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