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: src/sys/amd64/include/atomic.h,v 1.46 2008/11/22 05:55:56 kmacy Exp $
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) \
76 void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
77
78 int atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src);
79 int atomic_cmpset_long(volatile u_long *dst, u_long exp, u_long src);
80 u_int atomic_fetchadd_int(volatile u_int *p, u_int v);
81 u_long atomic_fetchadd_long(volatile u_long *p, u_long v);
82
83 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
84 u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p); \
85 void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
86
87 #else /* !KLD_MODULE && __GNUCLIKE_ASM */
88
89 /*
90 * For userland, always use lock prefixes so that the binaries will run
91 * on both SMP and !SMP systems.
92 */
93 #if defined(SMP) || !defined(_KERNEL)
94 #define MPLOCKED "lock ; "
95 #else
96 #define MPLOCKED
97 #endif
98
99 /*
100 * The assembly is volatilized to demark potential before-and-after side
101 * effects if an interrupt or SMP collision were to occur.
102 */
103 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \
104 static __inline void \
105 atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
106 { \
107 __asm __volatile(MPLOCKED OP \
108 : "=m" (*p) \
109 : CONS (V), "m" (*p)); \
110 } \
111 struct __hack
112
113 /*
114 * Atomic compare and set, used by the mutex functions
115 *
116 * if (*dst == exp) *dst = src (all 32 bit words)
117 *
118 * Returns 0 on failure, non-zero on success
119 */
120
121 static __inline int
122 atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
123 {
124 u_char res;
125
126 __asm __volatile(
127 " " MPLOCKED " "
128 " cmpxchgl %2,%1 ; "
129 " sete %0 ; "
130 "1: "
131 "# atomic_cmpset_int"
132 : "=a" (res), /* 0 */
133 "=m" (*dst) /* 1 */
134 : "r" (src), /* 2 */
135 "a" (exp), /* 3 */
136 "m" (*dst) /* 4 */
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 u_char res;
146
147 __asm __volatile(
148 " " MPLOCKED " "
149 " cmpxchgq %2,%1 ; "
150 " sete %0 ; "
151 "1: "
152 "# atomic_cmpset_long"
153 : "=a" (res), /* 0 */
154 "=m" (*dst) /* 1 */
155 : "r" (src), /* 2 */
156 "a" (exp), /* 3 */
157 "m" (*dst) /* 4 */
158 : "memory");
159
160 return (res);
161 }
162
163 /*
164 * Atomically add the value of v to the integer pointed to by p and return
165 * the previous value of *p.
166 */
167 static __inline u_int
168 atomic_fetchadd_int(volatile u_int *p, u_int v)
169 {
170
171 __asm __volatile(
172 " " MPLOCKED " "
173 " xaddl %0, %1 ; "
174 "# atomic_fetchadd_int"
175 : "+r" (v), /* 0 (result) */
176 "=m" (*p) /* 1 */
177 : "m" (*p)); /* 2 */
178
179 return (v);
180 }
181
182 /*
183 * Atomically add the value of v to the long integer pointed to by p and return
184 * the previous value of *p.
185 */
186 static __inline u_long
187 atomic_fetchadd_long(volatile u_long *p, u_long v)
188 {
189
190 __asm __volatile(
191 " " MPLOCKED " "
192 " xaddq %0, %1 ; "
193 "# atomic_fetchadd_long"
194 : "+r" (v), /* 0 (result) */
195 "=m" (*p) /* 1 */
196 : "m" (*p)); /* 2 */
197
198 return (v);
199 }
200
201 #if defined(_KERNEL) && !defined(SMP)
202
203 /*
204 * We assume that a = b will do atomic loads and stores. However, on a
205 * PentiumPro or higher, reads may pass writes, so for that case we have
206 * to use a serializing instruction (i.e. with LOCK) to do the load in
207 * SMP kernels. For UP kernels, however, the cache of the single processor
208 * is always consistent, so we don't need any memory barriers.
209 */
210 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
211 static __inline u_##TYPE \
212 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
213 { \
214 return (*p); \
215 } \
216 \
217 static __inline void \
218 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
219 { \
220 *p = v; \
221 } \
222 struct __hack
223
224 #else /* !(_KERNEL && !SMP) */
225
226 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
227 static __inline u_##TYPE \
228 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
229 { \
230 u_##TYPE res; \
231 \
232 __asm __volatile(MPLOCKED LOP \
233 : "=a" (res), /* 0 */ \
234 "=m" (*p) /* 1 */ \
235 : "m" (*p) /* 2 */ \
236 : "memory"); \
237 \
238 return (res); \
239 } \
240 \
241 /* \
242 * The XCHG instruction asserts LOCK automagically. \
243 */ \
244 static __inline void \
245 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
246 { \
247 __asm __volatile(SOP \
248 : "=m" (*p), /* 0 */ \
249 "+r" (v) /* 1 */ \
250 : "m" (*p)); /* 2 */ \
251 } \
252 struct __hack
253
254 #endif /* _KERNEL && !SMP */
255
256 #endif /* KLD_MODULE || !__GNUCLIKE_ASM */
257
258 ATOMIC_ASM(set, char, "orb %b1,%0", "iq", v);
259 ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v);
260 ATOMIC_ASM(add, char, "addb %b1,%0", "iq", v);
261 ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v);
262
263 ATOMIC_ASM(set, short, "orw %w1,%0", "ir", v);
264 ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v);
265 ATOMIC_ASM(add, short, "addw %w1,%0", "ir", v);
266 ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v);
267
268 ATOMIC_ASM(set, int, "orl %1,%0", "ir", v);
269 ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v);
270 ATOMIC_ASM(add, int, "addl %1,%0", "ir", v);
271 ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v);
272
273 ATOMIC_ASM(set, long, "orq %1,%0", "ir", v);
274 ATOMIC_ASM(clear, long, "andq %1,%0", "ir", ~v);
275 ATOMIC_ASM(add, long, "addq %1,%0", "ir", v);
276 ATOMIC_ASM(subtract, long, "subq %1,%0", "ir", v);
277
278 ATOMIC_STORE_LOAD(char, "cmpxchgb %b0,%1", "xchgb %b1,%0");
279 ATOMIC_STORE_LOAD(short,"cmpxchgw %w0,%1", "xchgw %w1,%0");
280 ATOMIC_STORE_LOAD(int, "cmpxchgl %0,%1", "xchgl %1,%0");
281 ATOMIC_STORE_LOAD(long, "cmpxchgq %0,%1", "xchgq %1,%0");
282
283 #undef ATOMIC_ASM
284 #undef ATOMIC_STORE_LOAD
285
286 #ifndef WANT_FUNCTIONS
287
288 /* Read the current value and store a zero in the destination. */
289 #ifdef __GNUCLIKE_ASM
290
291 static __inline u_int
292 atomic_readandclear_int(volatile u_int *addr)
293 {
294 u_int res;
295
296 res = 0;
297 __asm __volatile(
298 " xchgl %1,%0 ; "
299 "# atomic_readandclear_int"
300 : "+r" (res), /* 0 */
301 "=m" (*addr) /* 1 */
302 : "m" (*addr));
303
304 return (res);
305 }
306
307 static __inline u_long
308 atomic_readandclear_long(volatile u_long *addr)
309 {
310 u_long res;
311
312 res = 0;
313 __asm __volatile(
314 " xchgq %1,%0 ; "
315 "# atomic_readandclear_long"
316 : "+r" (res), /* 0 */
317 "=m" (*addr) /* 1 */
318 : "m" (*addr));
319
320 return (res);
321 }
322
323 #else /* !__GNUCLIKE_ASM */
324
325 u_int atomic_readandclear_int(volatile u_int *addr);
326 u_long atomic_readandclear_long(volatile u_long *addr);
327
328 #endif /* __GNUCLIKE_ASM */
329
330 /* Acquire and release variants are identical to the normal ones. */
331 #define atomic_set_acq_char atomic_set_char
332 #define atomic_set_rel_char atomic_set_char
333 #define atomic_clear_acq_char atomic_clear_char
334 #define atomic_clear_rel_char atomic_clear_char
335 #define atomic_add_acq_char atomic_add_char
336 #define atomic_add_rel_char atomic_add_char
337 #define atomic_subtract_acq_char atomic_subtract_char
338 #define atomic_subtract_rel_char atomic_subtract_char
339
340 #define atomic_set_acq_short atomic_set_short
341 #define atomic_set_rel_short atomic_set_short
342 #define atomic_clear_acq_short atomic_clear_short
343 #define atomic_clear_rel_short atomic_clear_short
344 #define atomic_add_acq_short atomic_add_short
345 #define atomic_add_rel_short atomic_add_short
346 #define atomic_subtract_acq_short atomic_subtract_short
347 #define atomic_subtract_rel_short atomic_subtract_short
348
349 #define atomic_set_acq_int atomic_set_int
350 #define atomic_set_rel_int atomic_set_int
351 #define atomic_clear_acq_int atomic_clear_int
352 #define atomic_clear_rel_int atomic_clear_int
353 #define atomic_add_acq_int atomic_add_int
354 #define atomic_add_rel_int atomic_add_int
355 #define atomic_subtract_acq_int atomic_subtract_int
356 #define atomic_subtract_rel_int atomic_subtract_int
357 #define atomic_cmpset_acq_int atomic_cmpset_int
358 #define atomic_cmpset_rel_int atomic_cmpset_int
359
360 #define atomic_set_acq_long atomic_set_long
361 #define atomic_set_rel_long atomic_set_long
362 #define atomic_clear_acq_long atomic_clear_long
363 #define atomic_clear_rel_long atomic_clear_long
364 #define atomic_add_acq_long atomic_add_long
365 #define atomic_add_rel_long atomic_add_long
366 #define atomic_subtract_acq_long atomic_subtract_long
367 #define atomic_subtract_rel_long atomic_subtract_long
368 #define atomic_cmpset_acq_long atomic_cmpset_long
369 #define atomic_cmpset_rel_long atomic_cmpset_long
370
371 /* Operations on 8-bit bytes. */
372 #define atomic_set_8 atomic_set_char
373 #define atomic_set_acq_8 atomic_set_acq_char
374 #define atomic_set_rel_8 atomic_set_rel_char
375 #define atomic_clear_8 atomic_clear_char
376 #define atomic_clear_acq_8 atomic_clear_acq_char
377 #define atomic_clear_rel_8 atomic_clear_rel_char
378 #define atomic_add_8 atomic_add_char
379 #define atomic_add_acq_8 atomic_add_acq_char
380 #define atomic_add_rel_8 atomic_add_rel_char
381 #define atomic_subtract_8 atomic_subtract_char
382 #define atomic_subtract_acq_8 atomic_subtract_acq_char
383 #define atomic_subtract_rel_8 atomic_subtract_rel_char
384 #define atomic_load_acq_8 atomic_load_acq_char
385 #define atomic_store_rel_8 atomic_store_rel_char
386
387 /* Operations on 16-bit words. */
388 #define atomic_set_16 atomic_set_short
389 #define atomic_set_acq_16 atomic_set_acq_short
390 #define atomic_set_rel_16 atomic_set_rel_short
391 #define atomic_clear_16 atomic_clear_short
392 #define atomic_clear_acq_16 atomic_clear_acq_short
393 #define atomic_clear_rel_16 atomic_clear_rel_short
394 #define atomic_add_16 atomic_add_short
395 #define atomic_add_acq_16 atomic_add_acq_short
396 #define atomic_add_rel_16 atomic_add_rel_short
397 #define atomic_subtract_16 atomic_subtract_short
398 #define atomic_subtract_acq_16 atomic_subtract_acq_short
399 #define atomic_subtract_rel_16 atomic_subtract_rel_short
400 #define atomic_load_acq_16 atomic_load_acq_short
401 #define atomic_store_rel_16 atomic_store_rel_short
402
403 /* Operations on 32-bit double words. */
404 #define atomic_set_32 atomic_set_int
405 #define atomic_set_acq_32 atomic_set_acq_int
406 #define atomic_set_rel_32 atomic_set_rel_int
407 #define atomic_clear_32 atomic_clear_int
408 #define atomic_clear_acq_32 atomic_clear_acq_int
409 #define atomic_clear_rel_32 atomic_clear_rel_int
410 #define atomic_add_32 atomic_add_int
411 #define atomic_add_acq_32 atomic_add_acq_int
412 #define atomic_add_rel_32 atomic_add_rel_int
413 #define atomic_subtract_32 atomic_subtract_int
414 #define atomic_subtract_acq_32 atomic_subtract_acq_int
415 #define atomic_subtract_rel_32 atomic_subtract_rel_int
416 #define atomic_load_acq_32 atomic_load_acq_int
417 #define atomic_store_rel_32 atomic_store_rel_int
418 #define atomic_cmpset_32 atomic_cmpset_int
419 #define atomic_cmpset_acq_32 atomic_cmpset_acq_int
420 #define atomic_cmpset_rel_32 atomic_cmpset_rel_int
421 #define atomic_readandclear_32 atomic_readandclear_int
422 #define atomic_fetchadd_32 atomic_fetchadd_int
423
424 /* Operations on 64-bit quad words. */
425 #define atomic_set_64 atomic_set_long
426 #define atomic_set_acq_64 atomic_set_acq_long
427 #define atomic_set_rel_64 atomic_set_rel_long
428 #define atomic_clear_64 atomic_clear_long
429 #define atomic_clear_acq_64 atomic_clear_acq_long
430 #define atomic_clear_rel_64 atomic_clear_rel_long
431 #define atomic_add_64 atomic_add_long
432 #define atomic_add_acq_64 atomic_add_acq_long
433 #define atomic_add_rel_64 atomic_add_rel_long
434 #define atomic_subtract_64 atomic_subtract_long
435 #define atomic_subtract_acq_64 atomic_subtract_acq_long
436 #define atomic_subtract_rel_64 atomic_subtract_rel_long
437 #define atomic_load_acq_64 atomic_load_acq_long
438 #define atomic_store_rel_64 atomic_store_rel_long
439 #define atomic_cmpset_64 atomic_cmpset_long
440 #define atomic_cmpset_acq_64 atomic_cmpset_acq_long
441 #define atomic_cmpset_rel_64 atomic_cmpset_rel_long
442 #define atomic_readandclear_64 atomic_readandclear_long
443
444 /* Operations on pointers. */
445 #define atomic_set_ptr atomic_set_long
446 #define atomic_set_acq_ptr atomic_set_acq_long
447 #define atomic_set_rel_ptr atomic_set_rel_long
448 #define atomic_clear_ptr atomic_clear_long
449 #define atomic_clear_acq_ptr atomic_clear_acq_long
450 #define atomic_clear_rel_ptr atomic_clear_rel_long
451 #define atomic_add_ptr atomic_add_long
452 #define atomic_add_acq_ptr atomic_add_acq_long
453 #define atomic_add_rel_ptr atomic_add_rel_long
454 #define atomic_subtract_ptr atomic_subtract_long
455 #define atomic_subtract_acq_ptr atomic_subtract_acq_long
456 #define atomic_subtract_rel_ptr atomic_subtract_rel_long
457 #define atomic_load_acq_ptr atomic_load_acq_long
458 #define atomic_store_rel_ptr atomic_store_rel_long
459 #define atomic_cmpset_ptr atomic_cmpset_long
460 #define atomic_cmpset_acq_ptr atomic_cmpset_acq_long
461 #define atomic_cmpset_rel_ptr atomic_cmpset_rel_long
462 #define atomic_readandclear_ptr atomic_readandclear_long
463
464 #endif /* !WANT_FUNCTIONS */
465
466 #endif /* !_MACHINE_ATOMIC_H_ */
467 Cache object: becc44dc30a5257f44d23681a1d22cea
|