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 #ifndef _SYS_CDEFS_H_
32 #error this file needs sys/cdefs.h as a prerequisite
33 #endif
34
35 /*
36 * Various simple operations on memory, each of which is atomic in the
37 * presence of interrupts and multiple processors.
38 *
39 * atomic_set_char(P, V) (*(u_char *)(P) |= (V))
40 * atomic_clear_char(P, V) (*(u_char *)(P) &= ~(V))
41 * atomic_add_char(P, V) (*(u_char *)(P) += (V))
42 * atomic_subtract_char(P, V) (*(u_char *)(P) -= (V))
43 *
44 * atomic_set_short(P, V) (*(u_short *)(P) |= (V))
45 * atomic_clear_short(P, V) (*(u_short *)(P) &= ~(V))
46 * atomic_add_short(P, V) (*(u_short *)(P) += (V))
47 * atomic_subtract_short(P, V) (*(u_short *)(P) -= (V))
48 *
49 * atomic_set_int(P, V) (*(u_int *)(P) |= (V))
50 * atomic_clear_int(P, V) (*(u_int *)(P) &= ~(V))
51 * atomic_add_int(P, V) (*(u_int *)(P) += (V))
52 * atomic_subtract_int(P, V) (*(u_int *)(P) -= (V))
53 * atomic_readandclear_int(P) (return (*(u_int *)(P)); *(u_int *)(P) = 0;)
54 *
55 * atomic_set_long(P, V) (*(u_long *)(P) |= (V))
56 * atomic_clear_long(P, V) (*(u_long *)(P) &= ~(V))
57 * atomic_add_long(P, V) (*(u_long *)(P) += (V))
58 * atomic_subtract_long(P, V) (*(u_long *)(P) -= (V))
59 * atomic_readandclear_long(P) (return (*(u_long *)(P)); *(u_long *)(P) = 0;)
60 */
61
62 /*
63 * The above functions are expanded inline in the statically-linked
64 * kernel. Lock prefixes are generated if an SMP kernel is being
65 * built.
66 *
67 * Kernel modules call real functions which are built into the kernel.
68 * This allows kernel modules to be portable between UP and SMP systems.
69 */
70 #if defined(KLD_MODULE) || !defined(__GNUCLIKE_ASM)
71 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \
72 void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v); \
73 void atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
74
75 int atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src);
76 u_int atomic_fetchadd_int(volatile u_int *p, u_int v);
77
78 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
79 u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p); \
80 void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
81
82 #else /* !KLD_MODULE && __GNUCLIKE_ASM */
83
84 /*
85 * For userland, always use lock prefixes so that the binaries will run
86 * on both SMP and !SMP systems.
87 */
88 #if defined(SMP) || !defined(_KERNEL)
89 #define MPLOCKED "lock ; "
90 #else
91 #define MPLOCKED
92 #endif
93
94 /*
95 * The assembly is volatilized to avoid code chunk removal by the compiler.
96 * GCC aggressively reorders operations and memory clobbering is necessary
97 * in order to avoid that for memory barriers.
98 */
99 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \
100 static __inline void \
101 atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
102 { \
103 __asm __volatile(MPLOCKED OP \
104 : "=m" (*p) \
105 : CONS (V), "m" (*p)); \
106 } \
107 \
108 static __inline void \
109 atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
110 { \
111 __asm __volatile(MPLOCKED OP \
112 : "=m" (*p) \
113 : CONS (V), "m" (*p) \
114 : "memory"); \
115 } \
116 struct __hack
117
118 /*
119 * Atomic compare and set, used by the mutex functions
120 *
121 * if (*dst == exp) *dst = src (all 32 bit words)
122 *
123 * Returns 0 on failure, non-zero on success
124 */
125
126 #ifdef CPU_DISABLE_CMPXCHG
127
128 static __inline int
129 atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
130 {
131 u_char res;
132
133 __asm __volatile(
134 " pushfl ; "
135 " cli ; "
136 " cmpl %3,%4 ; "
137 " jne 1f ; "
138 " movl %2,%1 ; "
139 "1: "
140 " sete %0 ; "
141 " popfl ; "
142 "# atomic_cmpset_int"
143 : "=q" (res), /* 0 */
144 "=m" (*dst) /* 1 */
145 : "r" (src), /* 2 */
146 "r" (exp), /* 3 */
147 "m" (*dst) /* 4 */
148 : "memory");
149
150 return (res);
151 }
152
153 #else /* !CPU_DISABLE_CMPXCHG */
154
155 static __inline int
156 atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
157 {
158 u_char res;
159
160 __asm __volatile(
161 " " MPLOCKED " "
162 " cmpxchgl %2,%1 ; "
163 " sete %0 ; "
164 "1: "
165 "# atomic_cmpset_int"
166 : "=a" (res), /* 0 */
167 "=m" (*dst) /* 1 */
168 : "r" (src), /* 2 */
169 "a" (exp), /* 3 */
170 "m" (*dst) /* 4 */
171 : "memory");
172
173 return (res);
174 }
175
176 #endif /* CPU_DISABLE_CMPXCHG */
177
178 /*
179 * Atomically add the value of v to the integer pointed to by p and return
180 * the previous value of *p.
181 */
182 static __inline u_int
183 atomic_fetchadd_int(volatile u_int *p, u_int v)
184 {
185
186 __asm __volatile(
187 " " MPLOCKED " "
188 " xaddl %0, %1 ; "
189 "# atomic_fetchadd_int"
190 : "+r" (v), /* 0 (result) */
191 "=m" (*p) /* 1 */
192 : "m" (*p)); /* 2 */
193
194 return (v);
195 }
196
197 #if defined(_KERNEL) && !defined(SMP)
198
199 /*
200 * We assume that a = b will do atomic loads and stores. However, on a
201 * PentiumPro or higher, reads may pass writes, so for that case we have
202 * to use a serializing instruction (i.e. with LOCK) to do the load in
203 * SMP kernels. For UP kernels, however, the cache of the single processor
204 * is always consistent, so we only need to take care of compiler.
205 */
206 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
207 static __inline u_##TYPE \
208 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
209 { \
210 u_##TYPE tmp; \
211 \
212 tmp = *p; \
213 __asm __volatile("" : : : "memory"); \
214 return (tmp); \
215 } \
216 \
217 static __inline void \
218 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
219 { \
220 __asm __volatile("" : : : "memory"); \
221 *p = v; \
222 } \
223 struct __hack
224
225 #else /* !(_KERNEL && !SMP) */
226
227 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
228 static __inline u_##TYPE \
229 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
230 { \
231 u_##TYPE res; \
232 \
233 __asm __volatile(MPLOCKED LOP \
234 : "=a" (res), /* 0 */ \
235 "=m" (*p) /* 1 */ \
236 : "m" (*p) /* 2 */ \
237 : "memory"); \
238 \
239 return (res); \
240 } \
241 \
242 /* \
243 * The XCHG instruction asserts LOCK automagically. \
244 */ \
245 static __inline void \
246 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
247 { \
248 __asm __volatile(SOP \
249 : "=m" (*p), /* 0 */ \
250 "+r" (v) /* 1 */ \
251 : "m" (*p) /* 2 */ \
252 : "memory"); \
253 } \
254 struct __hack
255
256 #endif /* _KERNEL && !SMP */
257
258 #endif /* KLD_MODULE || !__GNUCLIKE_ASM */
259
260 ATOMIC_ASM(set, char, "orb %b1,%0", "iq", v);
261 ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v);
262 ATOMIC_ASM(add, char, "addb %b1,%0", "iq", v);
263 ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v);
264
265 ATOMIC_ASM(set, short, "orw %w1,%0", "ir", v);
266 ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v);
267 ATOMIC_ASM(add, short, "addw %w1,%0", "ir", v);
268 ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v);
269
270 ATOMIC_ASM(set, int, "orl %1,%0", "ir", v);
271 ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v);
272 ATOMIC_ASM(add, int, "addl %1,%0", "ir", v);
273 ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v);
274
275 ATOMIC_ASM(set, long, "orl %1,%0", "ir", v);
276 ATOMIC_ASM(clear, long, "andl %1,%0", "ir", ~v);
277 ATOMIC_ASM(add, long, "addl %1,%0", "ir", v);
278 ATOMIC_ASM(subtract, long, "subl %1,%0", "ir", v);
279
280 ATOMIC_STORE_LOAD(char, "cmpxchgb %b0,%1", "xchgb %b1,%0");
281 ATOMIC_STORE_LOAD(short,"cmpxchgw %w0,%1", "xchgw %w1,%0");
282 ATOMIC_STORE_LOAD(int, "cmpxchgl %0,%1", "xchgl %1,%0");
283 ATOMIC_STORE_LOAD(long, "cmpxchgl %0,%1", "xchgl %1,%0");
284
285 #undef ATOMIC_ASM
286 #undef ATOMIC_STORE_LOAD
287
288 #ifndef WANT_FUNCTIONS
289
290 static __inline int
291 atomic_cmpset_long(volatile u_long *dst, u_long exp, u_long src)
292 {
293
294 return (atomic_cmpset_int((volatile u_int *)dst, (u_int)exp,
295 (u_int)src));
296 }
297
298 static __inline u_long
299 atomic_fetchadd_long(volatile u_long *p, u_long v)
300 {
301
302 return (atomic_fetchadd_int((volatile u_int *)p, (u_int)v));
303 }
304
305 /* Read the current value and store a zero in the destination. */
306 #ifdef __GNUCLIKE_ASM
307
308 static __inline u_int
309 atomic_readandclear_int(volatile u_int *addr)
310 {
311 u_int res;
312
313 res = 0;
314 __asm __volatile(
315 " xchgl %1,%0 ; "
316 "# atomic_readandclear_int"
317 : "+r" (res), /* 0 */
318 "=m" (*addr) /* 1 */
319 : "m" (*addr));
320
321 return (res);
322 }
323
324 static __inline u_long
325 atomic_readandclear_long(volatile u_long *addr)
326 {
327 u_long res;
328
329 res = 0;
330 __asm __volatile(
331 " xchgl %1,%0 ; "
332 "# atomic_readandclear_long"
333 : "+r" (res), /* 0 */
334 "=m" (*addr) /* 1 */
335 : "m" (*addr));
336
337 return (res);
338 }
339
340 #else /* !__GNUCLIKE_ASM */
341
342 u_int atomic_readandclear_int(volatile u_int *addr);
343 u_long atomic_readandclear_long(volatile u_long *addr);
344
345 #endif /* __GNUCLIKE_ASM */
346
347 #define atomic_set_acq_char atomic_set_barr_char
348 #define atomic_set_rel_char atomic_set_barr_char
349 #define atomic_clear_acq_char atomic_clear_barr_char
350 #define atomic_clear_rel_char atomic_clear_barr_char
351 #define atomic_add_acq_char atomic_add_barr_char
352 #define atomic_add_rel_char atomic_add_barr_char
353 #define atomic_subtract_acq_char atomic_subtract_barr_char
354 #define atomic_subtract_rel_char atomic_subtract_barr_char
355
356 #define atomic_set_acq_short atomic_set_barr_short
357 #define atomic_set_rel_short atomic_set_barr_short
358 #define atomic_clear_acq_short atomic_clear_barr_short
359 #define atomic_clear_rel_short atomic_clear_barr_short
360 #define atomic_add_acq_short atomic_add_barr_short
361 #define atomic_add_rel_short atomic_add_barr_short
362 #define atomic_subtract_acq_short atomic_subtract_barr_short
363 #define atomic_subtract_rel_short atomic_subtract_barr_short
364
365 #define atomic_set_acq_int atomic_set_barr_int
366 #define atomic_set_rel_int atomic_set_barr_int
367 #define atomic_clear_acq_int atomic_clear_barr_int
368 #define atomic_clear_rel_int atomic_clear_barr_int
369 #define atomic_add_acq_int atomic_add_barr_int
370 #define atomic_add_rel_int atomic_add_barr_int
371 #define atomic_subtract_acq_int atomic_subtract_barr_int
372 #define atomic_subtract_rel_int atomic_subtract_barr_int
373 #define atomic_cmpset_acq_int atomic_cmpset_int
374 #define atomic_cmpset_rel_int atomic_cmpset_int
375
376 #define atomic_set_acq_long atomic_set_barr_long
377 #define atomic_set_rel_long atomic_set_barr_long
378 #define atomic_clear_acq_long atomic_clear_barr_long
379 #define atomic_clear_rel_long atomic_clear_barr_long
380 #define atomic_add_acq_long atomic_add_barr_long
381 #define atomic_add_rel_long atomic_add_barr_long
382 #define atomic_subtract_acq_long atomic_subtract_barr_long
383 #define atomic_subtract_rel_long atomic_subtract_barr_long
384 #define atomic_cmpset_acq_long atomic_cmpset_long
385 #define atomic_cmpset_rel_long atomic_cmpset_long
386
387 /* Operations on 8-bit bytes. */
388 #define atomic_set_8 atomic_set_char
389 #define atomic_set_acq_8 atomic_set_acq_char
390 #define atomic_set_rel_8 atomic_set_rel_char
391 #define atomic_clear_8 atomic_clear_char
392 #define atomic_clear_acq_8 atomic_clear_acq_char
393 #define atomic_clear_rel_8 atomic_clear_rel_char
394 #define atomic_add_8 atomic_add_char
395 #define atomic_add_acq_8 atomic_add_acq_char
396 #define atomic_add_rel_8 atomic_add_rel_char
397 #define atomic_subtract_8 atomic_subtract_char
398 #define atomic_subtract_acq_8 atomic_subtract_acq_char
399 #define atomic_subtract_rel_8 atomic_subtract_rel_char
400 #define atomic_load_acq_8 atomic_load_acq_char
401 #define atomic_store_rel_8 atomic_store_rel_char
402
403 /* Operations on 16-bit words. */
404 #define atomic_set_16 atomic_set_short
405 #define atomic_set_acq_16 atomic_set_acq_short
406 #define atomic_set_rel_16 atomic_set_rel_short
407 #define atomic_clear_16 atomic_clear_short
408 #define atomic_clear_acq_16 atomic_clear_acq_short
409 #define atomic_clear_rel_16 atomic_clear_rel_short
410 #define atomic_add_16 atomic_add_short
411 #define atomic_add_acq_16 atomic_add_acq_short
412 #define atomic_add_rel_16 atomic_add_rel_short
413 #define atomic_subtract_16 atomic_subtract_short
414 #define atomic_subtract_acq_16 atomic_subtract_acq_short
415 #define atomic_subtract_rel_16 atomic_subtract_rel_short
416 #define atomic_load_acq_16 atomic_load_acq_short
417 #define atomic_store_rel_16 atomic_store_rel_short
418
419 /* Operations on 32-bit double words. */
420 #define atomic_set_32 atomic_set_int
421 #define atomic_set_acq_32 atomic_set_acq_int
422 #define atomic_set_rel_32 atomic_set_rel_int
423 #define atomic_clear_32 atomic_clear_int
424 #define atomic_clear_acq_32 atomic_clear_acq_int
425 #define atomic_clear_rel_32 atomic_clear_rel_int
426 #define atomic_add_32 atomic_add_int
427 #define atomic_add_acq_32 atomic_add_acq_int
428 #define atomic_add_rel_32 atomic_add_rel_int
429 #define atomic_subtract_32 atomic_subtract_int
430 #define atomic_subtract_acq_32 atomic_subtract_acq_int
431 #define atomic_subtract_rel_32 atomic_subtract_rel_int
432 #define atomic_load_acq_32 atomic_load_acq_int
433 #define atomic_store_rel_32 atomic_store_rel_int
434 #define atomic_cmpset_32 atomic_cmpset_int
435 #define atomic_cmpset_acq_32 atomic_cmpset_acq_int
436 #define atomic_cmpset_rel_32 atomic_cmpset_rel_int
437 #define atomic_readandclear_32 atomic_readandclear_int
438 #define atomic_fetchadd_32 atomic_fetchadd_int
439
440 /* Operations on pointers. */
441 #define atomic_set_ptr(p, v) \
442 atomic_set_int((volatile u_int *)(p), (u_int)(v))
443 #define atomic_set_acq_ptr(p, v) \
444 atomic_set_acq_int((volatile u_int *)(p), (u_int)(v))
445 #define atomic_set_rel_ptr(p, v) \
446 atomic_set_rel_int((volatile u_int *)(p), (u_int)(v))
447 #define atomic_clear_ptr(p, v) \
448 atomic_clear_int((volatile u_int *)(p), (u_int)(v))
449 #define atomic_clear_acq_ptr(p, v) \
450 atomic_clear_acq_int((volatile u_int *)(p), (u_int)(v))
451 #define atomic_clear_rel_ptr(p, v) \
452 atomic_clear_rel_int((volatile u_int *)(p), (u_int)(v))
453 #define atomic_add_ptr(p, v) \
454 atomic_add_int((volatile u_int *)(p), (u_int)(v))
455 #define atomic_add_acq_ptr(p, v) \
456 atomic_add_acq_int((volatile u_int *)(p), (u_int)(v))
457 #define atomic_add_rel_ptr(p, v) \
458 atomic_add_rel_int((volatile u_int *)(p), (u_int)(v))
459 #define atomic_subtract_ptr(p, v) \
460 atomic_subtract_int((volatile u_int *)(p), (u_int)(v))
461 #define atomic_subtract_acq_ptr(p, v) \
462 atomic_subtract_acq_int((volatile u_int *)(p), (u_int)(v))
463 #define atomic_subtract_rel_ptr(p, v) \
464 atomic_subtract_rel_int((volatile u_int *)(p), (u_int)(v))
465 #define atomic_load_acq_ptr(p) \
466 atomic_load_acq_int((volatile u_int *)(p))
467 #define atomic_store_rel_ptr(p, v) \
468 atomic_store_rel_int((volatile u_int *)(p), (v))
469 #define atomic_cmpset_ptr(dst, old, new) \
470 atomic_cmpset_int((volatile u_int *)(dst), (u_int)(old), (u_int)(new))
471 #define atomic_cmpset_acq_ptr(dst, old, new) \
472 atomic_cmpset_acq_int((volatile u_int *)(dst), (u_int)(old), \
473 (u_int)(new))
474 #define atomic_cmpset_rel_ptr(dst, old, new) \
475 atomic_cmpset_rel_int((volatile u_int *)(dst), (u_int)(old), \
476 (u_int)(new))
477 #define atomic_readandclear_ptr(p) \
478 atomic_readandclear_int((volatile u_int *)(p))
479
480 #endif /* !WANT_FUNCTIONS */
481
482 #endif /* !_MACHINE_ATOMIC_H_ */
Cache object: 2911b9f69ad6c51a9c48c210dccd9a09
|