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) 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 76 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \ 77 u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p); \ 78 void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v) 79 80 #else /* !KLD_MODULE */ 81 82 #ifdef __GNUCLIKE_ASM 83 84 /* 85 * For userland, assume the SMP case and use lock prefixes so that 86 * the binaries will run on both types of 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 demark potential before-and-after side 96 * effects if an interrupt or SMP collision were to occur. 97 */ 98 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \ 99 static __inline void \ 100 atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ 101 { \ 102 __asm __volatile(__XSTRING(MPLOCKED) OP \ 103 : "+m" (*p) \ 104 : CONS (V)); \ 105 } \ 106 struct __hack 107 108 #else /* !__GNUCLIKE_ASM */ 109 110 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \ 111 extern void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v) 112 113 #endif /* __GNUCLIKE_ASM */ 114 115 /* 116 * Atomic compare and set, used by the mutex functions 117 * 118 * if (*dst == exp) *dst = src (all 32 bit words) 119 * 120 * Returns 0 on failure, non-zero on success 121 */ 122 123 #ifdef __GNUCLIKE_ASM 124 125 #if defined(CPU_DISABLE_CMPXCHG) 126 127 static __inline int 128 atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src) 129 { 130 int res = exp; 131 132 __asm __volatile( 133 " pushfl ; " 134 " cli ; " 135 " cmpl %0,%2 ; " 136 " jne 1f ; " 137 " movl %1,%2 ; " 138 "1: " 139 " sete %%al; " 140 " movzbl %%al,%0 ; " 141 " popfl ; " 142 "# atomic_cmpset_int" 143 : "+a" (res) /* 0 (result) */ 144 : "r" (src), /* 1 */ 145 "m" (*(dst)) /* 2 */ 146 : "memory"); 147 148 return (res); 149 } 150 151 #else /* defined(CPU_DISABLE_CMPXCHG) */ 152 153 static __inline int 154 atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src) 155 { 156 int res = exp; 157 158 __asm __volatile ( 159 " " __XSTRING(MPLOCKED) " " 160 " cmpxchgl %1,%2 ; " 161 " setz %%al ; " 162 " movzbl %%al,%0 ; " 163 "1: " 164 "# atomic_cmpset_int" 165 : "+a" (res) /* 0 (result) */ 166 : "r" (src), /* 1 */ 167 "m" (*(dst)) /* 2 */ 168 : "memory"); 169 170 return (res); 171 } 172 173 #endif /* defined(CPU_DISABLE_CMPXCHG) */ 174 175 #endif /* __GNUCLIKE_ASM */ 176 177 #ifdef __GNUCLIKE_ASM 178 179 #if defined(_KERNEL) && !defined(SMP) 180 181 /* 182 * We assume that a = b will do atomic loads and stores. However, on a 183 * PentiumPro or higher, reads may pass writes, so for that case we have 184 * to use a serializing instruction (i.e. with LOCK) to do the load in 185 * SMP kernels. For UP kernels, however, the cache of the single processor 186 * is always consistent, so we don't need any memory barriers. 187 */ 188 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \ 189 static __inline u_##TYPE \ 190 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \ 191 { \ 192 return (*p); \ 193 } \ 194 \ 195 static __inline void \ 196 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ 197 { \ 198 *p = v; \ 199 } \ 200 struct __hack 201 202 #else /* defined(SMP) */ 203 204 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \ 205 static __inline u_##TYPE \ 206 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \ 207 { \ 208 u_##TYPE res; \ 209 \ 210 __asm __volatile(__XSTRING(MPLOCKED) LOP \ 211 : "=a" (res), /* 0 (result) */\ 212 "+m" (*p) /* 1 */ \ 213 : : "memory"); \ 214 \ 215 return (res); \ 216 } \ 217 \ 218 /* \ 219 * The XCHG instruction asserts LOCK automagically. \ 220 */ \ 221 static __inline void \ 222 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ 223 { \ 224 __asm __volatile(SOP \ 225 : "+m" (*p), /* 0 */ \ 226 "+r" (v) /* 1 */ \ 227 : : "memory"); \ 228 } \ 229 struct __hack 230 231 #endif /* !defined(SMP) */ 232 233 #else /* !__GNUCLIKE_ASM */ 234 235 extern int atomic_cmpset_int(volatile u_int *, u_int, u_int); 236 237 #define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \ 238 extern u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p); \ 239 extern void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v) 240 241 #endif /* __GNUCLIKE_ASM */ 242 243 #endif /* KLD_MODULE */ 244 245 ATOMIC_ASM(set, char, "orb %b1,%0", "iq", v); 246 ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v); 247 ATOMIC_ASM(add, char, "addb %b1,%0", "iq", v); 248 ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v); 249 250 ATOMIC_ASM(set, short, "orw %w1,%0", "ir", v); 251 ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v); 252 ATOMIC_ASM(add, short, "addw %w1,%0", "ir", v); 253 ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v); 254 255 ATOMIC_ASM(set, int, "orl %1,%0", "ir", v); 256 ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v); 257 ATOMIC_ASM(add, int, "addl %1,%0", "ir", v); 258 ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v); 259 260 ATOMIC_ASM(set, long, "orl %1,%0", "ir", v); 261 ATOMIC_ASM(clear, long, "andl %1,%0", "ir", ~v); 262 ATOMIC_ASM(add, long, "addl %1,%0", "ir", v); 263 ATOMIC_ASM(subtract, long, "subl %1,%0", "ir", v); 264 265 ATOMIC_STORE_LOAD(char, "cmpxchgb %b0,%1", "xchgb %b1,%0"); 266 ATOMIC_STORE_LOAD(short,"cmpxchgw %w0,%1", "xchgw %w1,%0"); 267 ATOMIC_STORE_LOAD(int, "cmpxchgl %0,%1", "xchgl %1,%0"); 268 ATOMIC_STORE_LOAD(long, "cmpxchgl %0,%1", "xchgl %1,%0"); 269 270 #undef ATOMIC_ASM 271 #undef ATOMIC_STORE_LOAD 272 273 #define atomic_set_acq_char atomic_set_char 274 #define atomic_set_rel_char atomic_set_char 275 #define atomic_clear_acq_char atomic_clear_char 276 #define atomic_clear_rel_char atomic_clear_char 277 #define atomic_add_acq_char atomic_add_char 278 #define atomic_add_rel_char atomic_add_char 279 #define atomic_subtract_acq_char atomic_subtract_char 280 #define atomic_subtract_rel_char atomic_subtract_char 281 282 #define atomic_set_acq_short atomic_set_short 283 #define atomic_set_rel_short atomic_set_short 284 #define atomic_clear_acq_short atomic_clear_short 285 #define atomic_clear_rel_short atomic_clear_short 286 #define atomic_add_acq_short atomic_add_short 287 #define atomic_add_rel_short atomic_add_short 288 #define atomic_subtract_acq_short atomic_subtract_short 289 #define atomic_subtract_rel_short atomic_subtract_short 290 291 #define atomic_set_acq_int atomic_set_int 292 #define atomic_set_rel_int atomic_set_int 293 #define atomic_clear_acq_int atomic_clear_int 294 #define atomic_clear_rel_int atomic_clear_int 295 #define atomic_add_acq_int atomic_add_int 296 #define atomic_add_rel_int atomic_add_int 297 #define atomic_subtract_acq_int atomic_subtract_int 298 #define atomic_subtract_rel_int atomic_subtract_int 299 #define atomic_cmpset_acq_int atomic_cmpset_int 300 #define atomic_cmpset_rel_int atomic_cmpset_int 301 302 #define atomic_set_acq_long atomic_set_long 303 #define atomic_set_rel_long atomic_set_long 304 #define atomic_clear_acq_long atomic_clear_long 305 #define atomic_clear_rel_long atomic_clear_long 306 #define atomic_add_acq_long atomic_add_long 307 #define atomic_add_rel_long atomic_add_long 308 #define atomic_subtract_acq_long atomic_subtract_long 309 #define atomic_subtract_rel_long atomic_subtract_long 310 #define atomic_cmpset_long atomic_cmpset_int 311 #define atomic_cmpset_acq_long atomic_cmpset_acq_int 312 #define atomic_cmpset_rel_long atomic_cmpset_rel_int 313 314 #define atomic_cmpset_acq_ptr atomic_cmpset_ptr 315 #define atomic_cmpset_rel_ptr atomic_cmpset_ptr 316 317 #define atomic_set_8 atomic_set_char 318 #define atomic_set_acq_8 atomic_set_acq_char 319 #define atomic_set_rel_8 atomic_set_rel_char 320 #define atomic_clear_8 atomic_clear_char 321 #define atomic_clear_acq_8 atomic_clear_acq_char 322 #define atomic_clear_rel_8 atomic_clear_rel_char 323 #define atomic_add_8 atomic_add_char 324 #define atomic_add_acq_8 atomic_add_acq_char 325 #define atomic_add_rel_8 atomic_add_rel_char 326 #define atomic_subtract_8 atomic_subtract_char 327 #define atomic_subtract_acq_8 atomic_subtract_acq_char 328 #define atomic_subtract_rel_8 atomic_subtract_rel_char 329 #define atomic_load_acq_8 atomic_load_acq_char 330 #define atomic_store_rel_8 atomic_store_rel_char 331 332 #define atomic_set_16 atomic_set_short 333 #define atomic_set_acq_16 atomic_set_acq_short 334 #define atomic_set_rel_16 atomic_set_rel_short 335 #define atomic_clear_16 atomic_clear_short 336 #define atomic_clear_acq_16 atomic_clear_acq_short 337 #define atomic_clear_rel_16 atomic_clear_rel_short 338 #define atomic_add_16 atomic_add_short 339 #define atomic_add_acq_16 atomic_add_acq_short 340 #define atomic_add_rel_16 atomic_add_rel_short 341 #define atomic_subtract_16 atomic_subtract_short 342 #define atomic_subtract_acq_16 atomic_subtract_acq_short 343 #define atomic_subtract_rel_16 atomic_subtract_rel_short 344 #define atomic_load_acq_16 atomic_load_acq_short 345 #define atomic_store_rel_16 atomic_store_rel_short 346 347 #define atomic_set_32 atomic_set_int 348 #define atomic_set_acq_32 atomic_set_acq_int 349 #define atomic_set_rel_32 atomic_set_rel_int 350 #define atomic_clear_32 atomic_clear_int 351 #define atomic_clear_acq_32 atomic_clear_acq_int 352 #define atomic_clear_rel_32 atomic_clear_rel_int 353 #define atomic_add_32 atomic_add_int 354 #define atomic_add_acq_32 atomic_add_acq_int 355 #define atomic_add_rel_32 atomic_add_rel_int 356 #define atomic_subtract_32 atomic_subtract_int 357 #define atomic_subtract_acq_32 atomic_subtract_acq_int 358 #define atomic_subtract_rel_32 atomic_subtract_rel_int 359 #define atomic_load_acq_32 atomic_load_acq_int 360 #define atomic_store_rel_32 atomic_store_rel_int 361 #define atomic_cmpset_32 atomic_cmpset_int 362 #define atomic_cmpset_acq_32 atomic_cmpset_acq_int 363 #define atomic_cmpset_rel_32 atomic_cmpset_rel_int 364 #define atomic_readandclear_32 atomic_readandclear_int 365 366 #if !defined(WANT_FUNCTIONS) 367 static __inline int 368 atomic_cmpset_ptr(volatile void *dst, void *exp, void *src) 369 { 370 371 return (atomic_cmpset_int((volatile u_int *)dst, (u_int)exp, 372 (u_int)src)); 373 } 374 375 static __inline void * 376 atomic_load_acq_ptr(volatile void *p) 377 { 378 /* 379 * The apparently-bogus cast to intptr_t in the following is to 380 * avoid a warning from "gcc -Wbad-function-cast". 381 */ 382 return ((void *)(intptr_t)atomic_load_acq_int((volatile u_int *)p)); 383 } 384 385 static __inline void 386 atomic_store_rel_ptr(volatile void *p, void *v) 387 { 388 atomic_store_rel_int((volatile u_int *)p, (u_int)v); 389 } 390 391 #define ATOMIC_PTR(NAME) \ 392 static __inline void \ 393 atomic_##NAME##_ptr(volatile void *p, uintptr_t v) \ 394 { \ 395 atomic_##NAME##_int((volatile u_int *)p, v); \ 396 } \ 397 \ 398 static __inline void \ 399 atomic_##NAME##_acq_ptr(volatile void *p, uintptr_t v) \ 400 { \ 401 atomic_##NAME##_acq_int((volatile u_int *)p, v);\ 402 } \ 403 \ 404 static __inline void \ 405 atomic_##NAME##_rel_ptr(volatile void *p, uintptr_t v) \ 406 { \ 407 atomic_##NAME##_rel_int((volatile u_int *)p, v);\ 408 } 409 410 ATOMIC_PTR(set) 411 ATOMIC_PTR(clear) 412 ATOMIC_PTR(add) 413 ATOMIC_PTR(subtract) 414 415 #undef ATOMIC_PTR 416 417 #ifdef __GNUCLIKE_ASM 418 419 static __inline u_int 420 atomic_readandclear_int(volatile u_int *addr) 421 { 422 u_int result; 423 424 __asm __volatile ( 425 " xorl %0,%0 ; " 426 " xchgl %1,%0 ; " 427 "# atomic_readandclear_int" 428 : "=&r" (result) /* 0 (result) */ 429 : "m" (*addr)); /* 1 (addr) */ 430 431 return (result); 432 } 433 434 static __inline u_long 435 atomic_readandclear_long(volatile u_long *addr) 436 { 437 u_long result; 438 439 __asm __volatile ( 440 " xorl %0,%0 ; " 441 " xchgl %1,%0 ; " 442 "# atomic_readandclear_int" 443 : "=&r" (result) /* 0 (result) */ 444 : "m" (*addr)); /* 1 (addr) */ 445 446 return (result); 447 } 448 449 #else /* !__GNUCLIKE_ASM */ 450 451 extern u_long atomic_readandclear_long(volatile u_long *); 452 extern u_int atomic_readandclear_int(volatile u_int *); 453 454 #endif /* __GNUCLIKE_ASM */ 455 456 #endif /* !defined(WANT_FUNCTIONS) */ 457 #endif /* ! _MACHINE_ATOMIC_H_ */ 458