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