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 #define __UBOOT__ 24 #ifndef __UBOOT__ 25 #include <linux/crc32.h> 26 #include <linux/kernel.h> 27 #include <linux/module.h> 28 #include <linux/compiler.h> 29 #endif 30 #include <linux/types.h> 31 32 #include <asm/byteorder.h> 33 34 #ifndef __UBOOT__ 35 #include <linux/slab.h> 36 #include <linux/init.h> 37 #include <asm/atomic.h> 38 #endif 39 #include "crc32defs.h" 40 #define CRC_LE_BITS 8 41 42 #if CRC_LE_BITS == 8 43 #define tole(x) cpu_to_le32(x) 44 #define tobe(x) cpu_to_be32(x) 45 #else 46 #define tole(x) (x) 47 #define tobe(x) (x) 48 #endif 49 #include "crc32table.h" 50 #ifndef __UBOOT__ 51 MODULE_AUTHOR("Matt Domsch <Matt_Domsch@dell.com>"); 52 MODULE_DESCRIPTION("Ethernet CRC32 calculations"); 53 MODULE_LICENSE("GPL"); 54 #endif 55 /** 56 * crc32_le() - Calculate bitwise little-endian Ethernet AUTODIN II CRC32 57 * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for 58 * other uses, or the previous crc32 value if computing incrementally. 59 * @p: pointer to buffer over which CRC is run 60 * @len: length of buffer @p 61 */ 62 u32 crc32_le(u32 crc, unsigned char const *p, size_t len); 63 64 #if CRC_LE_BITS == 1 65 /* 66 * In fact, the table-based code will work in this case, but it can be 67 * simplified by inlining the table in ?: form. 68 */ 69 70 u32 crc32_le(u32 crc, unsigned char const *p, size_t len) 71 { 72 int i; 73 while (len--) { 74 crc ^= *p++; 75 for (i = 0; i < 8; i++) 76 crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_LE : 0); 77 } 78 return crc; 79 } 80 #else /* Table-based approach */ 81 82 u32 crc32_le(u32 crc, unsigned char const *p, size_t len) 83 { 84 # if CRC_LE_BITS == 8 85 const u32 *b =(u32 *)p; 86 const u32 *tab = crc32table_le; 87 88 # ifdef __LITTLE_ENDIAN 89 # define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8) 90 # else 91 # define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8) 92 # endif 93 /* printf("Crc32_le crc=%x\n",crc); */ 94 crc = __cpu_to_le32(crc); 95 /* Align it */ 96 if((((long)b)&3 && len)){ 97 do { 98 u8 *p = (u8 *)b; 99 DO_CRC(*p++); 100 b = (void *)p; 101 } while ((--len) && ((long)b)&3 ); 102 } 103 if((len >= 4)){ 104 /* load data 32 bits wide, xor data 32 bits wide. */ 105 size_t save_len = len & 3; 106 len = len >> 2; 107 --b; /* use pre increment below(*++b) for speed */ 108 do { 109 crc ^= *++b; 110 DO_CRC(0); 111 DO_CRC(0); 112 DO_CRC(0); 113 DO_CRC(0); 114 } while (--len); 115 b++; /* point to next byte(s) */ 116 len = save_len; 117 } 118 /* And the last few bytes */ 119 if(len){ 120 do { 121 u8 *p = (u8 *)b; 122 DO_CRC(*p++); 123 b = (void *)p; 124 } while (--len); 125 } 126 127 return __le32_to_cpu(crc); 128 #undef ENDIAN_SHIFT 129 #undef DO_CRC 130 131 # elif CRC_LE_BITS == 4 132 while (len--) { 133 crc ^= *p++; 134 crc = (crc >> 4) ^ crc32table_le[crc & 15]; 135 crc = (crc >> 4) ^ crc32table_le[crc & 15]; 136 } 137 return crc; 138 # elif CRC_LE_BITS == 2 139 while (len--) { 140 crc ^= *p++; 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 crc = (crc >> 2) ^ crc32table_le[crc & 3]; 145 } 146 return crc; 147 # endif 148 } 149 #endif 150 #ifndef __UBOOT__ 151 /** 152 * crc32_be() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32 153 * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for 154 * other uses, or the previous crc32 value if computing incrementally. 155 * @p: pointer to buffer over which CRC is run 156 * @len: length of buffer @p 157 */ 158 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len); 159 160 #if CRC_BE_BITS == 1 161 /* 162 * In fact, the table-based code will work in this case, but it can be 163 * simplified by inlining the table in ?: form. 164 */ 165 166 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len) 167 { 168 int i; 169 while (len--) { 170 crc ^= *p++ << 24; 171 for (i = 0; i < 8; i++) 172 crc = 173 (crc << 1) ^ ((crc & 0x80000000) ? CRCPOLY_BE : 174 0); 175 } 176 return crc; 177 } 178 179 #else /* Table-based approach */ 180 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len) 181 { 182 # if CRC_BE_BITS == 8 183 const u32 *b =(u32 *)p; 184 const u32 *tab = crc32table_be; 185 186 # ifdef __LITTLE_ENDIAN 187 # define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8) 188 # else 189 # define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8) 190 # endif 191 192 crc = __cpu_to_be32(crc); 193 /* Align it */ 194 if(unlikely(((long)b)&3 && len)){ 195 do { 196 u8 *p = (u8 *)b; 197 DO_CRC(*p++); 198 b = (u32 *)p; 199 } while ((--len) && ((long)b)&3 ); 200 } 201 if(likely(len >= 4)){ 202 /* load data 32 bits wide, xor data 32 bits wide. */ 203 size_t save_len = len & 3; 204 len = len >> 2; 205 --b; /* use pre increment below(*++b) for speed */ 206 do { 207 crc ^= *++b; 208 DO_CRC(0); 209 DO_CRC(0); 210 DO_CRC(0); 211 DO_CRC(0); 212 } while (--len); 213 b++; /* point to next byte(s) */ 214 len = save_len; 215 } 216 /* And the last few bytes */ 217 if(len){ 218 do { 219 u8 *p = (u8 *)b; 220 DO_CRC(*p++); 221 b = (void *)p; 222 } while (--len); 223 } 224 return __be32_to_cpu(crc); 225 #undef ENDIAN_SHIFT 226 #undef DO_CRC 227 228 # elif CRC_BE_BITS == 4 229 while (len--) { 230 crc ^= *p++ << 24; 231 crc = (crc << 4) ^ crc32table_be[crc >> 28]; 232 crc = (crc << 4) ^ crc32table_be[crc >> 28]; 233 } 234 return crc; 235 # elif CRC_BE_BITS == 2 236 while (len--) { 237 crc ^= *p++ << 24; 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 crc = (crc << 2) ^ crc32table_be[crc >> 30]; 242 } 243 return crc; 244 # endif 245 } 246 #endif 247 248 EXPORT_SYMBOL(crc32_le); 249 EXPORT_SYMBOL(crc32_be); 250 #endif 251 /* 252 * A brief CRC tutorial. 253 * 254 * A CRC is a long-division remainder. You add the CRC to the message, 255 * and the whole thing (message+CRC) is a multiple of the given 256 * CRC polynomial. To check the CRC, you can either check that the 257 * CRC matches the recomputed value, *or* you can check that the 258 * remainder computed on the message+CRC is 0. This latter approach 259 * is used by a lot of hardware implementations, and is why so many 260 * protocols put the end-of-frame flag after the CRC. 261 * 262 * It's actually the same long division you learned in school, except that 263 * - We're working in binary, so the digits are only 0 and 1, and 264 * - When dividing polynomials, there are no carries. Rather than add and 265 * subtract, we just xor. Thus, we tend to get a bit sloppy about 266 * the difference between adding and subtracting. 267 * 268 * A 32-bit CRC polynomial is actually 33 bits long. But since it's 269 * 33 bits long, bit 32 is always going to be set, so usually the CRC 270 * is written in hex with the most significant bit omitted. (If you're 271 * familiar with the IEEE 754 floating-point format, it's the same idea.) 272 * 273 * Note that a CRC is computed over a string of *bits*, so you have 274 * to decide on the endianness of the bits within each byte. To get 275 * the best error-detecting properties, this should correspond to the 276 * order they're actually sent. For example, standard RS-232 serial is 277 * little-endian; the most significant bit (sometimes used for parity) 278 * is sent last. And when appending a CRC word to a message, you should 279 * do it in the right order, matching the endianness. 280 * 281 * Just like with ordinary division, the remainder is always smaller than 282 * the divisor (the CRC polynomial) you're dividing by. Each step of the 283 * division, you take one more digit (bit) of the dividend and append it 284 * to the current remainder. Then you figure out the appropriate multiple 285 * of the divisor to subtract to being the remainder back into range. 286 * In binary, it's easy - it has to be either 0 or 1, and to make the 287 * XOR cancel, it's just a copy of bit 32 of the remainder. 288 * 289 * When computing a CRC, we don't care about the quotient, so we can 290 * throw the quotient bit away, but subtract the appropriate multiple of 291 * the polynomial from the remainder and we're back to where we started, 292 * ready to process the next bit. 293 * 294 * A big-endian CRC written this way would be coded like: 295 * for (i = 0; i < input_bits; i++) { 296 * multiple = remainder & 0x80000000 ? CRCPOLY : 0; 297 * remainder = (remainder << 1 | next_input_bit()) ^ multiple; 298 * } 299 * Notice how, to get at bit 32 of the shifted remainder, we look 300 * at bit 31 of the remainder *before* shifting it. 301 * 302 * But also notice how the next_input_bit() bits we're shifting into 303 * the remainder don't actually affect any decision-making until 304 * 32 bits later. Thus, the first 32 cycles of this are pretty boring. 305 * Also, to add the CRC to a message, we need a 32-bit-long hole for it at 306 * the end, so we have to add 32 extra cycles shifting in zeros at the 307 * end of every message, 308 * 309 * So the standard trick is to rearrage merging in the next_input_bit() 310 * until the moment it's needed. Then the first 32 cycles can be precomputed, 311 * and merging in the final 32 zero bits to make room for the CRC can be 312 * skipped entirely. 313 * This changes the code to: 314 * for (i = 0; i < input_bits; i++) { 315 * remainder ^= next_input_bit() << 31; 316 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; 317 * remainder = (remainder << 1) ^ multiple; 318 * } 319 * With this optimization, the little-endian code is simpler: 320 * for (i = 0; i < input_bits; i++) { 321 * remainder ^= next_input_bit(); 322 * multiple = (remainder & 1) ? CRCPOLY : 0; 323 * remainder = (remainder >> 1) ^ multiple; 324 * } 325 * 326 * Note that the other details of endianness have been hidden in CRCPOLY 327 * (which must be bit-reversed) and next_input_bit(). 328 * 329 * However, as long as next_input_bit is returning the bits in a sensible 330 * order, we can actually do the merging 8 or more bits at a time rather 331 * than one bit at a time: 332 * for (i = 0; i < input_bytes; i++) { 333 * remainder ^= next_input_byte() << 24; 334 * for (j = 0; j < 8; j++) { 335 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; 336 * remainder = (remainder << 1) ^ multiple; 337 * } 338 * } 339 * Or in little-endian: 340 * for (i = 0; i < input_bytes; i++) { 341 * remainder ^= next_input_byte(); 342 * for (j = 0; j < 8; j++) { 343 * multiple = (remainder & 1) ? CRCPOLY : 0; 344 * remainder = (remainder << 1) ^ multiple; 345 * } 346 * } 347 * If the input is a multiple of 32 bits, you can even XOR in a 32-bit 348 * word at a time and increase the inner loop count to 32. 349 * 350 * You can also mix and match the two loop styles, for example doing the 351 * bulk of a message byte-at-a-time and adding bit-at-a-time processing 352 * for any fractional bytes at the end. 353 * 354 * The only remaining optimization is to the byte-at-a-time table method. 355 * Here, rather than just shifting one bit of the remainder to decide 356 * in the correct multiple to subtract, we can shift a byte at a time. 357 * This produces a 40-bit (rather than a 33-bit) intermediate remainder, 358 * but again the multiple of the polynomial to subtract depends only on 359 * the high bits, the high 8 bits in this case. 360 * 361 * The multile we need in that case is the low 32 bits of a 40-bit 362 * value whose high 8 bits are given, and which is a multiple of the 363 * generator polynomial. This is simply the CRC-32 of the given 364 * one-byte message. 365 * 366 * Two more details: normally, appending zero bits to a message which 367 * is already a multiple of a polynomial produces a larger multiple of that 368 * polynomial. To enable a CRC to detect this condition, it's common to 369 * invert the CRC before appending it. This makes the remainder of the 370 * message+crc come out not as zero, but some fixed non-zero value. 371 * 372 * The same problem applies to zero bits prepended to the message, and 373 * a similar solution is used. Instead of starting with a remainder of 374 * 0, an initial remainder of all ones is used. As long as you start 375 * the same way on decoding, it doesn't make a difference. 376 */ 377 378 #ifdef UNITTEST 379 380 #include <stdlib.h> 381 #include <stdio.h> 382 383 #ifndef __UBOOT__ 384 static void 385 buf_dump(char const *prefix, unsigned char const *buf, size_t len) 386 { 387 fputs(prefix, stdout); 388 while (len--) 389 printf(" %02x", *buf++); 390 putchar('\n'); 391 392 } 393 #endif 394 395 static void bytereverse(unsigned char *buf, size_t len) 396 { 397 while (len--) { 398 unsigned char x = bitrev8(*buf); 399 *buf++ = x; 400 } 401 } 402 403 static void random_garbage(unsigned char *buf, size_t len) 404 { 405 while (len--) 406 *buf++ = (unsigned char) random(); 407 } 408 409 #ifndef __UBOOT__ 410 static void store_le(u32 x, unsigned char *buf) 411 { 412 buf[0] = (unsigned char) x; 413 buf[1] = (unsigned char) (x >> 8); 414 buf[2] = (unsigned char) (x >> 16); 415 buf[3] = (unsigned char) (x >> 24); 416 } 417 #endif 418 419 static void store_be(u32 x, unsigned char *buf) 420 { 421 buf[0] = (unsigned char) (x >> 24); 422 buf[1] = (unsigned char) (x >> 16); 423 buf[2] = (unsigned char) (x >> 8); 424 buf[3] = (unsigned char) x; 425 } 426 427 /* 428 * This checks that CRC(buf + CRC(buf)) = 0, and that 429 * CRC commutes with bit-reversal. This has the side effect 430 * of bytewise bit-reversing the input buffer, and returns 431 * the CRC of the reversed buffer. 432 */ 433 static u32 test_step(u32 init, unsigned char *buf, size_t len) 434 { 435 u32 crc1, crc2; 436 size_t i; 437 438 crc1 = crc32_be(init, buf, len); 439 store_be(crc1, buf + len); 440 crc2 = crc32_be(init, buf, len + 4); 441 if (crc2) 442 printf("\nCRC cancellation fail: 0x%08x should be 0\n", 443 crc2); 444 445 for (i = 0; i <= len + 4; i++) { 446 crc2 = crc32_be(init, buf, i); 447 crc2 = crc32_be(crc2, buf + i, len + 4 - i); 448 if (crc2) 449 printf("\nCRC split fail: 0x%08x\n", crc2); 450 } 451 452 /* Now swap it around for the other test */ 453 454 bytereverse(buf, len + 4); 455 init = bitrev32(init); 456 crc2 = bitrev32(crc1); 457 if (crc1 != bitrev32(crc2)) 458 printf("\nBit reversal fail: 0x%08x -> 0x%08x -> 0x%08x\n", 459 crc1, crc2, bitrev32(crc2)); 460 crc1 = crc32_le(init, buf, len); 461 if (crc1 != crc2) 462 printf("\nCRC endianness fail: 0x%08x != 0x%08x\n", crc1, 463 crc2); 464 crc2 = crc32_le(init, buf, len + 4); 465 if (crc2) 466 printf("\nCRC cancellation fail: 0x%08x should be 0\n", 467 crc2); 468 469 for (i = 0; i <= len + 4; i++) { 470 crc2 = crc32_le(init, buf, i); 471 crc2 = crc32_le(crc2, buf + i, len + 4 - i); 472 if (crc2) 473 printf("\nCRC split fail: 0x%08x\n", crc2); 474 } 475 476 return crc1; 477 } 478 479 #define SIZE 64 480 #define INIT1 0 481 #define INIT2 0 482 483 int main(void) 484 { 485 unsigned char buf1[SIZE + 4]; 486 unsigned char buf2[SIZE + 4]; 487 unsigned char buf3[SIZE + 4]; 488 int i, j; 489 u32 crc1, crc2, crc3; 490 491 for (i = 0; i <= SIZE; i++) { 492 printf("\rTesting length %d...", i); 493 fflush(stdout); 494 random_garbage(buf1, i); 495 random_garbage(buf2, i); 496 for (j = 0; j < i; j++) 497 buf3[j] = buf1[j] ^ buf2[j]; 498 499 crc1 = test_step(INIT1, buf1, i); 500 crc2 = test_step(INIT2, buf2, i); 501 /* Now check that CRC(buf1 ^ buf2) = CRC(buf1) ^ CRC(buf2) */ 502 crc3 = test_step(INIT1 ^ INIT2, buf3, i); 503 if (crc3 != (crc1 ^ crc2)) 504 printf("CRC XOR fail: 0x%08x != 0x%08x ^ 0x%08x\n", 505 crc3, crc1, crc2); 506 } 507 printf("\nAll test complete. No failures expected.\n"); 508 return 0; 509 } 510 511 #endif /* UNITTEST */ 512