1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_BITOPS_H 3 #define _ASM_X86_BITOPS_H 4 5 /* 6 * Copyright 1992, Linus Torvalds. 7 * 8 * Note: inlines with more than a single statement should be marked 9 * __always_inline to avoid problems with older gcc's inlining heuristics. 10 */ 11 12 #ifndef _LINUX_BITOPS_H 13 #error only <linux/bitops.h> can be included directly 14 #endif 15 16 #include <linux/compiler.h> 17 #include <asm/alternative.h> 18 #include <asm/rmwcc.h> 19 #include <asm/barrier.h> 20 21 #if BITS_PER_LONG == 32 22 # define _BITOPS_LONG_SHIFT 5 23 #elif BITS_PER_LONG == 64 24 # define _BITOPS_LONG_SHIFT 6 25 #else 26 # error "Unexpected BITS_PER_LONG" 27 #endif 28 29 #define BIT_64(n) (U64_C(1) << (n)) 30 31 /* 32 * These have to be done with inline assembly: that way the bit-setting 33 * is guaranteed to be atomic. All bit operations return 0 if the bit 34 * was cleared before the operation and != 0 if it was not. 35 * 36 * bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1). 37 */ 38 39 #define RLONG_ADDR(x) "m" (*(volatile long *) (x)) 40 #define WBYTE_ADDR(x) "+m" (*(volatile char *) (x)) 41 42 #define ADDR RLONG_ADDR(addr) 43 44 /* 45 * We do the locked ops that don't return the old value as 46 * a mask operation on a byte. 47 */ 48 #define CONST_MASK_ADDR(nr, addr) WBYTE_ADDR((void *)(addr) + ((nr)>>3)) 49 #define CONST_MASK(nr) (1 << ((nr) & 7)) 50 51 static __always_inline void 52 arch_set_bit(long nr, volatile unsigned long *addr) 53 { 54 if (__builtin_constant_p(nr)) { 55 asm volatile(LOCK_PREFIX "orb %b1,%0" 56 : CONST_MASK_ADDR(nr, addr) 57 : "iq" (CONST_MASK(nr)) 58 : "memory"); 59 } else { 60 asm volatile(LOCK_PREFIX __ASM_SIZE(bts) " %1,%0" 61 : : RLONG_ADDR(addr), "Ir" (nr) : "memory"); 62 } 63 } 64 65 static __always_inline void 66 arch___set_bit(unsigned long nr, volatile unsigned long *addr) 67 { 68 asm volatile(__ASM_SIZE(bts) " %1,%0" : : ADDR, "Ir" (nr) : "memory"); 69 } 70 71 static __always_inline void 72 arch_clear_bit(long nr, volatile unsigned long *addr) 73 { 74 if (__builtin_constant_p(nr)) { 75 asm volatile(LOCK_PREFIX "andb %b1,%0" 76 : CONST_MASK_ADDR(nr, addr) 77 : "iq" (~CONST_MASK(nr))); 78 } else { 79 asm volatile(LOCK_PREFIX __ASM_SIZE(btr) " %1,%0" 80 : : RLONG_ADDR(addr), "Ir" (nr) : "memory"); 81 } 82 } 83 84 static __always_inline void 85 arch_clear_bit_unlock(long nr, volatile unsigned long *addr) 86 { 87 barrier(); 88 arch_clear_bit(nr, addr); 89 } 90 91 static __always_inline void 92 arch___clear_bit(unsigned long nr, volatile unsigned long *addr) 93 { 94 asm volatile(__ASM_SIZE(btr) " %1,%0" : : ADDR, "Ir" (nr) : "memory"); 95 } 96 97 static __always_inline bool 98 arch_clear_bit_unlock_is_negative_byte(long nr, volatile unsigned long *addr) 99 { 100 bool negative; 101 asm volatile(LOCK_PREFIX "andb %2,%1" 102 CC_SET(s) 103 : CC_OUT(s) (negative), WBYTE_ADDR(addr) 104 : "ir" ((char) ~(1 << nr)) : "memory"); 105 return negative; 106 } 107 #define arch_clear_bit_unlock_is_negative_byte \ 108 arch_clear_bit_unlock_is_negative_byte 109 110 static __always_inline void 111 arch___clear_bit_unlock(long nr, volatile unsigned long *addr) 112 { 113 arch___clear_bit(nr, addr); 114 } 115 116 static __always_inline void 117 arch___change_bit(unsigned long nr, volatile unsigned long *addr) 118 { 119 asm volatile(__ASM_SIZE(btc) " %1,%0" : : ADDR, "Ir" (nr) : "memory"); 120 } 121 122 static __always_inline void 123 arch_change_bit(long nr, volatile unsigned long *addr) 124 { 125 if (__builtin_constant_p(nr)) { 126 asm volatile(LOCK_PREFIX "xorb %b1,%0" 127 : CONST_MASK_ADDR(nr, addr) 128 : "iq" (CONST_MASK(nr))); 129 } else { 130 asm volatile(LOCK_PREFIX __ASM_SIZE(btc) " %1,%0" 131 : : RLONG_ADDR(addr), "Ir" (nr) : "memory"); 132 } 133 } 134 135 static __always_inline bool 136 arch_test_and_set_bit(long nr, volatile unsigned long *addr) 137 { 138 return GEN_BINARY_RMWcc(LOCK_PREFIX __ASM_SIZE(bts), *addr, c, "Ir", nr); 139 } 140 141 static __always_inline bool 142 arch_test_and_set_bit_lock(long nr, volatile unsigned long *addr) 143 { 144 return arch_test_and_set_bit(nr, addr); 145 } 146 147 static __always_inline bool 148 arch___test_and_set_bit(unsigned long nr, volatile unsigned long *addr) 149 { 150 bool oldbit; 151 152 asm(__ASM_SIZE(bts) " %2,%1" 153 CC_SET(c) 154 : CC_OUT(c) (oldbit) 155 : ADDR, "Ir" (nr) : "memory"); 156 return oldbit; 157 } 158 159 static __always_inline bool 160 arch_test_and_clear_bit(long nr, volatile unsigned long *addr) 161 { 162 return GEN_BINARY_RMWcc(LOCK_PREFIX __ASM_SIZE(btr), *addr, c, "Ir", nr); 163 } 164 165 /* 166 * Note: the operation is performed atomically with respect to 167 * the local CPU, but not other CPUs. Portable code should not 168 * rely on this behaviour. 169 * KVM relies on this behaviour on x86 for modifying memory that is also 170 * accessed from a hypervisor on the same CPU if running in a VM: don't change 171 * this without also updating arch/x86/kernel/kvm.c 172 */ 173 static __always_inline bool 174 arch___test_and_clear_bit(unsigned long nr, volatile unsigned long *addr) 175 { 176 bool oldbit; 177 178 asm volatile(__ASM_SIZE(btr) " %2,%1" 179 CC_SET(c) 180 : CC_OUT(c) (oldbit) 181 : ADDR, "Ir" (nr) : "memory"); 182 return oldbit; 183 } 184 185 static __always_inline bool 186 arch___test_and_change_bit(unsigned long nr, volatile unsigned long *addr) 187 { 188 bool oldbit; 189 190 asm volatile(__ASM_SIZE(btc) " %2,%1" 191 CC_SET(c) 192 : CC_OUT(c) (oldbit) 193 : ADDR, "Ir" (nr) : "memory"); 194 195 return oldbit; 196 } 197 198 static __always_inline bool 199 arch_test_and_change_bit(long nr, volatile unsigned long *addr) 200 { 201 return GEN_BINARY_RMWcc(LOCK_PREFIX __ASM_SIZE(btc), *addr, c, "Ir", nr); 202 } 203 204 static __always_inline bool constant_test_bit(long nr, const volatile unsigned long *addr) 205 { 206 return ((1UL << (nr & (BITS_PER_LONG-1))) & 207 (addr[nr >> _BITOPS_LONG_SHIFT])) != 0; 208 } 209 210 static __always_inline bool constant_test_bit_acquire(long nr, const volatile unsigned long *addr) 211 { 212 bool oldbit; 213 214 asm volatile("testb %2,%1" 215 CC_SET(nz) 216 : CC_OUT(nz) (oldbit) 217 : "m" (((unsigned char *)addr)[nr >> 3]), 218 "i" (1 << (nr & 7)) 219 :"memory"); 220 221 return oldbit; 222 } 223 224 static __always_inline bool variable_test_bit(long nr, volatile const unsigned long *addr) 225 { 226 bool oldbit; 227 228 asm volatile(__ASM_SIZE(bt) " %2,%1" 229 CC_SET(c) 230 : CC_OUT(c) (oldbit) 231 : "m" (*(unsigned long *)addr), "Ir" (nr) : "memory"); 232 233 return oldbit; 234 } 235 236 static __always_inline bool 237 arch_test_bit(unsigned long nr, const volatile unsigned long *addr) 238 { 239 return __builtin_constant_p(nr) ? constant_test_bit(nr, addr) : 240 variable_test_bit(nr, addr); 241 } 242 243 static __always_inline bool 244 arch_test_bit_acquire(unsigned long nr, const volatile unsigned long *addr) 245 { 246 return __builtin_constant_p(nr) ? constant_test_bit_acquire(nr, addr) : 247 variable_test_bit(nr, addr); 248 } 249 250 static __always_inline unsigned long variable__ffs(unsigned long word) 251 { 252 asm("rep; bsf %1,%0" 253 : "=r" (word) 254 : "rm" (word)); 255 return word; 256 } 257 258 /** 259 * __ffs - find first set bit in word 260 * @word: The word to search 261 * 262 * Undefined if no bit exists, so code should check against 0 first. 263 */ 264 #define __ffs(word) \ 265 (__builtin_constant_p(word) ? \ 266 (unsigned long)__builtin_ctzl(word) : \ 267 variable__ffs(word)) 268 269 static __always_inline unsigned long variable_ffz(unsigned long word) 270 { 271 asm("rep; bsf %1,%0" 272 : "=r" (word) 273 : "r" (~word)); 274 return word; 275 } 276 277 /** 278 * ffz - find first zero bit in word 279 * @word: The word to search 280 * 281 * Undefined if no zero exists, so code should check against ~0UL first. 282 */ 283 #define ffz(word) \ 284 (__builtin_constant_p(word) ? \ 285 (unsigned long)__builtin_ctzl(~word) : \ 286 variable_ffz(word)) 287 288 /* 289 * __fls: find last set bit in word 290 * @word: The word to search 291 * 292 * Undefined if no set bit exists, so code should check against 0 first. 293 */ 294 static __always_inline unsigned long __fls(unsigned long word) 295 { 296 asm("bsr %1,%0" 297 : "=r" (word) 298 : "rm" (word)); 299 return word; 300 } 301 302 #undef ADDR 303 304 #ifdef __KERNEL__ 305 static __always_inline int variable_ffs(int x) 306 { 307 int r; 308 309 #ifdef CONFIG_X86_64 310 /* 311 * AMD64 says BSFL won't clobber the dest reg if x==0; Intel64 says the 312 * dest reg is undefined if x==0, but their CPU architect says its 313 * value is written to set it to the same as before, except that the 314 * top 32 bits will be cleared. 315 * 316 * We cannot do this on 32 bits because at the very least some 317 * 486 CPUs did not behave this way. 318 */ 319 asm("bsfl %1,%0" 320 : "=r" (r) 321 : "rm" (x), "0" (-1)); 322 #elif defined(CONFIG_X86_CMOV) 323 asm("bsfl %1,%0\n\t" 324 "cmovzl %2,%0" 325 : "=&r" (r) : "rm" (x), "r" (-1)); 326 #else 327 asm("bsfl %1,%0\n\t" 328 "jnz 1f\n\t" 329 "movl $-1,%0\n" 330 "1:" : "=r" (r) : "rm" (x)); 331 #endif 332 return r + 1; 333 } 334 335 /** 336 * ffs - find first set bit in word 337 * @x: the word to search 338 * 339 * This is defined the same way as the libc and compiler builtin ffs 340 * routines, therefore differs in spirit from the other bitops. 341 * 342 * ffs(value) returns 0 if value is 0 or the position of the first 343 * set bit if value is nonzero. The first (least significant) bit 344 * is at position 1. 345 */ 346 #define ffs(x) (__builtin_constant_p(x) ? __builtin_ffs(x) : variable_ffs(x)) 347 348 /** 349 * fls - find last set bit in word 350 * @x: the word to search 351 * 352 * This is defined in a similar way as the libc and compiler builtin 353 * ffs, but returns the position of the most significant set bit. 354 * 355 * fls(value) returns 0 if value is 0 or the position of the last 356 * set bit if value is nonzero. The last (most significant) bit is 357 * at position 32. 358 */ 359 static __always_inline int fls(unsigned int x) 360 { 361 int r; 362 363 #ifdef CONFIG_X86_64 364 /* 365 * AMD64 says BSRL won't clobber the dest reg if x==0; Intel64 says the 366 * dest reg is undefined if x==0, but their CPU architect says its 367 * value is written to set it to the same as before, except that the 368 * top 32 bits will be cleared. 369 * 370 * We cannot do this on 32 bits because at the very least some 371 * 486 CPUs did not behave this way. 372 */ 373 asm("bsrl %1,%0" 374 : "=r" (r) 375 : "rm" (x), "0" (-1)); 376 #elif defined(CONFIG_X86_CMOV) 377 asm("bsrl %1,%0\n\t" 378 "cmovzl %2,%0" 379 : "=&r" (r) : "rm" (x), "rm" (-1)); 380 #else 381 asm("bsrl %1,%0\n\t" 382 "jnz 1f\n\t" 383 "movl $-1,%0\n" 384 "1:" : "=r" (r) : "rm" (x)); 385 #endif 386 return r + 1; 387 } 388 389 /** 390 * fls64 - find last set bit in a 64-bit word 391 * @x: the word to search 392 * 393 * This is defined in a similar way as the libc and compiler builtin 394 * ffsll, but returns the position of the most significant set bit. 395 * 396 * fls64(value) returns 0 if value is 0 or the position of the last 397 * set bit if value is nonzero. The last (most significant) bit is 398 * at position 64. 399 */ 400 #ifdef CONFIG_X86_64 401 static __always_inline int fls64(__u64 x) 402 { 403 int bitpos = -1; 404 /* 405 * AMD64 says BSRQ won't clobber the dest reg if x==0; Intel64 says the 406 * dest reg is undefined if x==0, but their CPU architect says its 407 * value is written to set it to the same as before. 408 */ 409 asm("bsrq %1,%q0" 410 : "+r" (bitpos) 411 : "rm" (x)); 412 return bitpos + 1; 413 } 414 #else 415 #include <asm-generic/bitops/fls64.h> 416 #endif 417 418 #include <asm-generic/bitops/sched.h> 419 420 #include <asm/arch_hweight.h> 421 422 #include <asm-generic/bitops/const_hweight.h> 423 424 #include <asm-generic/bitops/instrumented-atomic.h> 425 #include <asm-generic/bitops/instrumented-non-atomic.h> 426 #include <asm-generic/bitops/instrumented-lock.h> 427 428 #include <asm-generic/bitops/le.h> 429 430 #include <asm-generic/bitops/ext2-atomic-setbit.h> 431 432 #endif /* __KERNEL__ */ 433 #endif /* _ASM_X86_BITOPS_H */ 434