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 #if CRC_LE_BITS == 8 42 #define tole(x) cpu_to_le32(x) 43 #define tobe(x) cpu_to_be32(x) 44 #else 45 #define tole(x) (x) 46 #define tobe(x) (x) 47 #endif 48 #include "crc32table.h" 49 #ifdef UBI_LINUX 50 MODULE_AUTHOR("Matt Domsch <Matt_Domsch@dell.com>"); 51 MODULE_DESCRIPTION("Ethernet CRC32 calculations"); 52 MODULE_LICENSE("GPL"); 53 #endif 54 /** 55 * crc32_le() - Calculate bitwise little-endian Ethernet AUTODIN II CRC32 56 * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for 57 * other uses, or the previous crc32 value if computing incrementally. 58 * @p: pointer to buffer over which CRC is run 59 * @len: length of buffer @p 60 */ 61 u32 crc32_le(u32 crc, unsigned char const *p, size_t len); 62 63 #if CRC_LE_BITS == 1 64 /* 65 * In fact, the table-based code will work in this case, but it can be 66 * simplified by inlining the table in ?: form. 67 */ 68 69 u32 crc32_le(u32 crc, unsigned char const *p, size_t len) 70 { 71 int i; 72 while (len--) { 73 crc ^= *p++; 74 for (i = 0; i < 8; i++) 75 crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_LE : 0); 76 } 77 return crc; 78 } 79 #else /* Table-based approach */ 80 81 u32 crc32_le(u32 crc, unsigned char const *p, size_t len) 82 { 83 # if CRC_LE_BITS == 8 84 const u32 *b =(u32 *)p; 85 const u32 *tab = crc32table_le; 86 87 # ifdef __LITTLE_ENDIAN 88 # define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8) 89 # else 90 # define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8) 91 # endif 92 /* printf("Crc32_le crc=%x\n",crc); */ 93 crc = __cpu_to_le32(crc); 94 /* Align it */ 95 if((((long)b)&3 && len)){ 96 do { 97 u8 *p = (u8 *)b; 98 DO_CRC(*p++); 99 b = (void *)p; 100 } while ((--len) && ((long)b)&3 ); 101 } 102 if((len >= 4)){ 103 /* load data 32 bits wide, xor data 32 bits wide. */ 104 size_t save_len = len & 3; 105 len = len >> 2; 106 --b; /* use pre increment below(*++b) for speed */ 107 do { 108 crc ^= *++b; 109 DO_CRC(0); 110 DO_CRC(0); 111 DO_CRC(0); 112 DO_CRC(0); 113 } while (--len); 114 b++; /* point to next byte(s) */ 115 len = save_len; 116 } 117 /* And the last few bytes */ 118 if(len){ 119 do { 120 u8 *p = (u8 *)b; 121 DO_CRC(*p++); 122 b = (void *)p; 123 } while (--len); 124 } 125 126 return __le32_to_cpu(crc); 127 #undef ENDIAN_SHIFT 128 #undef DO_CRC 129 130 # elif CRC_LE_BITS == 4 131 while (len--) { 132 crc ^= *p++; 133 crc = (crc >> 4) ^ crc32table_le[crc & 15]; 134 crc = (crc >> 4) ^ crc32table_le[crc & 15]; 135 } 136 return crc; 137 # elif CRC_LE_BITS == 2 138 while (len--) { 139 crc ^= *p++; 140 crc = (crc >> 2) ^ crc32table_le[crc & 3]; 141 crc = (crc >> 2) ^ crc32table_le[crc & 3]; 142 crc = (crc >> 2) ^ crc32table_le[crc & 3]; 143 crc = (crc >> 2) ^ crc32table_le[crc & 3]; 144 } 145 return crc; 146 # endif 147 } 148 #endif 149 #ifdef UBI_LINUX 150 /** 151 * crc32_be() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32 152 * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for 153 * other uses, or the previous crc32 value if computing incrementally. 154 * @p: pointer to buffer over which CRC is run 155 * @len: length of buffer @p 156 */ 157 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len); 158 159 #if CRC_BE_BITS == 1 160 /* 161 * In fact, the table-based code will work in this case, but it can be 162 * simplified by inlining the table in ?: form. 163 */ 164 165 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len) 166 { 167 int i; 168 while (len--) { 169 crc ^= *p++ << 24; 170 for (i = 0; i < 8; i++) 171 crc = 172 (crc << 1) ^ ((crc & 0x80000000) ? CRCPOLY_BE : 173 0); 174 } 175 return crc; 176 } 177 178 #else /* Table-based approach */ 179 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len) 180 { 181 # if CRC_BE_BITS == 8 182 const u32 *b =(u32 *)p; 183 const u32 *tab = crc32table_be; 184 185 # ifdef __LITTLE_ENDIAN 186 # define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8) 187 # else 188 # define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8) 189 # endif 190 191 crc = __cpu_to_be32(crc); 192 /* Align it */ 193 if(unlikely(((long)b)&3 && len)){ 194 do { 195 u8 *p = (u8 *)b; 196 DO_CRC(*p++); 197 b = (u32 *)p; 198 } while ((--len) && ((long)b)&3 ); 199 } 200 if(likely(len >= 4)){ 201 /* load data 32 bits wide, xor data 32 bits wide. */ 202 size_t save_len = len & 3; 203 len = len >> 2; 204 --b; /* use pre increment below(*++b) for speed */ 205 do { 206 crc ^= *++b; 207 DO_CRC(0); 208 DO_CRC(0); 209 DO_CRC(0); 210 DO_CRC(0); 211 } while (--len); 212 b++; /* point to next byte(s) */ 213 len = save_len; 214 } 215 /* And the last few bytes */ 216 if(len){ 217 do { 218 u8 *p = (u8 *)b; 219 DO_CRC(*p++); 220 b = (void *)p; 221 } while (--len); 222 } 223 return __be32_to_cpu(crc); 224 #undef ENDIAN_SHIFT 225 #undef DO_CRC 226 227 # elif CRC_BE_BITS == 4 228 while (len--) { 229 crc ^= *p++ << 24; 230 crc = (crc << 4) ^ crc32table_be[crc >> 28]; 231 crc = (crc << 4) ^ crc32table_be[crc >> 28]; 232 } 233 return crc; 234 # elif CRC_BE_BITS == 2 235 while (len--) { 236 crc ^= *p++ << 24; 237 crc = (crc << 2) ^ crc32table_be[crc >> 30]; 238 crc = (crc << 2) ^ crc32table_be[crc >> 30]; 239 crc = (crc << 2) ^ crc32table_be[crc >> 30]; 240 crc = (crc << 2) ^ crc32table_be[crc >> 30]; 241 } 242 return crc; 243 # endif 244 } 245 #endif 246 247 EXPORT_SYMBOL(crc32_le); 248 EXPORT_SYMBOL(crc32_be); 249 #endif 250 /* 251 * A brief CRC tutorial. 252 * 253 * A CRC is a long-division remainder. You add the CRC to the message, 254 * and the whole thing (message+CRC) is a multiple of the given 255 * CRC polynomial. To check the CRC, you can either check that the 256 * CRC matches the recomputed value, *or* you can check that the 257 * remainder computed on the message+CRC is 0. This latter approach 258 * is used by a lot of hardware implementations, and is why so many 259 * protocols put the end-of-frame flag after the CRC. 260 * 261 * It's actually the same long division you learned in school, except that 262 * - We're working in binary, so the digits are only 0 and 1, and 263 * - When dividing polynomials, there are no carries. Rather than add and 264 * subtract, we just xor. Thus, we tend to get a bit sloppy about 265 * the difference between adding and subtracting. 266 * 267 * A 32-bit CRC polynomial is actually 33 bits long. But since it's 268 * 33 bits long, bit 32 is always going to be set, so usually the CRC 269 * is written in hex with the most significant bit omitted. (If you're 270 * familiar with the IEEE 754 floating-point format, it's the same idea.) 271 * 272 * Note that a CRC is computed over a string of *bits*, so you have 273 * to decide on the endianness of the bits within each byte. To get 274 * the best error-detecting properties, this should correspond to the 275 * order they're actually sent. For example, standard RS-232 serial is 276 * little-endian; the most significant bit (sometimes used for parity) 277 * is sent last. And when appending a CRC word to a message, you should 278 * do it in the right order, matching the endianness. 279 * 280 * Just like with ordinary division, the remainder is always smaller than 281 * the divisor (the CRC polynomial) you're dividing by. Each step of the 282 * division, you take one more digit (bit) of the dividend and append it 283 * to the current remainder. Then you figure out the appropriate multiple 284 * of the divisor to subtract to being the remainder back into range. 285 * In binary, it's easy - it has to be either 0 or 1, and to make the 286 * XOR cancel, it's just a copy of bit 32 of the remainder. 287 * 288 * When computing a CRC, we don't care about the quotient, so we can 289 * throw the quotient bit away, but subtract the appropriate multiple of 290 * the polynomial from the remainder and we're back to where we started, 291 * ready to process the next bit. 292 * 293 * A big-endian CRC written this way would be coded like: 294 * for (i = 0; i < input_bits; i++) { 295 * multiple = remainder & 0x80000000 ? CRCPOLY : 0; 296 * remainder = (remainder << 1 | next_input_bit()) ^ multiple; 297 * } 298 * Notice how, to get at bit 32 of the shifted remainder, we look 299 * at bit 31 of the remainder *before* shifting it. 300 * 301 * But also notice how the next_input_bit() bits we're shifting into 302 * the remainder don't actually affect any decision-making until 303 * 32 bits later. Thus, the first 32 cycles of this are pretty boring. 304 * Also, to add the CRC to a message, we need a 32-bit-long hole for it at 305 * the end, so we have to add 32 extra cycles shifting in zeros at the 306 * end of every message, 307 * 308 * So the standard trick is to rearrage merging in the next_input_bit() 309 * until the moment it's needed. Then the first 32 cycles can be precomputed, 310 * and merging in the final 32 zero bits to make room for the CRC can be 311 * skipped entirely. 312 * This changes the code to: 313 * for (i = 0; i < input_bits; i++) { 314 * remainder ^= next_input_bit() << 31; 315 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; 316 * remainder = (remainder << 1) ^ multiple; 317 * } 318 * With this optimization, the little-endian code is simpler: 319 * for (i = 0; i < input_bits; i++) { 320 * remainder ^= next_input_bit(); 321 * multiple = (remainder & 1) ? CRCPOLY : 0; 322 * remainder = (remainder >> 1) ^ multiple; 323 * } 324 * 325 * Note that the other details of endianness have been hidden in CRCPOLY 326 * (which must be bit-reversed) and next_input_bit(). 327 * 328 * However, as long as next_input_bit is returning the bits in a sensible 329 * order, we can actually do the merging 8 or more bits at a time rather 330 * than one bit at a time: 331 * for (i = 0; i < input_bytes; i++) { 332 * remainder ^= next_input_byte() << 24; 333 * for (j = 0; j < 8; j++) { 334 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; 335 * remainder = (remainder << 1) ^ multiple; 336 * } 337 * } 338 * Or in little-endian: 339 * for (i = 0; i < input_bytes; i++) { 340 * remainder ^= next_input_byte(); 341 * for (j = 0; j < 8; j++) { 342 * multiple = (remainder & 1) ? CRCPOLY : 0; 343 * remainder = (remainder << 1) ^ multiple; 344 * } 345 * } 346 * If the input is a multiple of 32 bits, you can even XOR in a 32-bit 347 * word at a time and increase the inner loop count to 32. 348 * 349 * You can also mix and match the two loop styles, for example doing the 350 * bulk of a message byte-at-a-time and adding bit-at-a-time processing 351 * for any fractional bytes at the end. 352 * 353 * The only remaining optimization is to the byte-at-a-time table method. 354 * Here, rather than just shifting one bit of the remainder to decide 355 * in the correct multiple to subtract, we can shift a byte at a time. 356 * This produces a 40-bit (rather than a 33-bit) intermediate remainder, 357 * but again the multiple of the polynomial to subtract depends only on 358 * the high bits, the high 8 bits in this case. 359 * 360 * The multile we need in that case is the low 32 bits of a 40-bit 361 * value whose high 8 bits are given, and which is a multiple of the 362 * generator polynomial. This is simply the CRC-32 of the given 363 * one-byte message. 364 * 365 * Two more details: normally, appending zero bits to a message which 366 * is already a multiple of a polynomial produces a larger multiple of that 367 * polynomial. To enable a CRC to detect this condition, it's common to 368 * invert the CRC before appending it. This makes the remainder of the 369 * message+crc come out not as zero, but some fixed non-zero value. 370 * 371 * The same problem applies to zero bits prepended to the message, and 372 * a similar solution is used. Instead of starting with a remainder of 373 * 0, an initial remainder of all ones is used. As long as you start 374 * the same way on decoding, it doesn't make a difference. 375 */ 376 377 #ifdef UNITTEST 378 379 #include <stdlib.h> 380 #include <stdio.h> 381 382 #ifdef UBI_LINUX /*Not used at present */ 383 static void 384 buf_dump(char const *prefix, unsigned char const *buf, size_t len) 385 { 386 fputs(prefix, stdout); 387 while (len--) 388 printf(" %02x", *buf++); 389 putchar('\n'); 390 391 } 392 #endif 393 394 static void bytereverse(unsigned char *buf, size_t len) 395 { 396 while (len--) { 397 unsigned char x = bitrev8(*buf); 398 *buf++ = x; 399 } 400 } 401 402 static void random_garbage(unsigned char *buf, size_t len) 403 { 404 while (len--) 405 *buf++ = (unsigned char) random(); 406 } 407 408 #ifdef UBI_LINUX /* Not used at present */ 409 static void store_le(u32 x, unsigned char *buf) 410 { 411 buf[0] = (unsigned char) x; 412 buf[1] = (unsigned char) (x >> 8); 413 buf[2] = (unsigned char) (x >> 16); 414 buf[3] = (unsigned char) (x >> 24); 415 } 416 #endif 417 418 static void store_be(u32 x, unsigned char *buf) 419 { 420 buf[0] = (unsigned char) (x >> 24); 421 buf[1] = (unsigned char) (x >> 16); 422 buf[2] = (unsigned char) (x >> 8); 423 buf[3] = (unsigned char) x; 424 } 425 426 /* 427 * This checks that CRC(buf + CRC(buf)) = 0, and that 428 * CRC commutes with bit-reversal. This has the side effect 429 * of bytewise bit-reversing the input buffer, and returns 430 * the CRC of the reversed buffer. 431 */ 432 static u32 test_step(u32 init, unsigned char *buf, size_t len) 433 { 434 u32 crc1, crc2; 435 size_t i; 436 437 crc1 = crc32_be(init, buf, len); 438 store_be(crc1, buf + len); 439 crc2 = crc32_be(init, buf, len + 4); 440 if (crc2) 441 printf("\nCRC cancellation fail: 0x%08x should be 0\n", 442 crc2); 443 444 for (i = 0; i <= len + 4; i++) { 445 crc2 = crc32_be(init, buf, i); 446 crc2 = crc32_be(crc2, buf + i, len + 4 - i); 447 if (crc2) 448 printf("\nCRC split fail: 0x%08x\n", crc2); 449 } 450 451 /* Now swap it around for the other test */ 452 453 bytereverse(buf, len + 4); 454 init = bitrev32(init); 455 crc2 = bitrev32(crc1); 456 if (crc1 != bitrev32(crc2)) 457 printf("\nBit reversal fail: 0x%08x -> 0x%08x -> 0x%08x\n", 458 crc1, crc2, bitrev32(crc2)); 459 crc1 = crc32_le(init, buf, len); 460 if (crc1 != crc2) 461 printf("\nCRC endianness fail: 0x%08x != 0x%08x\n", crc1, 462 crc2); 463 crc2 = crc32_le(init, buf, len + 4); 464 if (crc2) 465 printf("\nCRC cancellation fail: 0x%08x should be 0\n", 466 crc2); 467 468 for (i = 0; i <= len + 4; i++) { 469 crc2 = crc32_le(init, buf, i); 470 crc2 = crc32_le(crc2, buf + i, len + 4 - i); 471 if (crc2) 472 printf("\nCRC split fail: 0x%08x\n", crc2); 473 } 474 475 return crc1; 476 } 477 478 #define SIZE 64 479 #define INIT1 0 480 #define INIT2 0 481 482 int main(void) 483 { 484 unsigned char buf1[SIZE + 4]; 485 unsigned char buf2[SIZE + 4]; 486 unsigned char buf3[SIZE + 4]; 487 int i, j; 488 u32 crc1, crc2, crc3; 489 490 for (i = 0; i <= SIZE; i++) { 491 printf("\rTesting length %d...", i); 492 fflush(stdout); 493 random_garbage(buf1, i); 494 random_garbage(buf2, i); 495 for (j = 0; j < i; j++) 496 buf3[j] = buf1[j] ^ buf2[j]; 497 498 crc1 = test_step(INIT1, buf1, i); 499 crc2 = test_step(INIT2, buf2, i); 500 /* Now check that CRC(buf1 ^ buf2) = CRC(buf1) ^ CRC(buf2) */ 501 crc3 = test_step(INIT1 ^ INIT2, buf3, i); 502 if (crc3 != (crc1 ^ crc2)) 503 printf("CRC XOR fail: 0x%08x != 0x%08x ^ 0x%08x\n", 504 crc3, crc1, crc2); 505 } 506 printf("\nAll test complete. No failures expected.\n"); 507 return 0; 508 } 509 510 #endif /* UNITTEST */ 511