic_set_acq_32	atomic_set_acq_int
782#define	atomic_set_rel_32	atomic_set_rel_int
783#define	atomic_clear_32		atomic_clear_int
784#define	atomic_clear_acq_32	atomic_clear_acq_int
785#define	atomic_clear_rel_32	atomic_clear_rel_int
786#define	atomic_add_32		atomic_add_int
787#define	atomic_add_acq_32	atomic_add_acq_int
788#define	atomic_add_rel_32	atomic_add_rel_int
789#define	atomic_subtract_32	atomic_subtract_int
790#define	atomic_subtract_acq_32	atomic_subtract_acq_int
791#define	atomic_subtract_rel_32	atomic_subtract_rel_int
792#define	atomic_load_acq_32	atomic_load_acq_int
793#define	atomic_store_rel_32	atomic_store_rel_int
794#define	atomic_cmpset_32	atomic_cmpset_int
795#define	atomic_cmpset_acq_32	atomic_cmpset_acq_int
796#define	atomic_cmpset_rel_32	atomic_cmpset_rel_int
797#define	atomic_fcmpset_32	atomic_fcmpset_int
798#define	atomic_fcmpset_acq_32	atomic_fcmpset_acq_int
799#define	atomic_fcmpset_rel_32	atomic_fcmpset_rel_int
800#define	atomic_swap_32		atomic_swap_int
801#define	atomic_readandclear_32	atomic_readandclear_int
802#define	atomic_fetchadd_32	atomic_fetchadd_int
803#define	atomic_testandset_32	atomic_testandset_int
804#define	atomic_testandclear_32	atomic_testandclear_int
805
806/* Operations on pointers. */
807#define	atomic_set_ptr(p, v) \
808	atomic_set_int((volatile u_int *)(p), (u_int)(v))
809#define	atomic_set_acq_ptr(p, v) \
810	atomic_set_acq_int((volatile u_int *)(p), (u_int)(v))
811#define	atomic_set_rel_ptr(p, v) \
812	atomic_set_rel_int((volatile u_int *)(p), (u_int)(v))
813#define	atomic_clear_ptr(p, v) \
814	atomic_clear_int((volatile u_int *)(p), (u_int)(v))
815#define	atomic_clear_acq_ptr(p, v) \
816	atomic_clear_acq_int((volatile u_int *)(p), (u_int)(v))
817#define	atomic_clear_rel_ptr(p, v) \
818	atomic_clear_rel_int((volatile u_int *)(p), (u_int)(v))
819#define	atomic_add_ptr(p, v) \
820	atomic_add_int((volatile u_int *)(p), (u_int)(v))
821#define	atomic_add_acq_ptr(p, v) \
822	atomic_add_acq_int((volatile u_int *)(p), (u_int)(v))
823#define	atomic_add_rel_ptr(p, v) \
824	atomic_add_rel_int((volatile u_int *)(p), (u_int)(v))
825#define	atomic_subtract_ptr(p, v) \
826	atomic_subtract_int((volatile u_int *)(p), (u_int)(v))
827#define	atomic_subtract_acq_ptr(p, v) \
828	atomic_subtract_acq_int((volatile u_int *)(p), (u_int)(v))
829#define	atomic_subtract_rel_ptr(p, v) \
830	atomic_subtract_rel_int((volatile u_int *)(p), (u_int)(v))
831#define	atomic_load_acq_ptr(p) \
832	atomic_load_acq_int((volatile u_int *)(p))
833#define	atomic_store_rel_ptr(p, v) \
834	atomic_store_rel_int((volatile u_int *)(p), (v))
835#define	atomic_cmpset_ptr(dst, old, new) \
836	atomic_cmpset_int((volatile u_int *)(dst), (u_int)(old), (u_int)(new))
837#define	atomic_cmpset_acq_ptr(dst, old, new) \
838	atomic_cmpset_acq_int((volatile u_int *)(dst), (u_int)(old), \
839	    (u_int)(new))
840#define	atomic_cmpset_rel_ptr(dst, old, new) \
841	atomic_cmpset_rel_int((volatile u_int *)(dst), (u_int)(old), \
842	    (u_int)(new))
843#define	atomic_fcmpset_ptr(dst, old, new) \
844	atomic_fcmpset_int((volatile u_int *)(dst), (u_int *)(old), (u_int)(new))
845#define	atomic_fcmpset_acq_ptr(dst, old, new) \
846	atomic_fcmpset_acq_int((volatile u_int *)(dst), (u_int *)(old), \
847	    (u_int)(new))
848#define	atomic_fcmpset_rel_ptr(dst, old, new) \
849	atomic_fcmpset_rel_int((volatile u_int *)(dst), (u_int *)(old), \
850	    (u_int)(new))
851#define	atomic_swap_ptr(p, v) \
852	atomic_swap_int((volatile u_int *)(p), (u_int)(v))
853#define	atomic_readandclear_ptr(p) \
854	atomic_readandclear_int((volatile u_int *)(p))
855
856#endif /* !WANT_FUNCTIONS */
857
858#if defined(_KERNEL)
859#define	mb()	__mbk()
860#define	wmb()	__mbk()
861#define	rmb()	__mbk()
862#else
863#define	mb()	__mbu()
864#define	wmb()	__mbu()
865#define	rmb()	__mbu()
866#endif
867
868#endif /* !_MACHINE_ATOMIC_H_ */
869