1 #ifndef _ASM_X86_PERCPU_H 2 #define _ASM_X86_PERCPU_H 3 4 #ifdef CONFIG_X86_64 5 #define __percpu_seg gs 6 #define __percpu_mov_op movq 7 #else 8 #define __percpu_seg fs 9 #define __percpu_mov_op movl 10 #endif 11 12 #ifdef __ASSEMBLY__ 13 14 /* 15 * PER_CPU finds an address of a per-cpu variable. 16 * 17 * Args: 18 * var - variable name 19 * reg - 32bit register 20 * 21 * The resulting address is stored in the "reg" argument. 22 * 23 * Example: 24 * PER_CPU(cpu_gdt_descr, %ebx) 25 */ 26 #ifdef CONFIG_SMP 27 #define PER_CPU(var, reg) \ 28 __percpu_mov_op %__percpu_seg:this_cpu_off, reg; \ 29 lea var(reg), reg 30 #define PER_CPU_VAR(var) %__percpu_seg:var 31 #else /* ! SMP */ 32 #define PER_CPU(var, reg) __percpu_mov_op $var, reg 33 #define PER_CPU_VAR(var) var 34 #endif /* SMP */ 35 36 #ifdef CONFIG_X86_64_SMP 37 #define INIT_PER_CPU_VAR(var) init_per_cpu__##var 38 #else 39 #define INIT_PER_CPU_VAR(var) var 40 #endif 41 42 #else /* ...!ASSEMBLY */ 43 44 #include <linux/kernel.h> 45 #include <linux/stringify.h> 46 47 #ifdef CONFIG_SMP 48 #define __percpu_arg(x) "%%"__stringify(__percpu_seg)":%P" #x 49 #define __my_cpu_offset percpu_read(this_cpu_off) 50 #else 51 #define __percpu_arg(x) "%P" #x 52 #endif 53 54 /* 55 * Initialized pointers to per-cpu variables needed for the boot 56 * processor need to use these macros to get the proper address 57 * offset from __per_cpu_load on SMP. 58 * 59 * There also must be an entry in vmlinux_64.lds.S 60 */ 61 #define DECLARE_INIT_PER_CPU(var) \ 62 extern typeof(var) init_per_cpu_var(var) 63 64 #ifdef CONFIG_X86_64_SMP 65 #define init_per_cpu_var(var) init_per_cpu__##var 66 #else 67 #define init_per_cpu_var(var) var 68 #endif 69 70 /* For arch-specific code, we can use direct single-insn ops (they 71 * don't give an lvalue though). */ 72 extern void __bad_percpu_size(void); 73 74 #define percpu_to_op(op, var, val) \ 75 do { \ 76 typedef typeof(var) pto_T__; \ 77 if (0) { \ 78 pto_T__ pto_tmp__; \ 79 pto_tmp__ = (val); \ 80 } \ 81 switch (sizeof(var)) { \ 82 case 1: \ 83 asm(op "b %1,"__percpu_arg(0) \ 84 : "+m" (var) \ 85 : "qi" ((pto_T__)(val))); \ 86 break; \ 87 case 2: \ 88 asm(op "w %1,"__percpu_arg(0) \ 89 : "+m" (var) \ 90 : "ri" ((pto_T__)(val))); \ 91 break; \ 92 case 4: \ 93 asm(op "l %1,"__percpu_arg(0) \ 94 : "+m" (var) \ 95 : "ri" ((pto_T__)(val))); \ 96 break; \ 97 case 8: \ 98 asm(op "q %1,"__percpu_arg(0) \ 99 : "+m" (var) \ 100 : "re" ((pto_T__)(val))); \ 101 break; \ 102 default: __bad_percpu_size(); \ 103 } \ 104 } while (0) 105 106 /* 107 * Generate a percpu add to memory instruction and optimize code 108 * if one is added or subtracted. 109 */ 110 #define percpu_add_op(var, val) \ 111 do { \ 112 typedef typeof(var) pao_T__; \ 113 const int pao_ID__ = (__builtin_constant_p(val) && \ 114 ((val) == 1 || (val) == -1)) ? (val) : 0; \ 115 if (0) { \ 116 pao_T__ pao_tmp__; \ 117 pao_tmp__ = (val); \ 118 } \ 119 switch (sizeof(var)) { \ 120 case 1: \ 121 if (pao_ID__ == 1) \ 122 asm("incb "__percpu_arg(0) : "+m" (var)); \ 123 else if (pao_ID__ == -1) \ 124 asm("decb "__percpu_arg(0) : "+m" (var)); \ 125 else \ 126 asm("addb %1, "__percpu_arg(0) \ 127 : "+m" (var) \ 128 : "qi" ((pao_T__)(val))); \ 129 break; \ 130 case 2: \ 131 if (pao_ID__ == 1) \ 132 asm("incw "__percpu_arg(0) : "+m" (var)); \ 133 else if (pao_ID__ == -1) \ 134 asm("decw "__percpu_arg(0) : "+m" (var)); \ 135 else \ 136 asm("addw %1, "__percpu_arg(0) \ 137 : "+m" (var) \ 138 : "ri" ((pao_T__)(val))); \ 139 break; \ 140 case 4: \ 141 if (pao_ID__ == 1) \ 142 asm("incl "__percpu_arg(0) : "+m" (var)); \ 143 else if (pao_ID__ == -1) \ 144 asm("decl "__percpu_arg(0) : "+m" (var)); \ 145 else \ 146 asm("addl %1, "__percpu_arg(0) \ 147 : "+m" (var) \ 148 : "ri" ((pao_T__)(val))); \ 149 break; \ 150 case 8: \ 151 if (pao_ID__ == 1) \ 152 asm("incq "__percpu_arg(0) : "+m" (var)); \ 153 else if (pao_ID__ == -1) \ 154 asm("decq "__percpu_arg(0) : "+m" (var)); \ 155 else \ 156 asm("addq %1, "__percpu_arg(0) \ 157 : "+m" (var) \ 158 : "re" ((pao_T__)(val))); \ 159 break; \ 160 default: __bad_percpu_size(); \ 161 } \ 162 } while (0) 163 164 #define percpu_from_op(op, var, constraint) \ 165 ({ \ 166 typeof(var) pfo_ret__; \ 167 switch (sizeof(var)) { \ 168 case 1: \ 169 asm(op "b "__percpu_arg(1)",%0" \ 170 : "=q" (pfo_ret__) \ 171 : constraint); \ 172 break; \ 173 case 2: \ 174 asm(op "w "__percpu_arg(1)",%0" \ 175 : "=r" (pfo_ret__) \ 176 : constraint); \ 177 break; \ 178 case 4: \ 179 asm(op "l "__percpu_arg(1)",%0" \ 180 : "=r" (pfo_ret__) \ 181 : constraint); \ 182 break; \ 183 case 8: \ 184 asm(op "q "__percpu_arg(1)",%0" \ 185 : "=r" (pfo_ret__) \ 186 : constraint); \ 187 break; \ 188 default: __bad_percpu_size(); \ 189 } \ 190 pfo_ret__; \ 191 }) 192 193 #define percpu_unary_op(op, var) \ 194 ({ \ 195 switch (sizeof(var)) { \ 196 case 1: \ 197 asm(op "b "__percpu_arg(0) \ 198 : "+m" (var)); \ 199 break; \ 200 case 2: \ 201 asm(op "w "__percpu_arg(0) \ 202 : "+m" (var)); \ 203 break; \ 204 case 4: \ 205 asm(op "l "__percpu_arg(0) \ 206 : "+m" (var)); \ 207 break; \ 208 case 8: \ 209 asm(op "q "__percpu_arg(0) \ 210 : "+m" (var)); \ 211 break; \ 212 default: __bad_percpu_size(); \ 213 } \ 214 }) 215 216 /* 217 * percpu_read() makes gcc load the percpu variable every time it is 218 * accessed while percpu_read_stable() allows the value to be cached. 219 * percpu_read_stable() is more efficient and can be used if its value 220 * is guaranteed to be valid across cpus. The current users include 221 * get_current() and get_thread_info() both of which are actually 222 * per-thread variables implemented as per-cpu variables and thus 223 * stable for the duration of the respective task. 224 */ 225 #define percpu_read(var) percpu_from_op("mov", var, "m" (var)) 226 #define percpu_read_stable(var) percpu_from_op("mov", var, "p" (&(var))) 227 #define percpu_write(var, val) percpu_to_op("mov", var, val) 228 #define percpu_add(var, val) percpu_add_op(var, val) 229 #define percpu_sub(var, val) percpu_add_op(var, -(val)) 230 #define percpu_and(var, val) percpu_to_op("and", var, val) 231 #define percpu_or(var, val) percpu_to_op("or", var, val) 232 #define percpu_xor(var, val) percpu_to_op("xor", var, val) 233 #define percpu_inc(var) percpu_unary_op("inc", var) 234 235 #define __this_cpu_read_1(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 236 #define __this_cpu_read_2(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 237 #define __this_cpu_read_4(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 238 239 #define __this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val) 240 #define __this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val) 241 #define __this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val) 242 #define __this_cpu_add_1(pcp, val) percpu_add_op((pcp), val) 243 #define __this_cpu_add_2(pcp, val) percpu_add_op((pcp), val) 244 #define __this_cpu_add_4(pcp, val) percpu_add_op((pcp), val) 245 #define __this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val) 246 #define __this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val) 247 #define __this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val) 248 #define __this_cpu_or_1(pcp, val) percpu_to_op("or", (pcp), val) 249 #define __this_cpu_or_2(pcp, val) percpu_to_op("or", (pcp), val) 250 #define __this_cpu_or_4(pcp, val) percpu_to_op("or", (pcp), val) 251 #define __this_cpu_xor_1(pcp, val) percpu_to_op("xor", (pcp), val) 252 #define __this_cpu_xor_2(pcp, val) percpu_to_op("xor", (pcp), val) 253 #define __this_cpu_xor_4(pcp, val) percpu_to_op("xor", (pcp), val) 254 255 #define this_cpu_read_1(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 256 #define this_cpu_read_2(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 257 #define this_cpu_read_4(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 258 #define this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val) 259 #define this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val) 260 #define this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val) 261 #define this_cpu_add_1(pcp, val) percpu_add_op((pcp), val) 262 #define this_cpu_add_2(pcp, val) percpu_add_op((pcp), val) 263 #define this_cpu_add_4(pcp, val) percpu_add_op((pcp), val) 264 #define this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val) 265 #define this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val) 266 #define this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val) 267 #define this_cpu_or_1(pcp, val) percpu_to_op("or", (pcp), val) 268 #define this_cpu_or_2(pcp, val) percpu_to_op("or", (pcp), val) 269 #define this_cpu_or_4(pcp, val) percpu_to_op("or", (pcp), val) 270 #define this_cpu_xor_1(pcp, val) percpu_to_op("xor", (pcp), val) 271 #define this_cpu_xor_2(pcp, val) percpu_to_op("xor", (pcp), val) 272 #define this_cpu_xor_4(pcp, val) percpu_to_op("xor", (pcp), val) 273 274 #define irqsafe_cpu_add_1(pcp, val) percpu_add_op((pcp), val) 275 #define irqsafe_cpu_add_2(pcp, val) percpu_add_op((pcp), val) 276 #define irqsafe_cpu_add_4(pcp, val) percpu_add_op((pcp), val) 277 #define irqsafe_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val) 278 #define irqsafe_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val) 279 #define irqsafe_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val) 280 #define irqsafe_cpu_or_1(pcp, val) percpu_to_op("or", (pcp), val) 281 #define irqsafe_cpu_or_2(pcp, val) percpu_to_op("or", (pcp), val) 282 #define irqsafe_cpu_or_4(pcp, val) percpu_to_op("or", (pcp), val) 283 #define irqsafe_cpu_xor_1(pcp, val) percpu_to_op("xor", (pcp), val) 284 #define irqsafe_cpu_xor_2(pcp, val) percpu_to_op("xor", (pcp), val) 285 #define irqsafe_cpu_xor_4(pcp, val) percpu_to_op("xor", (pcp), val) 286 287 /* 288 * Per cpu atomic 64 bit operations are only available under 64 bit. 289 * 32 bit must fall back to generic operations. 290 */ 291 #ifdef CONFIG_X86_64 292 #define __this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 293 #define __this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val) 294 #define __this_cpu_add_8(pcp, val) percpu_add_op((pcp), val) 295 #define __this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val) 296 #define __this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val) 297 #define __this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val) 298 299 #define this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 300 #define this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val) 301 #define this_cpu_add_8(pcp, val) percpu_add_op((pcp), val) 302 #define this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val) 303 #define this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val) 304 #define this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val) 305 306 #define irqsafe_cpu_add_8(pcp, val) percpu_add_op((pcp), val) 307 #define irqsafe_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val) 308 #define irqsafe_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val) 309 #define irqsafe_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val) 310 311 #endif 312 313 /* This is not atomic against other CPUs -- CPU preemption needs to be off */ 314 #define x86_test_and_clear_bit_percpu(bit, var) \ 315 ({ \ 316 int old__; \ 317 asm volatile("btr %2,"__percpu_arg(1)"\n\tsbbl %0,%0" \ 318 : "=r" (old__), "+m" (var) \ 319 : "dIr" (bit)); \ 320 old__; \ 321 }) 322 323 #include <asm-generic/percpu.h> 324 325 /* We can use this directly for local CPU (faster). */ 326 DECLARE_PER_CPU(unsigned long, this_cpu_off); 327 328 #endif /* !__ASSEMBLY__ */ 329 330 #ifdef CONFIG_SMP 331 332 /* 333 * Define the "EARLY_PER_CPU" macros. These are used for some per_cpu 334 * variables that are initialized and accessed before there are per_cpu 335 * areas allocated. 336 */ 337 338 #define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \ 339 DEFINE_PER_CPU(_type, _name) = _initvalue; \ 340 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \ 341 { [0 ... NR_CPUS-1] = _initvalue }; \ 342 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map 343 344 #define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \ 345 EXPORT_PER_CPU_SYMBOL(_name) 346 347 #define DECLARE_EARLY_PER_CPU(_type, _name) \ 348 DECLARE_PER_CPU(_type, _name); \ 349 extern __typeof__(_type) *_name##_early_ptr; \ 350 extern __typeof__(_type) _name##_early_map[] 351 352 #define early_per_cpu_ptr(_name) (_name##_early_ptr) 353 #define early_per_cpu_map(_name, _idx) (_name##_early_map[_idx]) 354 #define early_per_cpu(_name, _cpu) \ 355 *(early_per_cpu_ptr(_name) ? \ 356 &early_per_cpu_ptr(_name)[_cpu] : \ 357 &per_cpu(_name, _cpu)) 358 359 #else /* !CONFIG_SMP */ 360 #define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \ 361 DEFINE_PER_CPU(_type, _name) = _initvalue 362 363 #define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \ 364 EXPORT_PER_CPU_SYMBOL(_name) 365 366 #define DECLARE_EARLY_PER_CPU(_type, _name) \ 367 DECLARE_PER_CPU(_type, _name) 368 369 #define early_per_cpu(_name, _cpu) per_cpu(_name, _cpu) 370 #define early_per_cpu_ptr(_name) NULL 371 /* no early_per_cpu_map() */ 372 373 #endif /* !CONFIG_SMP */ 374 375 #endif /* _ASM_X86_PERCPU_H */ 376