atomic.h revision 254614
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 254614 2013-08-21 21:30:06Z 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) \ 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) \ 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 %3,%1 ; " 142 " sete %0 ; " 143 "# atomic_cmpset_int" 144 : "=q" (res), /* 0 */ 145 "+m" (*dst), /* 1 */ 146 "+a" (expect) /* 2 */ 147 : "r" (src) /* 3 */ 148 : "memory", "cc"); 149 150 return (res); 151} 152 153static __inline int 154atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src) 155{ 156 u_char res; 157 158 __asm __volatile( 159 " " MPLOCKED " " 160 " cmpxchgq %3,%1 ; " 161 " sete %0 ; " 162 "# atomic_cmpset_long" 163 : "=q" (res), /* 0 */ 164 "+m" (*dst), /* 1 */ 165 "+a" (expect) /* 2 */ 166 : "r" (src) /* 3 */ 167 : "memory", "cc"); 168 169 return (res); 170} 171 172/* 173 * Atomically add the value of v to the integer pointed to by p and return 174 * the previous value of *p. 175 */ 176static __inline u_int 177atomic_fetchadd_int(volatile u_int *p, u_int v) 178{ 179 180 __asm __volatile( 181 " " MPLOCKED " " 182 " xaddl %0,%1 ; " 183 "# atomic_fetchadd_int" 184 : "+r" (v), /* 0 */ 185 "+m" (*p) /* 1 */ 186 : : "cc"); 187 return (v); 188} 189 190/* 191 * Atomically add the value of v to the long integer pointed to by p and return 192 * the previous value of *p. 193 */ 194static __inline u_long 195atomic_fetchadd_long(volatile u_long *p, u_long v) 196{ 197 198 __asm __volatile( 199 " " MPLOCKED " " 200 " xaddq %0,%1 ; " 201 "# atomic_fetchadd_long" 202 : "+r" (v), /* 0 */ 203 "+m" (*p) /* 1 */ 204 : : "cc"); 205 return (v); 206} 207 208/* 209 * We assume that a = b will do atomic loads and stores. Due to the 210 * IA32 memory model, a simple store guarantees release semantics. 211 * 212 * However, loads may pass stores, so for atomic_load_acq we have to 213 * ensure a Store/Load barrier to do the load in SMP kernels. We use 214 * "lock cmpxchg" as recommended by the AMD Software Optimization 215 * Guide, and not mfence. For UP kernels, however, the cache of the 216 * single processor is always consistent, so we only need to take care 217 * of the compiler. 218 */ 219#define ATOMIC_STORE(TYPE) \ 220static __inline void \ 221atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ 222{ \ 223 __compiler_membar(); \ 224 *p = v; \ 225} \ 226struct __hack 227 228#if defined(_KERNEL) && !defined(SMP) 229 230#define ATOMIC_LOAD(TYPE, LOP) \ 231static __inline u_##TYPE \ 232atomic_load_acq_##TYPE(volatile u_##TYPE *p) \ 233{ \ 234 u_##TYPE tmp; \ 235 \ 236 tmp = *p; \ 237 __compiler_membar(); \ 238 return (tmp); \ 239} \ 240struct __hack 241 242#else /* !(_KERNEL && !SMP) */ 243 244#define ATOMIC_LOAD(TYPE, LOP) \ 245static __inline u_##TYPE \ 246atomic_load_acq_##TYPE(volatile u_##TYPE *p) \ 247{ \ 248 u_##TYPE res; \ 249 \ 250 __asm __volatile(MPLOCKED LOP \ 251 : "=a" (res), /* 0 */ \ 252 "+m" (*p) /* 1 */ \ 253 : : "memory", "cc"); \ 254 \ 255 return (res); \ 256} \ 257struct __hack 258 259#endif /* _KERNEL && !SMP */ 260 261#endif /* KLD_MODULE || !__GNUCLIKE_ASM */ 262 263ATOMIC_ASM(set, char, "orb %b1,%0", "iq", v); 264ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v); 265ATOMIC_ASM(add, char, "addb %b1,%0", "iq", v); 266ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v); 267 268ATOMIC_ASM(set, short, "orw %w1,%0", "ir", v); 269ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v); 270ATOMIC_ASM(add, short, "addw %w1,%0", "ir", v); 271ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v); 272 273ATOMIC_ASM(set, int, "orl %1,%0", "ir", v); 274ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v); 275ATOMIC_ASM(add, int, "addl %1,%0", "ir", v); 276ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v); 277 278ATOMIC_ASM(set, long, "orq %1,%0", "ir", v); 279ATOMIC_ASM(clear, long, "andq %1,%0", "ir", ~v); 280ATOMIC_ASM(add, long, "addq %1,%0", "ir", v); 281ATOMIC_ASM(subtract, long, "subq %1,%0", "ir", v); 282 283ATOMIC_LOAD(char, "cmpxchgb %b0,%1"); 284ATOMIC_LOAD(short, "cmpxchgw %w0,%1"); 285ATOMIC_LOAD(int, "cmpxchgl %0,%1"); 286ATOMIC_LOAD(long, "cmpxchgq %0,%1"); 287 288ATOMIC_STORE(char); 289ATOMIC_STORE(short); 290ATOMIC_STORE(int); 291ATOMIC_STORE(long); 292 293#undef ATOMIC_ASM 294#undef ATOMIC_LOAD 295#undef ATOMIC_STORE 296 297#ifndef WANT_FUNCTIONS 298 299/* Read the current value and store a zero in the destination. */ 300#ifdef __GNUCLIKE_ASM 301 302static __inline u_int 303atomic_readandclear_int(volatile u_int *p) 304{ 305 u_int res; 306 307 res = 0; 308 __asm __volatile( 309 " xchgl %1,%0 ; " 310 "# atomic_readandclear_int" 311 : "+r" (res), /* 0 */ 312 "+m" (*p)); /* 1 */ 313 314 return (res); 315} 316 317static __inline u_long 318atomic_readandclear_long(volatile u_long *p) 319{ 320 u_long res; 321 322 res = 0; 323 __asm __volatile( 324 " xchgq %1,%0 ; " 325 "# atomic_readandclear_long" 326 : "+r" (res), /* 0 */ 327 "+m" (*p)); /* 1 */ 328 329 return (res); 330} 331 332#else /* !__GNUCLIKE_ASM */ 333 334u_int atomic_readandclear_int(volatile u_int *p); 335u_long atomic_readandclear_long(volatile u_long *p); 336 337#endif /* __GNUCLIKE_ASM */ 338 339#define atomic_set_acq_char atomic_set_barr_char 340#define atomic_set_rel_char atomic_set_barr_char 341#define atomic_clear_acq_char atomic_clear_barr_char 342#define atomic_clear_rel_char atomic_clear_barr_char 343#define atomic_add_acq_char atomic_add_barr_char 344#define atomic_add_rel_char atomic_add_barr_char 345#define atomic_subtract_acq_char atomic_subtract_barr_char 346#define atomic_subtract_rel_char atomic_subtract_barr_char 347 348#define atomic_set_acq_short atomic_set_barr_short 349#define atomic_set_rel_short atomic_set_barr_short 350#define atomic_clear_acq_short atomic_clear_barr_short 351#define atomic_clear_rel_short atomic_clear_barr_short 352#define atomic_add_acq_short atomic_add_barr_short 353#define atomic_add_rel_short atomic_add_barr_short 354#define atomic_subtract_acq_short atomic_subtract_barr_short 355#define atomic_subtract_rel_short atomic_subtract_barr_short 356 357#define atomic_set_acq_int atomic_set_barr_int 358#define atomic_set_rel_int atomic_set_barr_int 359#define atomic_clear_acq_int atomic_clear_barr_int 360#define atomic_clear_rel_int atomic_clear_barr_int 361#define atomic_add_acq_int atomic_add_barr_int 362#define atomic_add_rel_int atomic_add_barr_int 363#define atomic_subtract_acq_int atomic_subtract_barr_int 364#define atomic_subtract_rel_int atomic_subtract_barr_int 365#define atomic_cmpset_acq_int atomic_cmpset_int 366#define atomic_cmpset_rel_int atomic_cmpset_int 367 368#define atomic_set_acq_long atomic_set_barr_long 369#define atomic_set_rel_long atomic_set_barr_long 370#define atomic_clear_acq_long atomic_clear_barr_long 371#define atomic_clear_rel_long atomic_clear_barr_long 372#define atomic_add_acq_long atomic_add_barr_long 373#define atomic_add_rel_long atomic_add_barr_long 374#define atomic_subtract_acq_long atomic_subtract_barr_long 375#define atomic_subtract_rel_long atomic_subtract_barr_long 376#define atomic_cmpset_acq_long atomic_cmpset_long 377#define atomic_cmpset_rel_long atomic_cmpset_long 378 379/* Operations on 8-bit bytes. */ 380#define atomic_set_8 atomic_set_char 381#define atomic_set_acq_8 atomic_set_acq_char 382#define atomic_set_rel_8 atomic_set_rel_char 383#define atomic_clear_8 atomic_clear_char 384#define atomic_clear_acq_8 atomic_clear_acq_char 385#define atomic_clear_rel_8 atomic_clear_rel_char 386#define atomic_add_8 atomic_add_char 387#define atomic_add_acq_8 atomic_add_acq_char 388#define atomic_add_rel_8 atomic_add_rel_char 389#define atomic_subtract_8 atomic_subtract_char 390#define atomic_subtract_acq_8 atomic_subtract_acq_char 391#define atomic_subtract_rel_8 atomic_subtract_rel_char 392#define atomic_load_acq_8 atomic_load_acq_char 393#define atomic_store_rel_8 atomic_store_rel_char 394 395/* Operations on 16-bit words. */ 396#define atomic_set_16 atomic_set_short 397#define atomic_set_acq_16 atomic_set_acq_short 398#define atomic_set_rel_16 atomic_set_rel_short 399#define atomic_clear_16 atomic_clear_short 400#define atomic_clear_acq_16 atomic_clear_acq_short 401#define atomic_clear_rel_16 atomic_clear_rel_short 402#define atomic_add_16 atomic_add_short 403#define atomic_add_acq_16 atomic_add_acq_short 404#define atomic_add_rel_16 atomic_add_rel_short 405#define atomic_subtract_16 atomic_subtract_short 406#define atomic_subtract_acq_16 atomic_subtract_acq_short 407#define atomic_subtract_rel_16 atomic_subtract_rel_short 408#define atomic_load_acq_16 atomic_load_acq_short 409#define atomic_store_rel_16 atomic_store_rel_short 410 411/* Operations on 32-bit double words. */ 412#define atomic_set_32 atomic_set_int 413#define atomic_set_acq_32 atomic_set_acq_int 414#define atomic_set_rel_32 atomic_set_rel_int 415#define atomic_clear_32 atomic_clear_int 416#define atomic_clear_acq_32 atomic_clear_acq_int 417#define atomic_clear_rel_32 atomic_clear_rel_int 418#define atomic_add_32 atomic_add_int 419#define atomic_add_acq_32 atomic_add_acq_int 420#define atomic_add_rel_32 atomic_add_rel_int 421#define atomic_subtract_32 atomic_subtract_int 422#define atomic_subtract_acq_32 atomic_subtract_acq_int 423#define atomic_subtract_rel_32 atomic_subtract_rel_int 424#define atomic_load_acq_32 atomic_load_acq_int 425#define atomic_store_rel_32 atomic_store_rel_int 426#define atomic_cmpset_32 atomic_cmpset_int 427#define atomic_cmpset_acq_32 atomic_cmpset_acq_int 428#define atomic_cmpset_rel_32 atomic_cmpset_rel_int 429#define atomic_readandclear_32 atomic_readandclear_int 430#define atomic_fetchadd_32 atomic_fetchadd_int 431 432/* Operations on 64-bit quad words. */ 433#define atomic_set_64 atomic_set_long 434#define atomic_set_acq_64 atomic_set_acq_long 435#define atomic_set_rel_64 atomic_set_rel_long 436#define atomic_clear_64 atomic_clear_long 437#define atomic_clear_acq_64 atomic_clear_acq_long 438#define atomic_clear_rel_64 atomic_clear_rel_long 439#define atomic_add_64 atomic_add_long 440#define atomic_add_acq_64 atomic_add_acq_long 441#define atomic_add_rel_64 atomic_add_rel_long 442#define atomic_subtract_64 atomic_subtract_long 443#define atomic_subtract_acq_64 atomic_subtract_acq_long 444#define atomic_subtract_rel_64 atomic_subtract_rel_long 445#define atomic_load_acq_64 atomic_load_acq_long 446#define atomic_store_rel_64 atomic_store_rel_long 447#define atomic_cmpset_64 atomic_cmpset_long 448#define atomic_cmpset_acq_64 atomic_cmpset_acq_long 449#define atomic_cmpset_rel_64 atomic_cmpset_rel_long 450#define atomic_readandclear_64 atomic_readandclear_long 451 452/* Operations on pointers. */ 453#define atomic_set_ptr atomic_set_long 454#define atomic_set_acq_ptr atomic_set_acq_long 455#define atomic_set_rel_ptr atomic_set_rel_long 456#define atomic_clear_ptr atomic_clear_long 457#define atomic_clear_acq_ptr atomic_clear_acq_long 458#define atomic_clear_rel_ptr atomic_clear_rel_long 459#define atomic_add_ptr atomic_add_long 460#define atomic_add_acq_ptr atomic_add_acq_long 461#define atomic_add_rel_ptr atomic_add_rel_long 462#define atomic_subtract_ptr atomic_subtract_long 463#define atomic_subtract_acq_ptr atomic_subtract_acq_long 464#define atomic_subtract_rel_ptr atomic_subtract_rel_long 465#define atomic_load_acq_ptr atomic_load_acq_long 466#define atomic_store_rel_ptr atomic_store_rel_long 467#define atomic_cmpset_ptr atomic_cmpset_long 468#define atomic_cmpset_acq_ptr atomic_cmpset_acq_long 469#define atomic_cmpset_rel_ptr atomic_cmpset_rel_long 470#define atomic_readandclear_ptr atomic_readandclear_long 471 472#endif /* !WANT_FUNCTIONS */ 473 474#endif /* !_MACHINE_ATOMIC_H_ */ 475