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: releng/8.3/sys/amd64/include/atomic.h 216869 2011-01-01 08:47:38Z kib $
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 void atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
78
79 int atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src);
80 int atomic_cmpset_long(volatile u_long *dst, u_long exp, u_long src);
81 u_int atomic_fetchadd_int(volatile u_int *p, u_int v);
82 u_long atomic_fetchadd_long(volatile u_long *p, u_long v);
83
84 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
85 u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p); \
86 void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
87
88 #else /* !KLD_MODULE && __GNUCLIKE_ASM */
89
90 /*
91 * For userland, always use lock prefixes so that the binaries will run
92 * on both SMP and !SMP systems.
93 */
94 #if defined(SMP) || !defined(_KERNEL)
95 #define MPLOCKED "lock ; "
96 #else
97 #define MPLOCKED
98 #endif
99
100 /*
101 * The assembly is volatilized to avoid code chunk removal by the compiler.
102 * GCC aggressively reorders operations and memory clobbering is necessary
103 * in order to avoid that for memory barriers.
104 */
105 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \
106 static __inline void \
107 atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
108 { \
109 __asm __volatile(MPLOCKED OP \
110 : "=m" (*p) \
111 : CONS (V), "m" (*p) \
112 : "cc"); \
113 } \
114 \
115 static __inline void \
116 atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
117 { \
118 __asm __volatile(MPLOCKED OP \
119 : "=m" (*p) \
120 : CONS (V), "m" (*p) \
121 : "memory", "cc"); \
122 } \
123 struct __hack
124
125 /*
126 * Atomic compare and set, used by the mutex functions
127 *
128 * if (*dst == exp) *dst = src (all 32 bit words)
129 *
130 * Returns 0 on failure, non-zero on success
131 */
132
133 static __inline int
134 atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
135 {
136 u_char res;
137
138 __asm __volatile(
139 " " MPLOCKED " "
140 " cmpxchgl %2,%1 ; "
141 " sete %0 ; "
142 "1: "
143 "# atomic_cmpset_int"
144 : "=a" (res), /* 0 */
145 "=m" (*dst) /* 1 */
146 : "r" (src), /* 2 */
147 "a" (exp), /* 3 */
148 "m" (*dst) /* 4 */
149 : "memory", "cc");
150
151 return (res);
152 }
153
154 static __inline int
155 atomic_cmpset_long(volatile u_long *dst, u_long exp, u_long src)
156 {
157 u_char res;
158
159 __asm __volatile(
160 " " MPLOCKED " "
161 " cmpxchgq %2,%1 ; "
162 " sete %0 ; "
163 "1: "
164 "# atomic_cmpset_long"
165 : "=a" (res), /* 0 */
166 "=m" (*dst) /* 1 */
167 : "r" (src), /* 2 */
168 "a" (exp), /* 3 */
169 "m" (*dst) /* 4 */
170 : "memory", "cc");
171
172 return (res);
173 }
174
175 /*
176 * Atomically add the value of v to the integer pointed to by p and return
177 * the previous value of *p.
178 */
179 static __inline u_int
180 atomic_fetchadd_int(volatile u_int *p, u_int v)
181 {
182
183 __asm __volatile(
184 " " MPLOCKED " "
185 " xaddl %0, %1 ; "
186 "# atomic_fetchadd_int"
187 : "+r" (v), /* 0 (result) */
188 "=m" (*p) /* 1 */
189 : "m" (*p) /* 2 */
190 : "cc");
191 return (v);
192 }
193
194 /*
195 * Atomically add the value of v to the long integer pointed to by p and return
196 * the previous value of *p.
197 */
198 static __inline u_long
199 atomic_fetchadd_long(volatile u_long *p, u_long v)
200 {
201
202 __asm __volatile(
203 " " MPLOCKED " "
204 " xaddq %0, %1 ; "
205 "# atomic_fetchadd_long"
206 : "+r" (v), /* 0 (result) */
207 "=m" (*p) /* 1 */
208 : "m" (*p) /* 2 */
209 : "cc");
210 return (v);
211 }
212
213 #if defined(_KERNEL) && !defined(SMP)
214
215 /*
216 * We assume that a = b will do atomic loads and stores. However, on a
217 * PentiumPro or higher, reads may pass writes, so for that case we have
218 * to use a serializing instruction (i.e. with LOCK) to do the load in
219 * SMP kernels. For UP kernels, however, the cache of the single processor
220 * is always consistent, so we only need to take care of compiler.
221 */
222 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
223 static __inline u_##TYPE \
224 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
225 { \
226 u_##TYPE tmp; \
227 \
228 tmp = *p; \
229 __asm __volatile ("" : : : "memory"); \
230 return (tmp); \
231 } \
232 \
233 static __inline void \
234 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
235 { \
236 __asm __volatile ("" : : : "memory"); \
237 *p = v; \
238 } \
239 struct __hack
240
241 #else /* !(_KERNEL && !SMP) */
242
243 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
244 static __inline u_##TYPE \
245 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
246 { \
247 u_##TYPE res; \
248 \
249 __asm __volatile(MPLOCKED LOP \
250 : "=a" (res), /* 0 */ \
251 "=m" (*p) /* 1 */ \
252 : "m" (*p) /* 2 */ \
253 : "memory", "cc"); \
254 \
255 return (res); \
256 } \
257 \
258 /* \
259 * The XCHG instruction asserts LOCK automagically. \
260 */ \
261 static __inline void \
262 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
263 { \
264 __asm __volatile(SOP \
265 : "=m" (*p), /* 0 */ \
266 "+r" (v) /* 1 */ \
267 : "m" (*p) /* 2 */ \
268 : "memory"); \
269 } \
270 struct __hack
271
272 #endif /* _KERNEL && !SMP */
273
274 #endif /* KLD_MODULE || !__GNUCLIKE_ASM */
275
276 ATOMIC_ASM(set, char, "orb %b1,%0", "iq", v);
277 ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v);
278 ATOMIC_ASM(add, char, "addb %b1,%0", "iq", v);
279 ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v);
280
281 ATOMIC_ASM(set, short, "orw %w1,%0", "ir", v);
282 ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v);
283 ATOMIC_ASM(add, short, "addw %w1,%0", "ir", v);
284 ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v);
285
286 ATOMIC_ASM(set, int, "orl %1,%0", "ir", v);
287 ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v);
288 ATOMIC_ASM(add, int, "addl %1,%0", "ir", v);
289 ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v);
290
291 ATOMIC_ASM(set, long, "orq %1,%0", "ir", v);
292 ATOMIC_ASM(clear, long, "andq %1,%0", "ir", ~v);
293 ATOMIC_ASM(add, long, "addq %1,%0", "ir", v);
294 ATOMIC_ASM(subtract, long, "subq %1,%0", "ir", v);
295
296 ATOMIC_STORE_LOAD(char, "cmpxchgb %b0,%1", "xchgb %b1,%0");
297 ATOMIC_STORE_LOAD(short,"cmpxchgw %w0,%1", "xchgw %w1,%0");
298 ATOMIC_STORE_LOAD(int, "cmpxchgl %0,%1", "xchgl %1,%0");
299 ATOMIC_STORE_LOAD(long, "cmpxchgq %0,%1", "xchgq %1,%0");
300
301 #undef ATOMIC_ASM
302 #undef ATOMIC_STORE_LOAD
303
304 #ifndef WANT_FUNCTIONS
305
306 /* Read the current value and store a zero in the destination. */
307 #ifdef __GNUCLIKE_ASM
308
309 static __inline u_int
310 atomic_readandclear_int(volatile u_int *addr)
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" (*addr) /* 1 */
320 : "m" (*addr));
321
322 return (res);
323 }
324
325 static __inline u_long
326 atomic_readandclear_long(volatile u_long *addr)
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" (*addr) /* 1 */
336 : "m" (*addr));
337
338 return (res);
339 }
340
341 #else /* !__GNUCLIKE_ASM */
342
343 u_int atomic_readandclear_int(volatile u_int *addr);
344 u_long atomic_readandclear_long(volatile u_long *addr);
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_ */
Cache object: 30ebabaa370f4e2f494c41e48dadf069
|