1 /* 2 * Oct 15, 2000 Matt Domsch <Matt_Domsch@dell.com> 3 * Nicer crc32 functions/docs submitted by linux@horizon.com. Thanks! 4 * Code was from the public domain, copyright abandoned. Code was 5 * subsequently included in the kernel, thus was re-licensed under the 6 * GNU GPL v2. 7 * 8 * Oct 12, 2000 Matt Domsch <Matt_Domsch@dell.com> 9 * Same crc32 function was used in 5 other places in the kernel. 10 * I made one version, and deleted the others. 11 * There are various incantations of crc32(). Some use a seed of 0 or ~0. 12 * Some xor at the end with ~0. The generic crc32() function takes 13 * seed as an argument, and doesn't xor at the end. Then individual 14 * users can do whatever they need. 15 * drivers/net/smc9194.c uses seed ~0, doesn't xor with ~0. 16 * fs/jffs2 uses seed 0, doesn't xor with ~0. 17 * fs/partitions/efi.c uses seed ~0, xor's with ~0. 18 * 19 * This source code is licensed under the GNU General Public License, 20 * Version 2. See the file COPYING for more details. 21 */ 22 23 #ifdef UBI_LINUX 24 #include <linux/crc32.h> 25 #include <linux/kernel.h> 26 #include <linux/module.h> 27 #include <linux/compiler.h> 28 #endif 29 #include <linux/types.h> 30 31 #include <asm/byteorder.h> 32 33 #ifdef UBI_LINUX 34 #include <linux/slab.h> 35 #include <linux/init.h> 36 #include <asm/atomic.h> 37 #endif 38 #include "crc32defs.h" 39 #define CRC_LE_BITS 8 40 41 # define __force 42 #ifndef __constant_cpu_to_le32 43 #define __constant_cpu_to_le32(x) ((__force __le32)(__u32)(x)) 44 #endif 45 #ifndef __constant_le32_to_cpu 46 #define __constant_le32_to_cpu(x) ((__force __u32)(__le32)(x)) 47 #endif 48 49 #if CRC_LE_BITS == 8 50 #define tole(x) __constant_cpu_to_le32(x) 51 #define tobe(x) __constant_cpu_to_be32(x) 52 #else 53 #define tole(x) (x) 54 #define tobe(x) (x) 55 #endif 56 #include "crc32table.h" 57 #ifdef UBI_LINUX 58 MODULE_AUTHOR("Matt Domsch <Matt_Domsch@dell.com>"); 59 MODULE_DESCRIPTION("Ethernet CRC32 calculations"); 60 MODULE_LICENSE("GPL"); 61 #endif 62 /** 63 * crc32_le() - Calculate bitwise little-endian Ethernet AUTODIN II CRC32 64 * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for 65 * other uses, or the previous crc32 value if computing incrementally. 66 * @p: pointer to buffer over which CRC is run 67 * @len: length of buffer @p 68 */ 69 u32 crc32_le(u32 crc, unsigned char const *p, size_t len); 70 71 #if CRC_LE_BITS == 1 72 /* 73 * In fact, the table-based code will work in this case, but it can be 74 * simplified by inlining the table in ?: form. 75 */ 76 77 u32 crc32_le(u32 crc, unsigned char const *p, size_t len) 78 { 79 int i; 80 while (len--) { 81 crc ^= *p++; 82 for (i = 0; i < 8; i++) 83 crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_LE : 0); 84 } 85 return crc; 86 } 87 #else /* Table-based approach */ 88 89 u32 crc32_le(u32 crc, unsigned char const *p, size_t len) 90 { 91 # if CRC_LE_BITS == 8 92 const u32 *b =(u32 *)p; 93 const u32 *tab = crc32table_le; 94 95 # ifdef __LITTLE_ENDIAN 96 # define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8) 97 # else 98 # define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8) 99 # endif 100 /* printf("Crc32_le crc=%x\n",crc); */ 101 crc = __cpu_to_le32(crc); 102 /* Align it */ 103 if((((long)b)&3 && len)){ 104 do { 105 u8 *p = (u8 *)b; 106 DO_CRC(*p++); 107 b = (void *)p; 108 } while ((--len) && ((long)b)&3 ); 109 } 110 if((len >= 4)){ 111 /* load data 32 bits wide, xor data 32 bits wide. */ 112 size_t save_len = len & 3; 113 len = len >> 2; 114 --b; /* use pre increment below(*++b) for speed */ 115 do { 116 crc ^= *++b; 117 DO_CRC(0); 118 DO_CRC(0); 119 DO_CRC(0); 120 DO_CRC(0); 121 } while (--len); 122 b++; /* point to next byte(s) */ 123 len = save_len; 124 } 125 /* And the last few bytes */ 126 if(len){ 127 do { 128 u8 *p = (u8 *)b; 129 DO_CRC(*p++); 130 b = (void *)p; 131 } while (--len); 132 } 133 134 return __le32_to_cpu(crc); 135 #undef ENDIAN_SHIFT 136 #undef DO_CRC 137 138 # elif CRC_LE_BITS == 4 139 while (len--) { 140 crc ^= *p++; 141 crc = (crc >> 4) ^ crc32table_le[crc & 15]; 142 crc = (crc >> 4) ^ crc32table_le[crc & 15]; 143 } 144 return crc; 145 # elif CRC_LE_BITS == 2 146 while (len--) { 147 crc ^= *p++; 148 crc = (crc >> 2) ^ crc32table_le[crc & 3]; 149 crc = (crc >> 2) ^ crc32table_le[crc & 3]; 150 crc = (crc >> 2) ^ crc32table_le[crc & 3]; 151 crc = (crc >> 2) ^ crc32table_le[crc & 3]; 152 } 153 return crc; 154 # endif 155 } 156 #endif 157 #ifdef UBI_LINUX 158 /** 159 * crc32_be() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32 160 * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for 161 * other uses, or the previous crc32 value if computing incrementally. 162 * @p: pointer to buffer over which CRC is run 163 * @len: length of buffer @p 164 */ 165 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len); 166 167 #if CRC_BE_BITS == 1 168 /* 169 * In fact, the table-based code will work in this case, but it can be 170 * simplified by inlining the table in ?: form. 171 */ 172 173 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len) 174 { 175 int i; 176 while (len--) { 177 crc ^= *p++ << 24; 178 for (i = 0; i < 8; i++) 179 crc = 180 (crc << 1) ^ ((crc & 0x80000000) ? CRCPOLY_BE : 181 0); 182 } 183 return crc; 184 } 185 186 #else /* Table-based approach */ 187 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len) 188 { 189 # if CRC_BE_BITS == 8 190 const u32 *b =(u32 *)p; 191 const u32 *tab = crc32table_be; 192 193 # ifdef __LITTLE_ENDIAN 194 # define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8) 195 # else 196 # define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8) 197 # endif 198 199 crc = __cpu_to_be32(crc); 200 /* Align it */ 201 if(unlikely(((long)b)&3 && len)){ 202 do { 203 u8 *p = (u8 *)b; 204 DO_CRC(*p++); 205 b = (u32 *)p; 206 } while ((--len) && ((long)b)&3 ); 207 } 208 if(likely(len >= 4)){ 209 /* load data 32 bits wide, xor data 32 bits wide. */ 210 size_t save_len = len & 3; 211 len = len >> 2; 212 --b; /* use pre increment below(*++b) for speed */ 213 do { 214 crc ^= *++b; 215 DO_CRC(0); 216 DO_CRC(0); 217 DO_CRC(0); 218 DO_CRC(0); 219 } while (--len); 220 b++; /* point to next byte(s) */ 221 len = save_len; 222 } 223 /* And the last few bytes */ 224 if(len){ 225 do { 226 u8 *p = (u8 *)b; 227 DO_CRC(*p++); 228 b = (void *)p; 229 } while (--len); 230 } 231 return __be32_to_cpu(crc); 232 #undef ENDIAN_SHIFT 233 #undef DO_CRC 234 235 # elif CRC_BE_BITS == 4 236 while (len--) { 237 crc ^= *p++ << 24; 238 crc = (crc << 4) ^ crc32table_be[crc >> 28]; 239 crc = (crc << 4) ^ crc32table_be[crc >> 28]; 240 } 241 return crc; 242 # elif CRC_BE_BITS == 2 243 while (len--) { 244 crc ^= *p++ << 24; 245 crc = (crc << 2) ^ crc32table_be[crc >> 30]; 246 crc = (crc << 2) ^ crc32table_be[crc >> 30]; 247 crc = (crc << 2) ^ crc32table_be[crc >> 30]; 248 crc = (crc << 2) ^ crc32table_be[crc >> 30]; 249 } 250 return crc; 251 # endif 252 } 253 #endif 254 255 EXPORT_SYMBOL(crc32_le); 256 EXPORT_SYMBOL(crc32_be); 257 #endif 258 /* 259 * A brief CRC tutorial. 260 * 261 * A CRC is a long-division remainder. You add the CRC to the message, 262 * and the whole thing (message+CRC) is a multiple of the given 263 * CRC polynomial. To check the CRC, you can either check that the 264 * CRC matches the recomputed value, *or* you can check that the 265 * remainder computed on the message+CRC is 0. This latter approach 266 * is used by a lot of hardware implementations, and is why so many 267 * protocols put the end-of-frame flag after the CRC. 268 * 269 * It's actually the same long division you learned in school, except that 270 * - We're working in binary, so the digits are only 0 and 1, and 271 * - When dividing polynomials, there are no carries. Rather than add and 272 * subtract, we just xor. Thus, we tend to get a bit sloppy about 273 * the difference between adding and subtracting. 274 * 275 * A 32-bit CRC polynomial is actually 33 bits long. But since it's 276 * 33 bits long, bit 32 is always going to be set, so usually the CRC 277 * is written in hex with the most significant bit omitted. (If you're 278 * familiar with the IEEE 754 floating-point format, it's the same idea.) 279 * 280 * Note that a CRC is computed over a string of *bits*, so you have 281 * to decide on the endianness of the bits within each byte. To get 282 * the best error-detecting properties, this should correspond to the 283 * order they're actually sent. For example, standard RS-232 serial is 284 * little-endian; the most significant bit (sometimes used for parity) 285 * is sent last. And when appending a CRC word to a message, you should 286 * do it in the right order, matching the endianness. 287 * 288 * Just like with ordinary division, the remainder is always smaller than 289 * the divisor (the CRC polynomial) you're dividing by. Each step of the 290 * division, you take one more digit (bit) of the dividend and append it 291 * to the current remainder. Then you figure out the appropriate multiple 292 * of the divisor to subtract to being the remainder back into range. 293 * In binary, it's easy - it has to be either 0 or 1, and to make the 294 * XOR cancel, it's just a copy of bit 32 of the remainder. 295 * 296 * When computing a CRC, we don't care about the quotient, so we can 297 * throw the quotient bit away, but subtract the appropriate multiple of 298 * the polynomial from the remainder and we're back to where we started, 299 * ready to process the next bit. 300 * 301 * A big-endian CRC written this way would be coded like: 302 * for (i = 0; i < input_bits; i++) { 303 * multiple = remainder & 0x80000000 ? CRCPOLY : 0; 304 * remainder = (remainder << 1 | next_input_bit()) ^ multiple; 305 * } 306 * Notice how, to get at bit 32 of the shifted remainder, we look 307 * at bit 31 of the remainder *before* shifting it. 308 * 309 * But also notice how the next_input_bit() bits we're shifting into 310 * the remainder don't actually affect any decision-making until 311 * 32 bits later. Thus, the first 32 cycles of this are pretty boring. 312 * Also, to add the CRC to a message, we need a 32-bit-long hole for it at 313 * the end, so we have to add 32 extra cycles shifting in zeros at the 314 * end of every message, 315 * 316 * So the standard trick is to rearrage merging in the next_input_bit() 317 * until the moment it's needed. Then the first 32 cycles can be precomputed, 318 * and merging in the final 32 zero bits to make room for the CRC can be 319 * skipped entirely. 320 * This changes the code to: 321 * for (i = 0; i < input_bits; i++) { 322 * remainder ^= next_input_bit() << 31; 323 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; 324 * remainder = (remainder << 1) ^ multiple; 325 * } 326 * With this optimization, the little-endian code is simpler: 327 * for (i = 0; i < input_bits; i++) { 328 * remainder ^= next_input_bit(); 329 * multiple = (remainder & 1) ? CRCPOLY : 0; 330 * remainder = (remainder >> 1) ^ multiple; 331 * } 332 * 333 * Note that the other details of endianness have been hidden in CRCPOLY 334 * (which must be bit-reversed) and next_input_bit(). 335 * 336 * However, as long as next_input_bit is returning the bits in a sensible 337 * order, we can actually do the merging 8 or more bits at a time rather 338 * than one bit at a time: 339 * for (i = 0; i < input_bytes; i++) { 340 * remainder ^= next_input_byte() << 24; 341 * for (j = 0; j < 8; j++) { 342 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; 343 * remainder = (remainder << 1) ^ multiple; 344 * } 345 * } 346 * Or in little-endian: 347 * for (i = 0; i < input_bytes; i++) { 348 * remainder ^= next_input_byte(); 349 * for (j = 0; j < 8; j++) { 350 * multiple = (remainder & 1) ? CRCPOLY : 0; 351 * remainder = (remainder << 1) ^ multiple; 352 * } 353 * } 354 * If the input is a multiple of 32 bits, you can even XOR in a 32-bit 355 * word at a time and increase the inner loop count to 32. 356 * 357 * You can also mix and match the two loop styles, for example doing the 358 * bulk of a message byte-at-a-time and adding bit-at-a-time processing 359 * for any fractional bytes at the end. 360 * 361 * The only remaining optimization is to the byte-at-a-time table method. 362 * Here, rather than just shifting one bit of the remainder to decide 363 * in the correct multiple to subtract, we can shift a byte at a time. 364 * This produces a 40-bit (rather than a 33-bit) intermediate remainder, 365 * but again the multiple of the polynomial to subtract depends only on 366 * the high bits, the high 8 bits in this case. 367 * 368 * The multile we need in that case is the low 32 bits of a 40-bit 369 * value whose high 8 bits are given, and which is a multiple of the 370 * generator polynomial. This is simply the CRC-32 of the given 371 * one-byte message. 372 * 373 * Two more details: normally, appending zero bits to a message which 374 * is already a multiple of a polynomial produces a larger multiple of that 375 * polynomial. To enable a CRC to detect this condition, it's common to 376 * invert the CRC before appending it. This makes the remainder of the 377 * message+crc come out not as zero, but some fixed non-zero value. 378 * 379 * The same problem applies to zero bits prepended to the message, and 380 * a similar solution is used. Instead of starting with a remainder of 381 * 0, an initial remainder of all ones is used. As long as you start 382 * the same way on decoding, it doesn't make a difference. 383 */ 384 385 #ifdef UNITTEST 386 387 #include <stdlib.h> 388 #include <stdio.h> 389 390 #ifdef UBI_LINUX /*Not used at present */ 391 static void 392 buf_dump(char const *prefix, unsigned char const *buf, size_t len) 393 { 394 fputs(prefix, stdout); 395 while (len--) 396 printf(" %02x", *buf++); 397 putchar('\n'); 398 399 } 400 #endif 401 402 static void bytereverse(unsigned char *buf, size_t len) 403 { 404 while (len--) { 405 unsigned char x = bitrev8(*buf); 406 *buf++ = x; 407 } 408 } 409 410 static void random_garbage(unsigned char *buf, size_t len) 411 { 412 while (len--) 413 *buf++ = (unsigned char) random(); 414 } 415 416 #ifdef UBI_LINUX /* Not used at present */ 417 static void store_le(u32 x, unsigned char *buf) 418 { 419 buf[0] = (unsigned char) x; 420 buf[1] = (unsigned char) (x >> 8); 421 buf[2] = (unsigned char) (x >> 16); 422 buf[3] = (unsigned char) (x >> 24); 423 } 424 #endif 425 426 static void store_be(u32 x, unsigned char *buf) 427 { 428 buf[0] = (unsigned char) (x >> 24); 429 buf[1] = (unsigned char) (x >> 16); 430 buf[2] = (unsigned char) (x >> 8); 431 buf[3] = (unsigned char) x; 432 } 433 434 /* 435 * This checks that CRC(buf + CRC(buf)) = 0, and that 436 * CRC commutes with bit-reversal. This has the side effect 437 * of bytewise bit-reversing the input buffer, and returns 438 * the CRC of the reversed buffer. 439 */ 440 static u32 test_step(u32 init, unsigned char *buf, size_t len) 441 { 442 u32 crc1, crc2; 443 size_t i; 444 445 crc1 = crc32_be(init, buf, len); 446 store_be(crc1, buf + len); 447 crc2 = crc32_be(init, buf, len + 4); 448 if (crc2) 449 printf("\nCRC cancellation fail: 0x%08x should be 0\n", 450 crc2); 451 452 for (i = 0; i <= len + 4; i++) { 453 crc2 = crc32_be(init, buf, i); 454 crc2 = crc32_be(crc2, buf + i, len + 4 - i); 455 if (crc2) 456 printf("\nCRC split fail: 0x%08x\n", crc2); 457 } 458 459 /* Now swap it around for the other test */ 460 461 bytereverse(buf, len + 4); 462 init = bitrev32(init); 463 crc2 = bitrev32(crc1); 464 if (crc1 != bitrev32(crc2)) 465 printf("\nBit reversal fail: 0x%08x -> 0x%08x -> 0x%08x\n", 466 crc1, crc2, bitrev32(crc2)); 467 crc1 = crc32_le(init, buf, len); 468 if (crc1 != crc2) 469 printf("\nCRC endianness fail: 0x%08x != 0x%08x\n", crc1, 470 crc2); 471 crc2 = crc32_le(init, buf, len + 4); 472 if (crc2) 473 printf("\nCRC cancellation fail: 0x%08x should be 0\n", 474 crc2); 475 476 for (i = 0; i <= len + 4; i++) { 477 crc2 = crc32_le(init, buf, i); 478 crc2 = crc32_le(crc2, buf + i, len + 4 - i); 479 if (crc2) 480 printf("\nCRC split fail: 0x%08x\n", crc2); 481 } 482 483 return crc1; 484 } 485 486 #define SIZE 64 487 #define INIT1 0 488 #define INIT2 0 489 490 int main(void) 491 { 492 unsigned char buf1[SIZE + 4]; 493 unsigned char buf2[SIZE + 4]; 494 unsigned char buf3[SIZE + 4]; 495 int i, j; 496 u32 crc1, crc2, crc3; 497 498 for (i = 0; i <= SIZE; i++) { 499 printf("\rTesting length %d...", i); 500 fflush(stdout); 501 random_garbage(buf1, i); 502 random_garbage(buf2, i); 503 for (j = 0; j < i; j++) 504 buf3[j] = buf1[j] ^ buf2[j]; 505 506 crc1 = test_step(INIT1, buf1, i); 507 crc2 = test_step(INIT2, buf2, i); 508 /* Now check that CRC(buf1 ^ buf2) = CRC(buf1) ^ CRC(buf2) */ 509 crc3 = test_step(INIT1 ^ INIT2, buf3, i); 510 if (crc3 != (crc1 ^ crc2)) 511 printf("CRC XOR fail: 0x%08x != 0x%08x ^ 0x%08x\n", 512 crc3, crc1, crc2); 513 } 514 printf("\nAll test complete. No failures expected.\n"); 515 return 0; 516 } 517 518 #endif /* UNITTEST */ 519