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