xref: /freebsd/sys/amd64/include/atomic.h (revision c98323078dede7579020518ec84cdcb478e5c142)
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 /*
32  * Various simple arithmetic on memory which is atomic in the presence
33  * of interrupts and multiple processors.
34  *
35  * atomic_set_char(P, V)	(*(u_char*)(P) |= (V))
36  * atomic_clear_char(P, V)	(*(u_char*)(P) &= ~(V))
37  * atomic_add_char(P, V)	(*(u_char*)(P) += (V))
38  * atomic_subtract_char(P, V)	(*(u_char*)(P) -= (V))
39  *
40  * atomic_set_short(P, V)	(*(u_short*)(P) |= (V))
41  * atomic_clear_short(P, V)	(*(u_short*)(P) &= ~(V))
42  * atomic_add_short(P, V)	(*(u_short*)(P) += (V))
43  * atomic_subtract_short(P, V)	(*(u_short*)(P) -= (V))
44  *
45  * atomic_set_int(P, V)		(*(u_int*)(P) |= (V))
46  * atomic_clear_int(P, V)	(*(u_int*)(P) &= ~(V))
47  * atomic_add_int(P, V)		(*(u_int*)(P) += (V))
48  * atomic_subtract_int(P, V)	(*(u_int*)(P) -= (V))
49  * atomic_readandclear_int(P)	(return  *(u_int*)P; *(u_int*)P = 0;)
50  *
51  * atomic_set_long(P, V)	(*(u_long*)(P) |= (V))
52  * atomic_clear_long(P, V)	(*(u_long*)(P) &= ~(V))
53  * atomic_add_long(P, V)	(*(u_long*)(P) += (V))
54  * atomic_subtract_long(P, V)	(*(u_long*)(P) -= (V))
55  * atomic_readandclear_long(P)	(return  *(u_long*)P; *(u_long*)P = 0;)
56  */
57 
58 /*
59  * The above functions are expanded inline in the statically-linked
60  * kernel.  Lock prefixes are generated if an SMP kernel is being
61  * built.
62  *
63  * Kernel modules call real functions which are built into the kernel.
64  * This allows kernel modules to be portable between UP and SMP systems.
65  */
66 #if defined(KLD_MODULE)
67 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V)			\
68 void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
69 
70 int atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src);
71 int atomic_cmpset_long(volatile u_long *dst, u_long exp, u_long src);
72 
73 #define	ATOMIC_STORE_LOAD(TYPE, LOP, SOP)			\
74 u_##TYPE	atomic_load_acq_##TYPE(volatile u_##TYPE *p);	\
75 void		atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
76 
77 #else /* !KLD_MODULE */
78 
79 #ifdef __GNUC__
80 
81 /*
82  * For userland, assume the SMP case and use lock prefixes so that
83  * the binaries will run on both types of systems.
84  */
85 #if defined(SMP) || !defined(_KERNEL)
86 #define MPLOCKED	lock ;
87 #else
88 #define MPLOCKED
89 #endif
90 
91 /*
92  * The assembly is volatilized to demark potential before-and-after side
93  * effects if an interrupt or SMP collision were to occur.
94  */
95 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V)		\
96 static __inline void					\
97 atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
98 {							\
99 	__asm __volatile(__XSTRING(MPLOCKED) OP		\
100 			 : "+m" (*p)			\
101 			 : CONS (V));			\
102 }							\
103 struct __hack
104 
105 #else /* !__GNUC__ */
106 
107 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V)				\
108 extern void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
109 
110 #endif /* __GNUC__ */
111 
112 /*
113  * Atomic compare and set, used by the mutex functions
114  *
115  * if (*dst == exp) *dst = src (all 32 bit words)
116  *
117  * Returns 0 on failure, non-zero on success
118  */
119 
120 #if defined(__GNUC__)
121 
122 static __inline int
123 atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
124 {
125 	int res = exp;
126 
127 	__asm __volatile (
128 	"	" __XSTRING(MPLOCKED) "	"
129 	"	cmpxchgl %1,%2 ;	"
130 	"       setz	%%al ;		"
131 	"	movzbl	%%al,%0 ;	"
132 	"1:				"
133 	"# atomic_cmpset_int"
134 	: "+a" (res)			/* 0 (result) */
135 	: "r" (src),			/* 1 */
136 	  "m" (*(dst))			/* 2 */
137 	: "memory");
138 
139 	return (res);
140 }
141 
142 static __inline int
143 atomic_cmpset_long(volatile u_long *dst, u_long exp, u_long src)
144 {
145 	long res = exp;
146 
147 	__asm __volatile (
148 	"	" __XSTRING(MPLOCKED) "	"
149 	"	cmpxchgq %1,%2 ;	"
150 	"       setz	%%al ;		"
151 	"	movzbq	%%al,%0 ;	"
152 	"1:				"
153 	"# atomic_cmpset_long"
154 	: "+a" (res)			/* 0 (result) */
155 	: "r" (src),			/* 1 */
156 	  "m" (*(dst))			/* 2 */
157 	: "memory");
158 
159 	return (res);
160 }
161 #endif /* defined(__GNUC__) */
162 
163 #if defined(__GNUC__)
164 
165 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP)		\
166 static __inline u_##TYPE				\
167 atomic_load_acq_##TYPE(volatile u_##TYPE *p)		\
168 {							\
169 	u_##TYPE res;					\
170 							\
171 	__asm __volatile(__XSTRING(MPLOCKED) LOP	\
172 	: "=a" (res),			/* 0 (result) */\
173 	  "+m" (*p)			/* 1 */		\
174 	: : "memory");				 	\
175 							\
176 	return (res);					\
177 }							\
178 							\
179 /*							\
180  * The XCHG instruction asserts LOCK automagically.	\
181  */							\
182 static __inline void					\
183 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
184 {							\
185 	__asm __volatile(SOP				\
186 	: "+m" (*p),			/* 0 */		\
187 	  "+r" (v)			/* 1 */		\
188 	: : "memory");				 	\
189 }							\
190 struct __hack
191 
192 #else /* !defined(__GNUC__) */
193 
194 extern int atomic_cmpset_int(volatile u_int *, u_int, u_int);
195 extern int atomic_cmpset_long(volatile u_long *, u_long, u_long);
196 
197 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP)				\
198 extern u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p);		\
199 extern void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
200 
201 #endif /* defined(__GNUC__) */
202 
203 #endif /* KLD_MODULE */
204 
205 ATOMIC_ASM(set,	     char,  "orb %b1,%0",  "iq",  v);
206 ATOMIC_ASM(clear,    char,  "andb %b1,%0", "iq", ~v);
207 ATOMIC_ASM(add,	     char,  "addb %b1,%0", "iq",  v);
208 ATOMIC_ASM(subtract, char,  "subb %b1,%0", "iq",  v);
209 
210 ATOMIC_ASM(set,	     short, "orw %w1,%0",  "ir",  v);
211 ATOMIC_ASM(clear,    short, "andw %w1,%0", "ir", ~v);
212 ATOMIC_ASM(add,	     short, "addw %w1,%0", "ir",  v);
213 ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir",  v);
214 
215 ATOMIC_ASM(set,	     int,   "orl %1,%0",   "ir",  v);
216 ATOMIC_ASM(clear,    int,   "andl %1,%0",  "ir", ~v);
217 ATOMIC_ASM(add,	     int,   "addl %1,%0",  "ir",  v);
218 ATOMIC_ASM(subtract, int,   "subl %1,%0",  "ir",  v);
219 
220 ATOMIC_ASM(set,	     long,  "orq %1,%0",   "ir",  v);
221 ATOMIC_ASM(clear,    long,  "andq %1,%0",  "ir", ~v);
222 ATOMIC_ASM(add,	     long,  "addq %1,%0",  "ir",  v);
223 ATOMIC_ASM(subtract, long,  "subq %1,%0",  "ir",  v);
224 
225 ATOMIC_STORE_LOAD(char,	"cmpxchgb %b0,%1", "xchgb %b1,%0");
226 ATOMIC_STORE_LOAD(short,"cmpxchgw %w0,%1", "xchgw %w1,%0");
227 ATOMIC_STORE_LOAD(int,	"cmpxchgl %0,%1",  "xchgl %1,%0");
228 ATOMIC_STORE_LOAD(long,	"cmpxchgq %0,%1",  "xchgq %1,%0");
229 
230 #undef ATOMIC_ASM
231 #undef ATOMIC_STORE_LOAD
232 
233 #define	atomic_set_acq_char		atomic_set_char
234 #define	atomic_set_rel_char		atomic_set_char
235 #define	atomic_clear_acq_char		atomic_clear_char
236 #define	atomic_clear_rel_char		atomic_clear_char
237 #define	atomic_add_acq_char		atomic_add_char
238 #define	atomic_add_rel_char		atomic_add_char
239 #define	atomic_subtract_acq_char	atomic_subtract_char
240 #define	atomic_subtract_rel_char	atomic_subtract_char
241 
242 #define	atomic_set_acq_short		atomic_set_short
243 #define	atomic_set_rel_short		atomic_set_short
244 #define	atomic_clear_acq_short		atomic_clear_short
245 #define	atomic_clear_rel_short		atomic_clear_short
246 #define	atomic_add_acq_short		atomic_add_short
247 #define	atomic_add_rel_short		atomic_add_short
248 #define	atomic_subtract_acq_short	atomic_subtract_short
249 #define	atomic_subtract_rel_short	atomic_subtract_short
250 
251 #define	atomic_set_acq_int		atomic_set_int
252 #define	atomic_set_rel_int		atomic_set_int
253 #define	atomic_clear_acq_int		atomic_clear_int
254 #define	atomic_clear_rel_int		atomic_clear_int
255 #define	atomic_add_acq_int		atomic_add_int
256 #define	atomic_add_rel_int		atomic_add_int
257 #define	atomic_subtract_acq_int		atomic_subtract_int
258 #define	atomic_subtract_rel_int		atomic_subtract_int
259 #define atomic_cmpset_acq_int		atomic_cmpset_int
260 #define atomic_cmpset_rel_int		atomic_cmpset_int
261 
262 #define	atomic_set_acq_long		atomic_set_long
263 #define	atomic_set_rel_long		atomic_set_long
264 #define	atomic_clear_acq_long		atomic_clear_long
265 #define	atomic_clear_rel_long		atomic_clear_long
266 #define	atomic_add_acq_long		atomic_add_long
267 #define	atomic_add_rel_long		atomic_add_long
268 #define	atomic_subtract_acq_long	atomic_subtract_long
269 #define	atomic_subtract_rel_long	atomic_subtract_long
270 
271 #define atomic_cmpset_acq_ptr		atomic_cmpset_ptr
272 #define atomic_cmpset_rel_ptr		atomic_cmpset_ptr
273 
274 #define	atomic_set_8		atomic_set_char
275 #define	atomic_set_acq_8	atomic_set_acq_char
276 #define	atomic_set_rel_8	atomic_set_rel_char
277 #define	atomic_clear_8		atomic_clear_char
278 #define	atomic_clear_acq_8	atomic_clear_acq_char
279 #define	atomic_clear_rel_8	atomic_clear_rel_char
280 #define	atomic_add_8		atomic_add_char
281 #define	atomic_add_acq_8	atomic_add_acq_char
282 #define	atomic_add_rel_8	atomic_add_rel_char
283 #define	atomic_subtract_8	atomic_subtract_char
284 #define	atomic_subtract_acq_8	atomic_subtract_acq_char
285 #define	atomic_subtract_rel_8	atomic_subtract_rel_char
286 #define	atomic_load_acq_8	atomic_load_acq_char
287 #define	atomic_store_rel_8	atomic_store_rel_char
288 
289 #define	atomic_set_16		atomic_set_short
290 #define	atomic_set_acq_16	atomic_set_acq_short
291 #define	atomic_set_rel_16	atomic_set_rel_short
292 #define	atomic_clear_16		atomic_clear_short
293 #define	atomic_clear_acq_16	atomic_clear_acq_short
294 #define	atomic_clear_rel_16	atomic_clear_rel_short
295 #define	atomic_add_16		atomic_add_short
296 #define	atomic_add_acq_16	atomic_add_acq_short
297 #define	atomic_add_rel_16	atomic_add_rel_short
298 #define	atomic_subtract_16	atomic_subtract_short
299 #define	atomic_subtract_acq_16	atomic_subtract_acq_short
300 #define	atomic_subtract_rel_16	atomic_subtract_rel_short
301 #define	atomic_load_acq_16	atomic_load_acq_short
302 #define	atomic_store_rel_16	atomic_store_rel_short
303 
304 #define	atomic_set_32		atomic_set_int
305 #define	atomic_set_acq_32	atomic_set_acq_int
306 #define	atomic_set_rel_32	atomic_set_rel_int
307 #define	atomic_clear_32		atomic_clear_int
308 #define	atomic_clear_acq_32	atomic_clear_acq_int
309 #define	atomic_clear_rel_32	atomic_clear_rel_int
310 #define	atomic_add_32		atomic_add_int
311 #define	atomic_add_acq_32	atomic_add_acq_int
312 #define	atomic_add_rel_32	atomic_add_rel_int
313 #define	atomic_subtract_32	atomic_subtract_int
314 #define	atomic_subtract_acq_32	atomic_subtract_acq_int
315 #define	atomic_subtract_rel_32	atomic_subtract_rel_int
316 #define	atomic_load_acq_32	atomic_load_acq_int
317 #define	atomic_store_rel_32	atomic_store_rel_int
318 #define	atomic_cmpset_32	atomic_cmpset_int
319 #define	atomic_cmpset_acq_32	atomic_cmpset_acq_int
320 #define	atomic_cmpset_rel_32	atomic_cmpset_rel_int
321 #define	atomic_readandclear_32	atomic_readandclear_int
322 
323 #if !defined(WANT_FUNCTIONS)
324 static __inline int
325 atomic_cmpset_ptr(volatile void *dst, void *exp, void *src)
326 {
327 
328 	return (atomic_cmpset_long((volatile u_long *)dst,
329 	    (u_long)exp, (u_long)src));
330 }
331 
332 static __inline void *
333 atomic_load_acq_ptr(volatile void *p)
334 {
335 	/*
336 	 * The apparently-bogus cast to intptr_t in the following is to
337 	 * avoid a warning from "gcc -Wbad-function-cast".
338 	 */
339 	return ((void *)(intptr_t)atomic_load_acq_long((volatile u_long *)p));
340 }
341 
342 static __inline void
343 atomic_store_rel_ptr(volatile void *p, void *v)
344 {
345 	atomic_store_rel_long((volatile u_long *)p, (u_long)v);
346 }
347 
348 #define ATOMIC_PTR(NAME)				\
349 static __inline void					\
350 atomic_##NAME##_ptr(volatile void *p, uintptr_t v)	\
351 {							\
352 	atomic_##NAME##_long((volatile u_long *)p, v);	\
353 }							\
354 							\
355 static __inline void					\
356 atomic_##NAME##_acq_ptr(volatile void *p, uintptr_t v)	\
357 {							\
358 	atomic_##NAME##_acq_long((volatile u_long *)p, v);\
359 }							\
360 							\
361 static __inline void					\
362 atomic_##NAME##_rel_ptr(volatile void *p, uintptr_t v)	\
363 {							\
364 	atomic_##NAME##_rel_long((volatile u_long *)p, v);\
365 }
366 
367 ATOMIC_PTR(set)
368 ATOMIC_PTR(clear)
369 ATOMIC_PTR(add)
370 ATOMIC_PTR(subtract)
371 
372 #undef ATOMIC_PTR
373 
374 #if defined(__GNUC__)
375 
376 static __inline u_int
377 atomic_readandclear_int(volatile u_int *addr)
378 {
379 	u_int result;
380 
381 	__asm __volatile (
382 	"	xorl	%0,%0 ;		"
383 	"	xchgl	%1,%0 ;		"
384 	"# atomic_readandclear_int"
385 	: "=&r" (result)		/* 0 (result) */
386 	: "m" (*addr));			/* 1 (addr) */
387 
388 	return (result);
389 }
390 
391 static __inline u_long
392 atomic_readandclear_long(volatile u_long *addr)
393 {
394 	u_long result;
395 
396 	__asm __volatile (
397 	"	xorq	%0,%0 ;		"
398 	"	xchgq	%1,%0 ;		"
399 	"# atomic_readandclear_int"
400 	: "=&r" (result)		/* 0 (result) */
401 	: "m" (*addr));			/* 1 (addr) */
402 
403 	return (result);
404 }
405 
406 #else /* !defined(__GNUC__) */
407 
408 extern u_long	atomic_readandclear_long(volatile u_long *);
409 extern u_int	atomic_readandclear_int(volatile u_int *);
410 
411 #endif /* defined(__GNUC__) */
412 
413 #endif	/* !defined(WANT_FUNCTIONS) */
414 #endif /* ! _MACHINE_ATOMIC_H_ */
415