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