1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_PERCPU_H 3 #define _ASM_X86_PERCPU_H 4 5 #ifdef CONFIG_X86_64 6 #define __percpu_seg gs 7 #define __percpu_rel (%rip) 8 #else 9 #define __percpu_seg fs 10 #define __percpu_rel 11 #endif 12 13 #ifdef __ASSEMBLY__ 14 15 #ifdef CONFIG_SMP 16 #define __percpu %__percpu_seg: 17 #else 18 #define __percpu 19 #endif 20 21 #define PER_CPU_VAR(var) __percpu(var)__percpu_rel 22 23 #ifdef CONFIG_X86_64_SMP 24 #define INIT_PER_CPU_VAR(var) init_per_cpu__##var 25 #else 26 #define INIT_PER_CPU_VAR(var) var 27 #endif 28 29 #else /* ...!ASSEMBLY */ 30 31 #include <linux/build_bug.h> 32 #include <linux/stringify.h> 33 #include <asm/asm.h> 34 35 #ifdef CONFIG_SMP 36 37 #ifdef CONFIG_CC_HAS_NAMED_AS 38 39 #ifdef __CHECKER__ 40 #define __seg_gs __attribute__((address_space(__seg_gs))) 41 #define __seg_fs __attribute__((address_space(__seg_fs))) 42 #endif 43 44 #ifdef CONFIG_X86_64 45 #define __percpu_seg_override __seg_gs 46 #else 47 #define __percpu_seg_override __seg_fs 48 #endif 49 50 #define __percpu_prefix "" 51 52 #else /* CONFIG_CC_HAS_NAMED_AS */ 53 54 #define __percpu_seg_override 55 #define __percpu_prefix "%%"__stringify(__percpu_seg)":" 56 57 #endif /* CONFIG_CC_HAS_NAMED_AS */ 58 59 #define __force_percpu_prefix "%%"__stringify(__percpu_seg)":" 60 #define __my_cpu_offset this_cpu_read(this_cpu_off) 61 62 #ifdef CONFIG_USE_X86_SEG_SUPPORT 63 /* 64 * Efficient implementation for cases in which the compiler supports 65 * named address spaces. Allows the compiler to perform additional 66 * optimizations that can save more instructions. 67 */ 68 #define arch_raw_cpu_ptr(ptr) \ 69 ({ \ 70 unsigned long tcp_ptr__; \ 71 tcp_ptr__ = __raw_cpu_read(, this_cpu_off); \ 72 \ 73 tcp_ptr__ += (unsigned long)(ptr); \ 74 (typeof(*(ptr)) __kernel __force *)tcp_ptr__; \ 75 }) 76 #else /* CONFIG_USE_X86_SEG_SUPPORT */ 77 /* 78 * Compared to the generic __my_cpu_offset version, the following 79 * saves one instruction and avoids clobbering a temp register. 80 */ 81 #define arch_raw_cpu_ptr(ptr) \ 82 ({ \ 83 unsigned long tcp_ptr__; \ 84 asm ("mov " __percpu_arg(1) ", %0" \ 85 : "=r" (tcp_ptr__) \ 86 : "m" (__my_cpu_var(this_cpu_off))); \ 87 \ 88 tcp_ptr__ += (unsigned long)(ptr); \ 89 (typeof(*(ptr)) __kernel __force *)tcp_ptr__; \ 90 }) 91 #endif /* CONFIG_USE_X86_SEG_SUPPORT */ 92 93 #define PER_CPU_VAR(var) %__percpu_seg:(var)__percpu_rel 94 95 #else /* CONFIG_SMP */ 96 #define __percpu_seg_override 97 #define __percpu_prefix "" 98 #define __force_percpu_prefix "" 99 100 #define PER_CPU_VAR(var) (var)__percpu_rel 101 102 #endif /* CONFIG_SMP */ 103 104 #define __my_cpu_type(var) typeof(var) __percpu_seg_override 105 #define __my_cpu_ptr(ptr) (__my_cpu_type(*ptr) *)(uintptr_t)(ptr) 106 #define __my_cpu_var(var) (*__my_cpu_ptr(&var)) 107 #define __percpu_arg(x) __percpu_prefix "%" #x 108 #define __force_percpu_arg(x) __force_percpu_prefix "%" #x 109 110 /* 111 * Initialized pointers to per-cpu variables needed for the boot 112 * processor need to use these macros to get the proper address 113 * offset from __per_cpu_load on SMP. 114 * 115 * There also must be an entry in vmlinux_64.lds.S 116 */ 117 #define DECLARE_INIT_PER_CPU(var) \ 118 extern typeof(var) init_per_cpu_var(var) 119 120 #ifdef CONFIG_X86_64_SMP 121 #define init_per_cpu_var(var) init_per_cpu__##var 122 #else 123 #define init_per_cpu_var(var) var 124 #endif 125 126 /* For arch-specific code, we can use direct single-insn ops (they 127 * don't give an lvalue though). */ 128 129 #define __pcpu_type_1 u8 130 #define __pcpu_type_2 u16 131 #define __pcpu_type_4 u32 132 #define __pcpu_type_8 u64 133 134 #define __pcpu_cast_1(val) ((u8)(((unsigned long) val) & 0xff)) 135 #define __pcpu_cast_2(val) ((u16)(((unsigned long) val) & 0xffff)) 136 #define __pcpu_cast_4(val) ((u32)(((unsigned long) val) & 0xffffffff)) 137 #define __pcpu_cast_8(val) ((u64)(val)) 138 139 #define __pcpu_op1_1(op, dst) op "b " dst 140 #define __pcpu_op1_2(op, dst) op "w " dst 141 #define __pcpu_op1_4(op, dst) op "l " dst 142 #define __pcpu_op1_8(op, dst) op "q " dst 143 144 #define __pcpu_op2_1(op, src, dst) op "b " src ", " dst 145 #define __pcpu_op2_2(op, src, dst) op "w " src ", " dst 146 #define __pcpu_op2_4(op, src, dst) op "l " src ", " dst 147 #define __pcpu_op2_8(op, src, dst) op "q " src ", " dst 148 149 #define __pcpu_reg_1(mod, x) mod "q" (x) 150 #define __pcpu_reg_2(mod, x) mod "r" (x) 151 #define __pcpu_reg_4(mod, x) mod "r" (x) 152 #define __pcpu_reg_8(mod, x) mod "r" (x) 153 154 #define __pcpu_reg_imm_1(x) "qi" (x) 155 #define __pcpu_reg_imm_2(x) "ri" (x) 156 #define __pcpu_reg_imm_4(x) "ri" (x) 157 #define __pcpu_reg_imm_8(x) "re" (x) 158 159 #define percpu_to_op(size, qual, op, _var, _val) \ 160 do { \ 161 __pcpu_type_##size pto_val__ = __pcpu_cast_##size(_val); \ 162 if (0) { \ 163 typeof(_var) pto_tmp__; \ 164 pto_tmp__ = (_val); \ 165 (void)pto_tmp__; \ 166 } \ 167 asm qual(__pcpu_op2_##size(op, "%[val]", __percpu_arg([var])) \ 168 : [var] "+m" (__my_cpu_var(_var)) \ 169 : [val] __pcpu_reg_imm_##size(pto_val__)); \ 170 } while (0) 171 172 #define percpu_unary_op(size, qual, op, _var) \ 173 ({ \ 174 asm qual (__pcpu_op1_##size(op, __percpu_arg([var])) \ 175 : [var] "+m" (__my_cpu_var(_var))); \ 176 }) 177 178 /* 179 * Generate a percpu add to memory instruction and optimize code 180 * if one is added or subtracted. 181 */ 182 #define percpu_add_op(size, qual, var, val) \ 183 do { \ 184 const int pao_ID__ = (__builtin_constant_p(val) && \ 185 ((val) == 1 || (val) == -1)) ? \ 186 (int)(val) : 0; \ 187 if (0) { \ 188 typeof(var) pao_tmp__; \ 189 pao_tmp__ = (val); \ 190 (void)pao_tmp__; \ 191 } \ 192 if (pao_ID__ == 1) \ 193 percpu_unary_op(size, qual, "inc", var); \ 194 else if (pao_ID__ == -1) \ 195 percpu_unary_op(size, qual, "dec", var); \ 196 else \ 197 percpu_to_op(size, qual, "add", var, val); \ 198 } while (0) 199 200 #define percpu_from_op(size, qual, op, _var) \ 201 ({ \ 202 __pcpu_type_##size pfo_val__; \ 203 asm qual (__pcpu_op2_##size(op, __percpu_arg([var]), "%[val]") \ 204 : [val] __pcpu_reg_##size("=", pfo_val__) \ 205 : [var] "m" (__my_cpu_var(_var))); \ 206 (typeof(_var))(unsigned long) pfo_val__; \ 207 }) 208 209 #define percpu_stable_op(size, op, _var) \ 210 ({ \ 211 __pcpu_type_##size pfo_val__; \ 212 asm(__pcpu_op2_##size(op, __force_percpu_arg(a[var]), "%[val]") \ 213 : [val] __pcpu_reg_##size("=", pfo_val__) \ 214 : [var] "i" (&(_var))); \ 215 (typeof(_var))(unsigned long) pfo_val__; \ 216 }) 217 218 /* 219 * Add return operation 220 */ 221 #define percpu_add_return_op(size, qual, _var, _val) \ 222 ({ \ 223 __pcpu_type_##size paro_tmp__ = __pcpu_cast_##size(_val); \ 224 asm qual (__pcpu_op2_##size("xadd", "%[tmp]", \ 225 __percpu_arg([var])) \ 226 : [tmp] __pcpu_reg_##size("+", paro_tmp__), \ 227 [var] "+m" (__my_cpu_var(_var)) \ 228 : : "memory"); \ 229 (typeof(_var))(unsigned long) (paro_tmp__ + _val); \ 230 }) 231 232 /* 233 * xchg is implemented using cmpxchg without a lock prefix. xchg is 234 * expensive due to the implied lock prefix. The processor cannot prefetch 235 * cachelines if xchg is used. 236 */ 237 #define percpu_xchg_op(size, qual, _var, _nval) \ 238 ({ \ 239 __pcpu_type_##size pxo_old__; \ 240 __pcpu_type_##size pxo_new__ = __pcpu_cast_##size(_nval); \ 241 asm qual (__pcpu_op2_##size("mov", __percpu_arg([var]), \ 242 "%[oval]") \ 243 "\n1:\t" \ 244 __pcpu_op2_##size("cmpxchg", "%[nval]", \ 245 __percpu_arg([var])) \ 246 "\n\tjnz 1b" \ 247 : [oval] "=&a" (pxo_old__), \ 248 [var] "+m" (__my_cpu_var(_var)) \ 249 : [nval] __pcpu_reg_##size(, pxo_new__) \ 250 : "memory"); \ 251 (typeof(_var))(unsigned long) pxo_old__; \ 252 }) 253 254 /* 255 * cmpxchg has no such implied lock semantics as a result it is much 256 * more efficient for cpu local operations. 257 */ 258 #define percpu_cmpxchg_op(size, qual, _var, _oval, _nval) \ 259 ({ \ 260 __pcpu_type_##size pco_old__ = __pcpu_cast_##size(_oval); \ 261 __pcpu_type_##size pco_new__ = __pcpu_cast_##size(_nval); \ 262 asm qual (__pcpu_op2_##size("cmpxchg", "%[nval]", \ 263 __percpu_arg([var])) \ 264 : [oval] "+a" (pco_old__), \ 265 [var] "+m" (__my_cpu_var(_var)) \ 266 : [nval] __pcpu_reg_##size(, pco_new__) \ 267 : "memory"); \ 268 (typeof(_var))(unsigned long) pco_old__; \ 269 }) 270 271 #define percpu_try_cmpxchg_op(size, qual, _var, _ovalp, _nval) \ 272 ({ \ 273 bool success; \ 274 __pcpu_type_##size *pco_oval__ = (__pcpu_type_##size *)(_ovalp); \ 275 __pcpu_type_##size pco_old__ = *pco_oval__; \ 276 __pcpu_type_##size pco_new__ = __pcpu_cast_##size(_nval); \ 277 asm qual (__pcpu_op2_##size("cmpxchg", "%[nval]", \ 278 __percpu_arg([var])) \ 279 CC_SET(z) \ 280 : CC_OUT(z) (success), \ 281 [oval] "+a" (pco_old__), \ 282 [var] "+m" (__my_cpu_var(_var)) \ 283 : [nval] __pcpu_reg_##size(, pco_new__) \ 284 : "memory"); \ 285 if (unlikely(!success)) \ 286 *pco_oval__ = pco_old__; \ 287 likely(success); \ 288 }) 289 290 #if defined(CONFIG_X86_32) && !defined(CONFIG_UML) 291 #define percpu_cmpxchg64_op(size, qual, _var, _oval, _nval) \ 292 ({ \ 293 union { \ 294 u64 var; \ 295 struct { \ 296 u32 low, high; \ 297 }; \ 298 } old__, new__; \ 299 \ 300 old__.var = _oval; \ 301 new__.var = _nval; \ 302 \ 303 asm qual (ALTERNATIVE("call this_cpu_cmpxchg8b_emu", \ 304 "cmpxchg8b " __percpu_arg([var]), X86_FEATURE_CX8) \ 305 : [var] "+m" (__my_cpu_var(_var)), \ 306 "+a" (old__.low), \ 307 "+d" (old__.high) \ 308 : "b" (new__.low), \ 309 "c" (new__.high), \ 310 "S" (&(_var)) \ 311 : "memory"); \ 312 \ 313 old__.var; \ 314 }) 315 316 #define raw_cpu_cmpxchg64(pcp, oval, nval) percpu_cmpxchg64_op(8, , pcp, oval, nval) 317 #define this_cpu_cmpxchg64(pcp, oval, nval) percpu_cmpxchg64_op(8, volatile, pcp, oval, nval) 318 319 #define percpu_try_cmpxchg64_op(size, qual, _var, _ovalp, _nval) \ 320 ({ \ 321 bool success; \ 322 u64 *_oval = (u64 *)(_ovalp); \ 323 union { \ 324 u64 var; \ 325 struct { \ 326 u32 low, high; \ 327 }; \ 328 } old__, new__; \ 329 \ 330 old__.var = *_oval; \ 331 new__.var = _nval; \ 332 \ 333 asm qual (ALTERNATIVE("call this_cpu_cmpxchg8b_emu", \ 334 "cmpxchg8b " __percpu_arg([var]), X86_FEATURE_CX8) \ 335 CC_SET(z) \ 336 : CC_OUT(z) (success), \ 337 [var] "+m" (__my_cpu_var(_var)), \ 338 "+a" (old__.low), \ 339 "+d" (old__.high) \ 340 : "b" (new__.low), \ 341 "c" (new__.high), \ 342 "S" (&(_var)) \ 343 : "memory"); \ 344 if (unlikely(!success)) \ 345 *_oval = old__.var; \ 346 likely(success); \ 347 }) 348 349 #define raw_cpu_try_cmpxchg64(pcp, ovalp, nval) percpu_try_cmpxchg64_op(8, , pcp, ovalp, nval) 350 #define this_cpu_try_cmpxchg64(pcp, ovalp, nval) percpu_try_cmpxchg64_op(8, volatile, pcp, ovalp, nval) 351 #endif 352 353 #ifdef CONFIG_X86_64 354 #define raw_cpu_cmpxchg64(pcp, oval, nval) percpu_cmpxchg_op(8, , pcp, oval, nval); 355 #define this_cpu_cmpxchg64(pcp, oval, nval) percpu_cmpxchg_op(8, volatile, pcp, oval, nval); 356 357 #define raw_cpu_try_cmpxchg64(pcp, ovalp, nval) percpu_try_cmpxchg_op(8, , pcp, ovalp, nval); 358 #define this_cpu_try_cmpxchg64(pcp, ovalp, nval) percpu_try_cmpxchg_op(8, volatile, pcp, ovalp, nval); 359 360 #define percpu_cmpxchg128_op(size, qual, _var, _oval, _nval) \ 361 ({ \ 362 union { \ 363 u128 var; \ 364 struct { \ 365 u64 low, high; \ 366 }; \ 367 } old__, new__; \ 368 \ 369 old__.var = _oval; \ 370 new__.var = _nval; \ 371 \ 372 asm qual (ALTERNATIVE("call this_cpu_cmpxchg16b_emu", \ 373 "cmpxchg16b " __percpu_arg([var]), X86_FEATURE_CX16) \ 374 : [var] "+m" (__my_cpu_var(_var)), \ 375 "+a" (old__.low), \ 376 "+d" (old__.high) \ 377 : "b" (new__.low), \ 378 "c" (new__.high), \ 379 "S" (&(_var)) \ 380 : "memory"); \ 381 \ 382 old__.var; \ 383 }) 384 385 #define raw_cpu_cmpxchg128(pcp, oval, nval) percpu_cmpxchg128_op(16, , pcp, oval, nval) 386 #define this_cpu_cmpxchg128(pcp, oval, nval) percpu_cmpxchg128_op(16, volatile, pcp, oval, nval) 387 388 #define percpu_try_cmpxchg128_op(size, qual, _var, _ovalp, _nval) \ 389 ({ \ 390 bool success; \ 391 u128 *_oval = (u128 *)(_ovalp); \ 392 union { \ 393 u128 var; \ 394 struct { \ 395 u64 low, high; \ 396 }; \ 397 } old__, new__; \ 398 \ 399 old__.var = *_oval; \ 400 new__.var = _nval; \ 401 \ 402 asm qual (ALTERNATIVE("call this_cpu_cmpxchg16b_emu", \ 403 "cmpxchg16b " __percpu_arg([var]), X86_FEATURE_CX16) \ 404 CC_SET(z) \ 405 : CC_OUT(z) (success), \ 406 [var] "+m" (__my_cpu_var(_var)), \ 407 "+a" (old__.low), \ 408 "+d" (old__.high) \ 409 : "b" (new__.low), \ 410 "c" (new__.high), \ 411 "S" (&(_var)) \ 412 : "memory"); \ 413 if (unlikely(!success)) \ 414 *_oval = old__.var; \ 415 likely(success); \ 416 }) 417 418 #define raw_cpu_try_cmpxchg128(pcp, ovalp, nval) percpu_try_cmpxchg128_op(16, , pcp, ovalp, nval) 419 #define this_cpu_try_cmpxchg128(pcp, ovalp, nval) percpu_try_cmpxchg128_op(16, volatile, pcp, ovalp, nval) 420 #endif 421 422 /* 423 * this_cpu_read() makes gcc load the percpu variable every time it is 424 * accessed while this_cpu_read_stable() allows the value to be cached. 425 * this_cpu_read_stable() is more efficient and can be used if its value 426 * is guaranteed to be valid across cpus. The current users include 427 * pcpu_hot.current_task and pcpu_hot.top_of_stack, both of which are 428 * actually per-thread variables implemented as per-CPU variables and 429 * thus stable for the duration of the respective task. 430 */ 431 #define this_cpu_read_stable_1(pcp) percpu_stable_op(1, "mov", pcp) 432 #define this_cpu_read_stable_2(pcp) percpu_stable_op(2, "mov", pcp) 433 #define this_cpu_read_stable_4(pcp) percpu_stable_op(4, "mov", pcp) 434 #define this_cpu_read_stable_8(pcp) percpu_stable_op(8, "mov", pcp) 435 #define this_cpu_read_stable(pcp) __pcpu_size_call_return(this_cpu_read_stable_, pcp) 436 437 #ifdef CONFIG_USE_X86_SEG_SUPPORT 438 439 #define __raw_cpu_read(qual, pcp) \ 440 ({ \ 441 *(qual __my_cpu_type(pcp) *)__my_cpu_ptr(&(pcp)); \ 442 }) 443 444 #define __raw_cpu_write(qual, pcp, val) \ 445 do { \ 446 *(qual __my_cpu_type(pcp) *)__my_cpu_ptr(&(pcp)) = (val); \ 447 } while (0) 448 449 #define raw_cpu_read_1(pcp) __raw_cpu_read(, pcp) 450 #define raw_cpu_read_2(pcp) __raw_cpu_read(, pcp) 451 #define raw_cpu_read_4(pcp) __raw_cpu_read(, pcp) 452 #define raw_cpu_write_1(pcp, val) __raw_cpu_write(, pcp, val) 453 #define raw_cpu_write_2(pcp, val) __raw_cpu_write(, pcp, val) 454 #define raw_cpu_write_4(pcp, val) __raw_cpu_write(, pcp, val) 455 456 #define this_cpu_read_1(pcp) __raw_cpu_read(volatile, pcp) 457 #define this_cpu_read_2(pcp) __raw_cpu_read(volatile, pcp) 458 #define this_cpu_read_4(pcp) __raw_cpu_read(volatile, pcp) 459 #define this_cpu_write_1(pcp, val) __raw_cpu_write(volatile, pcp, val) 460 #define this_cpu_write_2(pcp, val) __raw_cpu_write(volatile, pcp, val) 461 #define this_cpu_write_4(pcp, val) __raw_cpu_write(volatile, pcp, val) 462 463 #ifdef CONFIG_X86_64 464 #define raw_cpu_read_8(pcp) __raw_cpu_read(, pcp) 465 #define raw_cpu_write_8(pcp, val) __raw_cpu_write(, pcp, val) 466 467 #define this_cpu_read_8(pcp) __raw_cpu_read(volatile, pcp) 468 #define this_cpu_write_8(pcp, val) __raw_cpu_write(volatile, pcp, val) 469 #endif 470 471 #define this_cpu_read_const(pcp) __raw_cpu_read(, pcp) 472 #else /* CONFIG_USE_X86_SEG_SUPPORT */ 473 474 #define raw_cpu_read_1(pcp) percpu_from_op(1, , "mov", pcp) 475 #define raw_cpu_read_2(pcp) percpu_from_op(2, , "mov", pcp) 476 #define raw_cpu_read_4(pcp) percpu_from_op(4, , "mov", pcp) 477 #define raw_cpu_write_1(pcp, val) percpu_to_op(1, , "mov", (pcp), val) 478 #define raw_cpu_write_2(pcp, val) percpu_to_op(2, , "mov", (pcp), val) 479 #define raw_cpu_write_4(pcp, val) percpu_to_op(4, , "mov", (pcp), val) 480 481 #define this_cpu_read_1(pcp) percpu_from_op(1, volatile, "mov", pcp) 482 #define this_cpu_read_2(pcp) percpu_from_op(2, volatile, "mov", pcp) 483 #define this_cpu_read_4(pcp) percpu_from_op(4, volatile, "mov", pcp) 484 #define this_cpu_write_1(pcp, val) percpu_to_op(1, volatile, "mov", (pcp), val) 485 #define this_cpu_write_2(pcp, val) percpu_to_op(2, volatile, "mov", (pcp), val) 486 #define this_cpu_write_4(pcp, val) percpu_to_op(4, volatile, "mov", (pcp), val) 487 488 #ifdef CONFIG_X86_64 489 #define raw_cpu_read_8(pcp) percpu_from_op(8, , "mov", pcp) 490 #define raw_cpu_write_8(pcp, val) percpu_to_op(8, , "mov", (pcp), val) 491 492 #define this_cpu_read_8(pcp) percpu_from_op(8, volatile, "mov", pcp) 493 #define this_cpu_write_8(pcp, val) percpu_to_op(8, volatile, "mov", (pcp), val) 494 #endif 495 496 /* 497 * The generic per-cpu infrastrucutre is not suitable for 498 * reading const-qualified variables. 499 */ 500 #define this_cpu_read_const(pcp) ({ BUILD_BUG(); (typeof(pcp))0; }) 501 #endif /* CONFIG_USE_X86_SEG_SUPPORT */ 502 503 #define raw_cpu_add_1(pcp, val) percpu_add_op(1, , (pcp), val) 504 #define raw_cpu_add_2(pcp, val) percpu_add_op(2, , (pcp), val) 505 #define raw_cpu_add_4(pcp, val) percpu_add_op(4, , (pcp), val) 506 #define raw_cpu_and_1(pcp, val) percpu_to_op(1, , "and", (pcp), val) 507 #define raw_cpu_and_2(pcp, val) percpu_to_op(2, , "and", (pcp), val) 508 #define raw_cpu_and_4(pcp, val) percpu_to_op(4, , "and", (pcp), val) 509 #define raw_cpu_or_1(pcp, val) percpu_to_op(1, , "or", (pcp), val) 510 #define raw_cpu_or_2(pcp, val) percpu_to_op(2, , "or", (pcp), val) 511 #define raw_cpu_or_4(pcp, val) percpu_to_op(4, , "or", (pcp), val) 512 513 /* 514 * raw_cpu_xchg() can use a load-store since it is not required to be 515 * IRQ-safe. 516 */ 517 #define raw_percpu_xchg_op(var, nval) \ 518 ({ \ 519 typeof(var) pxo_ret__ = raw_cpu_read(var); \ 520 raw_cpu_write(var, (nval)); \ 521 pxo_ret__; \ 522 }) 523 524 #define raw_cpu_xchg_1(pcp, val) raw_percpu_xchg_op(pcp, val) 525 #define raw_cpu_xchg_2(pcp, val) raw_percpu_xchg_op(pcp, val) 526 #define raw_cpu_xchg_4(pcp, val) raw_percpu_xchg_op(pcp, val) 527 528 #define this_cpu_add_1(pcp, val) percpu_add_op(1, volatile, (pcp), val) 529 #define this_cpu_add_2(pcp, val) percpu_add_op(2, volatile, (pcp), val) 530 #define this_cpu_add_4(pcp, val) percpu_add_op(4, volatile, (pcp), val) 531 #define this_cpu_and_1(pcp, val) percpu_to_op(1, volatile, "and", (pcp), val) 532 #define this_cpu_and_2(pcp, val) percpu_to_op(2, volatile, "and", (pcp), val) 533 #define this_cpu_and_4(pcp, val) percpu_to_op(4, volatile, "and", (pcp), val) 534 #define this_cpu_or_1(pcp, val) percpu_to_op(1, volatile, "or", (pcp), val) 535 #define this_cpu_or_2(pcp, val) percpu_to_op(2, volatile, "or", (pcp), val) 536 #define this_cpu_or_4(pcp, val) percpu_to_op(4, volatile, "or", (pcp), val) 537 #define this_cpu_xchg_1(pcp, nval) percpu_xchg_op(1, volatile, pcp, nval) 538 #define this_cpu_xchg_2(pcp, nval) percpu_xchg_op(2, volatile, pcp, nval) 539 #define this_cpu_xchg_4(pcp, nval) percpu_xchg_op(4, volatile, pcp, nval) 540 541 #define raw_cpu_add_return_1(pcp, val) percpu_add_return_op(1, , pcp, val) 542 #define raw_cpu_add_return_2(pcp, val) percpu_add_return_op(2, , pcp, val) 543 #define raw_cpu_add_return_4(pcp, val) percpu_add_return_op(4, , pcp, val) 544 #define raw_cpu_cmpxchg_1(pcp, oval, nval) percpu_cmpxchg_op(1, , pcp, oval, nval) 545 #define raw_cpu_cmpxchg_2(pcp, oval, nval) percpu_cmpxchg_op(2, , pcp, oval, nval) 546 #define raw_cpu_cmpxchg_4(pcp, oval, nval) percpu_cmpxchg_op(4, , pcp, oval, nval) 547 #define raw_cpu_try_cmpxchg_1(pcp, ovalp, nval) percpu_try_cmpxchg_op(1, , pcp, ovalp, nval) 548 #define raw_cpu_try_cmpxchg_2(pcp, ovalp, nval) percpu_try_cmpxchg_op(2, , pcp, ovalp, nval) 549 #define raw_cpu_try_cmpxchg_4(pcp, ovalp, nval) percpu_try_cmpxchg_op(4, , pcp, ovalp, nval) 550 551 #define this_cpu_add_return_1(pcp, val) percpu_add_return_op(1, volatile, pcp, val) 552 #define this_cpu_add_return_2(pcp, val) percpu_add_return_op(2, volatile, pcp, val) 553 #define this_cpu_add_return_4(pcp, val) percpu_add_return_op(4, volatile, pcp, val) 554 #define this_cpu_cmpxchg_1(pcp, oval, nval) percpu_cmpxchg_op(1, volatile, pcp, oval, nval) 555 #define this_cpu_cmpxchg_2(pcp, oval, nval) percpu_cmpxchg_op(2, volatile, pcp, oval, nval) 556 #define this_cpu_cmpxchg_4(pcp, oval, nval) percpu_cmpxchg_op(4, volatile, pcp, oval, nval) 557 #define this_cpu_try_cmpxchg_1(pcp, ovalp, nval) percpu_try_cmpxchg_op(1, volatile, pcp, ovalp, nval) 558 #define this_cpu_try_cmpxchg_2(pcp, ovalp, nval) percpu_try_cmpxchg_op(2, volatile, pcp, ovalp, nval) 559 #define this_cpu_try_cmpxchg_4(pcp, ovalp, nval) percpu_try_cmpxchg_op(4, volatile, pcp, ovalp, nval) 560 561 /* 562 * Per cpu atomic 64 bit operations are only available under 64 bit. 563 * 32 bit must fall back to generic operations. 564 */ 565 #ifdef CONFIG_X86_64 566 #define raw_cpu_add_8(pcp, val) percpu_add_op(8, , (pcp), val) 567 #define raw_cpu_and_8(pcp, val) percpu_to_op(8, , "and", (pcp), val) 568 #define raw_cpu_or_8(pcp, val) percpu_to_op(8, , "or", (pcp), val) 569 #define raw_cpu_add_return_8(pcp, val) percpu_add_return_op(8, , pcp, val) 570 #define raw_cpu_xchg_8(pcp, nval) raw_percpu_xchg_op(pcp, nval) 571 #define raw_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(8, , pcp, oval, nval) 572 #define raw_cpu_try_cmpxchg_8(pcp, ovalp, nval) percpu_try_cmpxchg_op(8, , pcp, ovalp, nval) 573 574 #define this_cpu_add_8(pcp, val) percpu_add_op(8, volatile, (pcp), val) 575 #define this_cpu_and_8(pcp, val) percpu_to_op(8, volatile, "and", (pcp), val) 576 #define this_cpu_or_8(pcp, val) percpu_to_op(8, volatile, "or", (pcp), val) 577 #define this_cpu_add_return_8(pcp, val) percpu_add_return_op(8, volatile, pcp, val) 578 #define this_cpu_xchg_8(pcp, nval) percpu_xchg_op(8, volatile, pcp, nval) 579 #define this_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(8, volatile, pcp, oval, nval) 580 #define this_cpu_try_cmpxchg_8(pcp, ovalp, nval) percpu_try_cmpxchg_op(8, volatile, pcp, ovalp, nval) 581 #endif 582 583 static __always_inline bool x86_this_cpu_constant_test_bit(unsigned int nr, 584 const unsigned long __percpu *addr) 585 { 586 unsigned long __percpu *a = 587 (unsigned long __percpu *)addr + nr / BITS_PER_LONG; 588 589 #ifdef CONFIG_X86_64 590 return ((1UL << (nr % BITS_PER_LONG)) & raw_cpu_read_8(*a)) != 0; 591 #else 592 return ((1UL << (nr % BITS_PER_LONG)) & raw_cpu_read_4(*a)) != 0; 593 #endif 594 } 595 596 static inline bool x86_this_cpu_variable_test_bit(int nr, 597 const unsigned long __percpu *addr) 598 { 599 bool oldbit; 600 601 asm volatile("btl "__percpu_arg(2)",%1" 602 CC_SET(c) 603 : CC_OUT(c) (oldbit) 604 : "m" (*__my_cpu_ptr((unsigned long __percpu *)(addr))), "Ir" (nr)); 605 606 return oldbit; 607 } 608 609 #define x86_this_cpu_test_bit(nr, addr) \ 610 (__builtin_constant_p((nr)) \ 611 ? x86_this_cpu_constant_test_bit((nr), (addr)) \ 612 : x86_this_cpu_variable_test_bit((nr), (addr))) 613 614 615 #include <asm-generic/percpu.h> 616 617 /* We can use this directly for local CPU (faster). */ 618 DECLARE_PER_CPU_READ_MOSTLY(unsigned long, this_cpu_off); 619 620 #endif /* !__ASSEMBLY__ */ 621 622 #ifdef CONFIG_SMP 623 624 /* 625 * Define the "EARLY_PER_CPU" macros. These are used for some per_cpu 626 * variables that are initialized and accessed before there are per_cpu 627 * areas allocated. 628 */ 629 630 #define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \ 631 DEFINE_PER_CPU(_type, _name) = _initvalue; \ 632 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \ 633 { [0 ... NR_CPUS-1] = _initvalue }; \ 634 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map 635 636 #define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \ 637 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue; \ 638 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \ 639 { [0 ... NR_CPUS-1] = _initvalue }; \ 640 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map 641 642 #define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \ 643 EXPORT_PER_CPU_SYMBOL(_name) 644 645 #define DECLARE_EARLY_PER_CPU(_type, _name) \ 646 DECLARE_PER_CPU(_type, _name); \ 647 extern __typeof__(_type) *_name##_early_ptr; \ 648 extern __typeof__(_type) _name##_early_map[] 649 650 #define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \ 651 DECLARE_PER_CPU_READ_MOSTLY(_type, _name); \ 652 extern __typeof__(_type) *_name##_early_ptr; \ 653 extern __typeof__(_type) _name##_early_map[] 654 655 #define early_per_cpu_ptr(_name) (_name##_early_ptr) 656 #define early_per_cpu_map(_name, _idx) (_name##_early_map[_idx]) 657 #define early_per_cpu(_name, _cpu) \ 658 *(early_per_cpu_ptr(_name) ? \ 659 &early_per_cpu_ptr(_name)[_cpu] : \ 660 &per_cpu(_name, _cpu)) 661 662 #else /* !CONFIG_SMP */ 663 #define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \ 664 DEFINE_PER_CPU(_type, _name) = _initvalue 665 666 #define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \ 667 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue 668 669 #define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \ 670 EXPORT_PER_CPU_SYMBOL(_name) 671 672 #define DECLARE_EARLY_PER_CPU(_type, _name) \ 673 DECLARE_PER_CPU(_type, _name) 674 675 #define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \ 676 DECLARE_PER_CPU_READ_MOSTLY(_type, _name) 677 678 #define early_per_cpu(_name, _cpu) per_cpu(_name, _cpu) 679 #define early_per_cpu_ptr(_name) NULL 680 /* no early_per_cpu_map() */ 681 682 #endif /* !CONFIG_SMP */ 683 684 #endif /* _ASM_X86_PERCPU_H */ 685