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