atomic.h revision 254610
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: head/sys/amd64/include/atomic.h 254610 2013-08-21 20:43:50Z jkim $
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#define	mb()	__asm __volatile("mfence;" : : : "memory")
36#define	wmb()	__asm __volatile("sfence;" : : : "memory")
37#define	rmb()	__asm __volatile("lfence;" : : : "memory")
38
39/*
40 * Various simple operations on memory, each of which is atomic in the
41 * presence of interrupts and multiple processors.
42 *
43 * atomic_set_char(P, V)	(*(u_char *)(P) |= (V))
44 * atomic_clear_char(P, V)	(*(u_char *)(P) &= ~(V))
45 * atomic_add_char(P, V)	(*(u_char *)(P) += (V))
46 * atomic_subtract_char(P, V)	(*(u_char *)(P) -= (V))
47 *
48 * atomic_set_short(P, V)	(*(u_short *)(P) |= (V))
49 * atomic_clear_short(P, V)	(*(u_short *)(P) &= ~(V))
50 * atomic_add_short(P, V)	(*(u_short *)(P) += (V))
51 * atomic_subtract_short(P, V)	(*(u_short *)(P) -= (V))
52 *
53 * atomic_set_int(P, V)		(*(u_int *)(P) |= (V))
54 * atomic_clear_int(P, V)	(*(u_int *)(P) &= ~(V))
55 * atomic_add_int(P, V)		(*(u_int *)(P) += (V))
56 * atomic_subtract_int(P, V)	(*(u_int *)(P) -= (V))
57 * atomic_readandclear_int(P)	(return (*(u_int *)(P)); *(u_int *)(P) = 0;)
58 *
59 * atomic_set_long(P, V)	(*(u_long *)(P) |= (V))
60 * atomic_clear_long(P, V)	(*(u_long *)(P) &= ~(V))
61 * atomic_add_long(P, V)	(*(u_long *)(P) += (V))
62 * atomic_subtract_long(P, V)	(*(u_long *)(P) -= (V))
63 * atomic_readandclear_long(P)	(return (*(u_long *)(P)); *(u_long *)(P) = 0;)
64 */
65
66/*
67 * The above functions are expanded inline in the statically-linked
68 * kernel.  Lock prefixes are generated if an SMP kernel is being
69 * built.
70 *
71 * Kernel modules call real functions which are built into the kernel.
72 * This allows kernel modules to be portable between UP and SMP systems.
73 */
74#if defined(KLD_MODULE) || !defined(__GNUCLIKE_ASM)
75#define	ATOMIC_ASM(NAME, TYPE, OP, CONS, V)			\
76void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v);	\
77void atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
78
79int	atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src);
80int	atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src);
81u_int	atomic_fetchadd_int(volatile u_int *p, u_int v);
82u_long	atomic_fetchadd_long(volatile u_long *p, u_long v);
83
84#define	ATOMIC_LOAD(TYPE, LOP)					\
85u_##TYPE	atomic_load_acq_##TYPE(volatile u_##TYPE *p)
86#define	ATOMIC_STORE(TYPE)					\
87void		atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
88
89#else /* !KLD_MODULE && __GNUCLIKE_ASM */
90
91/*
92 * For userland, always use lock prefixes so that the binaries will run
93 * on both SMP and !SMP systems.
94 */
95#if defined(SMP) || !defined(_KERNEL)
96#define	MPLOCKED	"lock ; "
97#else
98#define	MPLOCKED
99#endif
100
101/*
102 * The assembly is volatilized to avoid code chunk removal by the compiler.
103 * GCC aggressively reorders operations and memory clobbering is necessary
104 * in order to avoid that for memory barriers.
105 */
106#define	ATOMIC_ASM(NAME, TYPE, OP, CONS, V)		\
107static __inline void					\
108atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
109{							\
110	__asm __volatile(MPLOCKED OP			\
111	: "=m" (*p)					\
112	: CONS (V), "m" (*p)				\
113	: "cc");					\
114}							\
115							\
116static __inline void					\
117atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
118{							\
119	__asm __volatile(MPLOCKED OP			\
120	: "=m" (*p)					\
121	: CONS (V), "m" (*p)				\
122	: "memory", "cc");				\
123}							\
124struct __hack
125
126/*
127 * Atomic compare and set, used by the mutex functions
128 *
129 * if (*dst == expect) *dst = src (all 32 bit words)
130 *
131 * Returns 0 on failure, non-zero on success
132 */
133
134static __inline int
135atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src)
136{
137	u_char res;
138
139	__asm __volatile(
140	"	" MPLOCKED "		"
141	"	cmpxchgl %2,%1 ;	"
142	"       sete	%0 ;		"
143	"1:				"
144	"# atomic_cmpset_int"
145	: "=a" (res),			/* 0 */
146	  "=m" (*dst)			/* 1 */
147	: "r" (src),			/* 2 */
148	  "a" (expect),			/* 3 */
149	  "m" (*dst)			/* 4 */
150	: "memory", "cc");
151
152	return (res);
153}
154
155static __inline int
156atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src)
157{
158	u_char res;
159
160	__asm __volatile(
161	"	" MPLOCKED "		"
162	"	cmpxchgq %2,%1 ;	"
163	"       sete	%0 ;		"
164	"1:				"
165	"# atomic_cmpset_long"
166	: "=a" (res),			/* 0 */
167	  "=m" (*dst)			/* 1 */
168	: "r" (src),			/* 2 */
169	  "a" (expect),			/* 3 */
170	  "m" (*dst)			/* 4 */
171	: "memory", "cc");
172
173	return (res);
174}
175
176/*
177 * Atomically add the value of v to the integer pointed to by p and return
178 * the previous value of *p.
179 */
180static __inline u_int
181atomic_fetchadd_int(volatile u_int *p, u_int v)
182{
183
184	__asm __volatile(
185	"	" MPLOCKED "		"
186	"	xaddl	%0,%1 ;		"
187	"# atomic_fetchadd_int"
188	: "+r" (v),			/* 0 */
189	  "=m" (*p)			/* 1 */
190	: "m" (*p)			/* 2 */
191	: "cc");
192	return (v);
193}
194
195/*
196 * Atomically add the value of v to the long integer pointed to by p and return
197 * the previous value of *p.
198 */
199static __inline u_long
200atomic_fetchadd_long(volatile u_long *p, u_long v)
201{
202
203	__asm __volatile(
204	"	" MPLOCKED "		"
205	"	xaddq	%0,%1 ;		"
206	"# atomic_fetchadd_long"
207	: "+r" (v),			/* 0 */
208	  "=m" (*p)			/* 1 */
209	: "m" (*p)			/* 2 */
210	: "cc");
211	return (v);
212}
213
214/*
215 * We assume that a = b will do atomic loads and stores.  Due to the
216 * IA32 memory model, a simple store guarantees release semantics.
217 *
218 * However, loads may pass stores, so for atomic_load_acq we have to
219 * ensure a Store/Load barrier to do the load in SMP kernels.  We use
220 * "lock cmpxchg" as recommended by the AMD Software Optimization
221 * Guide, and not mfence.  For UP kernels, however, the cache of the
222 * single processor is always consistent, so we only need to take care
223 * of the compiler.
224 */
225#define	ATOMIC_STORE(TYPE)				\
226static __inline void					\
227atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
228{							\
229	__compiler_membar();				\
230	*p = v;						\
231}							\
232struct __hack
233
234#if defined(_KERNEL) && !defined(SMP)
235
236#define	ATOMIC_LOAD(TYPE, LOP)				\
237static __inline u_##TYPE				\
238atomic_load_acq_##TYPE(volatile u_##TYPE *p)		\
239{							\
240	u_##TYPE tmp;					\
241							\
242	tmp = *p;					\
243	__compiler_membar();				\
244	return (tmp);					\
245}							\
246struct __hack
247
248#else /* !(_KERNEL && !SMP) */
249
250#define	ATOMIC_LOAD(TYPE, LOP)				\
251static __inline u_##TYPE				\
252atomic_load_acq_##TYPE(volatile u_##TYPE *p)		\
253{							\
254	u_##TYPE res;					\
255							\
256	__asm __volatile(MPLOCKED LOP			\
257	: "=a" (res),			/* 0 */		\
258	  "=m" (*p)			/* 1 */		\
259	: "m" (*p)			/* 2 */		\
260	: "memory", "cc");				\
261							\
262	return (res);					\
263}							\
264struct __hack
265
266#endif /* _KERNEL && !SMP */
267
268#endif /* KLD_MODULE || !__GNUCLIKE_ASM */
269
270ATOMIC_ASM(set,	     char,  "orb %b1,%0",  "iq",  v);
271ATOMIC_ASM(clear,    char,  "andb %b1,%0", "iq", ~v);
272ATOMIC_ASM(add,	     char,  "addb %b1,%0", "iq",  v);
273ATOMIC_ASM(subtract, char,  "subb %b1,%0", "iq",  v);
274
275ATOMIC_ASM(set,	     short, "orw %w1,%0",  "ir",  v);
276ATOMIC_ASM(clear,    short, "andw %w1,%0", "ir", ~v);
277ATOMIC_ASM(add,	     short, "addw %w1,%0", "ir",  v);
278ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir",  v);
279
280ATOMIC_ASM(set,	     int,   "orl %1,%0",   "ir",  v);
281ATOMIC_ASM(clear,    int,   "andl %1,%0",  "ir", ~v);
282ATOMIC_ASM(add,	     int,   "addl %1,%0",  "ir",  v);
283ATOMIC_ASM(subtract, int,   "subl %1,%0",  "ir",  v);
284
285ATOMIC_ASM(set,	     long,  "orq %1,%0",   "ir",  v);
286ATOMIC_ASM(clear,    long,  "andq %1,%0",  "ir", ~v);
287ATOMIC_ASM(add,	     long,  "addq %1,%0",  "ir",  v);
288ATOMIC_ASM(subtract, long,  "subq %1,%0",  "ir",  v);
289
290ATOMIC_LOAD(char,  "cmpxchgb %b0,%1");
291ATOMIC_LOAD(short, "cmpxchgw %w0,%1");
292ATOMIC_LOAD(int,   "cmpxchgl %0,%1");
293ATOMIC_LOAD(long,  "cmpxchgq %0,%1");
294
295ATOMIC_STORE(char);
296ATOMIC_STORE(short);
297ATOMIC_STORE(int);
298ATOMIC_STORE(long);
299
300#undef ATOMIC_ASM
301#undef ATOMIC_LOAD
302#undef ATOMIC_STORE
303
304#ifndef WANT_FUNCTIONS
305
306/* Read the current value and store a zero in the destination. */
307#ifdef __GNUCLIKE_ASM
308
309static __inline u_int
310atomic_readandclear_int(volatile u_int *p)
311{
312	u_int res;
313
314	res = 0;
315	__asm __volatile(
316	"	xchgl	%1,%0 ;		"
317	"# atomic_readandclear_int"
318	: "+r" (res),			/* 0 */
319	  "=m" (*p)			/* 1 */
320	: "m" (*p));
321
322	return (res);
323}
324
325static __inline u_long
326atomic_readandclear_long(volatile u_long *p)
327{
328	u_long res;
329
330	res = 0;
331	__asm __volatile(
332	"	xchgq	%1,%0 ;		"
333	"# atomic_readandclear_long"
334	: "+r" (res),			/* 0 */
335	  "=m" (*p)			/* 1 */
336	: "m" (*p));
337
338	return (res);
339}
340
341#else /* !__GNUCLIKE_ASM */
342
343u_int	atomic_readandclear_int(volatile u_int *p);
344u_long	atomic_readandclear_long(volatile u_long *p);
345
346#endif /* __GNUCLIKE_ASM */
347
348#define	atomic_set_acq_char		atomic_set_barr_char
349#define	atomic_set_rel_char		atomic_set_barr_char
350#define	atomic_clear_acq_char		atomic_clear_barr_char
351#define	atomic_clear_rel_char		atomic_clear_barr_char
352#define	atomic_add_acq_char		atomic_add_barr_char
353#define	atomic_add_rel_char		atomic_add_barr_char
354#define	atomic_subtract_acq_char	atomic_subtract_barr_char
355#define	atomic_subtract_rel_char	atomic_subtract_barr_char
356
357#define	atomic_set_acq_short		atomic_set_barr_short
358#define	atomic_set_rel_short		atomic_set_barr_short
359#define	atomic_clear_acq_short		atomic_clear_barr_short
360#define	atomic_clear_rel_short		atomic_clear_barr_short
361#define	atomic_add_acq_short		atomic_add_barr_short
362#define	atomic_add_rel_short		atomic_add_barr_short
363#define	atomic_subtract_acq_short	atomic_subtract_barr_short
364#define	atomic_subtract_rel_short	atomic_subtract_barr_short
365
366#define	atomic_set_acq_int		atomic_set_barr_int
367#define	atomic_set_rel_int		atomic_set_barr_int
368#define	atomic_clear_acq_int		atomic_clear_barr_int
369#define	atomic_clear_rel_int		atomic_clear_barr_int
370#define	atomic_add_acq_int		atomic_add_barr_int
371#define	atomic_add_rel_int		atomic_add_barr_int
372#define	atomic_subtract_acq_int		atomic_subtract_barr_int
373#define	atomic_subtract_rel_int		atomic_subtract_barr_int
374#define	atomic_cmpset_acq_int		atomic_cmpset_int
375#define	atomic_cmpset_rel_int		atomic_cmpset_int
376
377#define	atomic_set_acq_long		atomic_set_barr_long
378#define	atomic_set_rel_long		atomic_set_barr_long
379#define	atomic_clear_acq_long		atomic_clear_barr_long
380#define	atomic_clear_rel_long		atomic_clear_barr_long
381#define	atomic_add_acq_long		atomic_add_barr_long
382#define	atomic_add_rel_long		atomic_add_barr_long
383#define	atomic_subtract_acq_long	atomic_subtract_barr_long
384#define	atomic_subtract_rel_long	atomic_subtract_barr_long
385#define	atomic_cmpset_acq_long		atomic_cmpset_long
386#define	atomic_cmpset_rel_long		atomic_cmpset_long
387
388/* Operations on 8-bit bytes. */
389#define	atomic_set_8		atomic_set_char
390#define	atomic_set_acq_8	atomic_set_acq_char
391#define	atomic_set_rel_8	atomic_set_rel_char
392#define	atomic_clear_8		atomic_clear_char
393#define	atomic_clear_acq_8	atomic_clear_acq_char
394#define	atomic_clear_rel_8	atomic_clear_rel_char
395#define	atomic_add_8		atomic_add_char
396#define	atomic_add_acq_8	atomic_add_acq_char
397#define	atomic_add_rel_8	atomic_add_rel_char
398#define	atomic_subtract_8	atomic_subtract_char
399#define	atomic_subtract_acq_8	atomic_subtract_acq_char
400#define	atomic_subtract_rel_8	atomic_subtract_rel_char
401#define	atomic_load_acq_8	atomic_load_acq_char
402#define	atomic_store_rel_8	atomic_store_rel_char
403
404/* Operations on 16-bit words. */
405#define	atomic_set_16		atomic_set_short
406#define	atomic_set_acq_16	atomic_set_acq_short
407#define	atomic_set_rel_16	atomic_set_rel_short
408#define	atomic_clear_16		atomic_clear_short
409#define	atomic_clear_acq_16	atomic_clear_acq_short
410#define	atomic_clear_rel_16	atomic_clear_rel_short
411#define	atomic_add_16		atomic_add_short
412#define	atomic_add_acq_16	atomic_add_acq_short
413#define	atomic_add_rel_16	atomic_add_rel_short
414#define	atomic_subtract_16	atomic_subtract_short
415#define	atomic_subtract_acq_16	atomic_subtract_acq_short
416#define	atomic_subtract_rel_16	atomic_subtract_rel_short
417#define	atomic_load_acq_16	atomic_load_acq_short
418#define	atomic_store_rel_16	atomic_store_rel_short
419
420/* Operations on 32-bit double words. */
421#define	atomic_set_32		atomic_set_int
422#define	atomic_set_acq_32	atomic_set_acq_int
423#define	atomic_set_rel_32	atomic_set_rel_int
424#define	atomic_clear_32		atomic_clear_int
425#define	atomic_clear_acq_32	atomic_clear_acq_int
426#define	atomic_clear_rel_32	atomic_clear_rel_int
427#define	atomic_add_32		atomic_add_int
428#define	atomic_add_acq_32	atomic_add_acq_int
429#define	atomic_add_rel_32	atomic_add_rel_int
430#define	atomic_subtract_32	atomic_subtract_int
431#define	atomic_subtract_acq_32	atomic_subtract_acq_int
432#define	atomic_subtract_rel_32	atomic_subtract_rel_int
433#define	atomic_load_acq_32	atomic_load_acq_int
434#define	atomic_store_rel_32	atomic_store_rel_int
435#define	atomic_cmpset_32	atomic_cmpset_int
436#define	atomic_cmpset_acq_32	atomic_cmpset_acq_int
437#define	atomic_cmpset_rel_32	atomic_cmpset_rel_int
438#define	atomic_readandclear_32	atomic_readandclear_int
439#define	atomic_fetchadd_32	atomic_fetchadd_int
440
441/* Operations on 64-bit quad words. */
442#define	atomic_set_64		atomic_set_long
443#define	atomic_set_acq_64	atomic_set_acq_long
444#define	atomic_set_rel_64	atomic_set_rel_long
445#define	atomic_clear_64		atomic_clear_long
446#define	atomic_clear_acq_64	atomic_clear_acq_long
447#define	atomic_clear_rel_64	atomic_clear_rel_long
448#define	atomic_add_64		atomic_add_long
449#define	atomic_add_acq_64	atomic_add_acq_long
450#define	atomic_add_rel_64	atomic_add_rel_long
451#define	atomic_subtract_64	atomic_subtract_long
452#define	atomic_subtract_acq_64	atomic_subtract_acq_long
453#define	atomic_subtract_rel_64	atomic_subtract_rel_long
454#define	atomic_load_acq_64	atomic_load_acq_long
455#define	atomic_store_rel_64	atomic_store_rel_long
456#define	atomic_cmpset_64	atomic_cmpset_long
457#define	atomic_cmpset_acq_64	atomic_cmpset_acq_long
458#define	atomic_cmpset_rel_64	atomic_cmpset_rel_long
459#define	atomic_readandclear_64	atomic_readandclear_long
460
461/* Operations on pointers. */
462#define	atomic_set_ptr		atomic_set_long
463#define	atomic_set_acq_ptr	atomic_set_acq_long
464#define	atomic_set_rel_ptr	atomic_set_rel_long
465#define	atomic_clear_ptr	atomic_clear_long
466#define	atomic_clear_acq_ptr	atomic_clear_acq_long
467#define	atomic_clear_rel_ptr	atomic_clear_rel_long
468#define	atomic_add_ptr		atomic_add_long
469#define	atomic_add_acq_ptr	atomic_add_acq_long
470#define	atomic_add_rel_ptr	atomic_add_rel_long
471#define	atomic_subtract_ptr	atomic_subtract_long
472#define	atomic_subtract_acq_ptr	atomic_subtract_acq_long
473#define	atomic_subtract_rel_ptr	atomic_subtract_rel_long
474#define	atomic_load_acq_ptr	atomic_load_acq_long
475#define	atomic_store_rel_ptr	atomic_store_rel_long
476#define	atomic_cmpset_ptr	atomic_cmpset_long
477#define	atomic_cmpset_acq_ptr	atomic_cmpset_acq_long
478#define	atomic_cmpset_rel_ptr	atomic_cmpset_rel_long
479#define	atomic_readandclear_ptr	atomic_readandclear_long
480
481#endif /* !WANT_FUNCTIONS */
482
483#endif /* !_MACHINE_ATOMIC_H_ */
484