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
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.
19 * This source code is licensed under the GNU General Public License,
20 * Version 2. See the file COPYING for more details.
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"
33 #define tole(x) __constant_cpu_to_le32(x)
34 #define tobe(x) __constant_cpu_to_be32(x)
39 #include "crc32table.h"
41 MODULE_AUTHOR("Matt Domsch <Matt_Domsch@dell.com>");
42 MODULE_DESCRIPTION("Ethernet CRC32 calculations");
43 MODULE_LICENSE("GPL");
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
52 u32 __attribute_pure__ crc32_le(u32 crc, unsigned char const *p, size_t len);
56 * In fact, the table-based code will work in this case, but it can be
57 * simplified by inlining the table in ?: form.
60 u32 __attribute_pure__ crc32_le(u32 crc, unsigned char const *p, size_t len)
65 for (i = 0; i < 8; i++)
66 crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_LE : 0);
70 #else /* Table-based approach */
72 u32 __attribute_pure__ crc32_le(u32 crc, unsigned char const *p, size_t len)
75 const u32 *b =(u32 *)p;
76 const u32 *tab = crc32table_le;
78 # ifdef __LITTLE_ENDIAN
79 # define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8)
81 # define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8)
84 crc = __cpu_to_le32(crc);
86 if(unlikely(((long)b)&3 && len)){
91 } while ((--len) && ((long)b)&3 );
94 /* load data 32 bits wide, xor data 32 bits wide. */
95 size_t save_len = len & 3;
97 --b; /* use pre increment below(*++b) for speed */
105 b++; /* point to next byte(s) */
108 /* And the last few bytes */
117 return __le32_to_cpu(crc);
121 # elif CRC_LE_BITS == 4
124 crc = (crc >> 4) ^ crc32table_le[crc & 15];
125 crc = (crc >> 4) ^ crc32table_le[crc & 15];
128 # elif CRC_LE_BITS == 2
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];
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
148 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len);
152 * In fact, the table-based code will work in this case, but it can be
153 * simplified by inlining the table in ?: form.
156 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len)
161 for (i = 0; i < 8; i++)
163 (crc << 1) ^ ((crc & 0x80000000) ? CRCPOLY_BE :
169 #else /* Table-based approach */
170 u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len)
172 # if CRC_BE_BITS == 8
173 const u32 *b =(u32 *)p;
174 const u32 *tab = crc32table_be;
176 # ifdef __LITTLE_ENDIAN
177 # define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8)
179 # define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8)
182 crc = __cpu_to_be32(crc);
184 if(unlikely(((long)b)&3 && len)){
189 } while ((--len) && ((long)b)&3 );
191 if(likely(len >= 4)){
192 /* load data 32 bits wide, xor data 32 bits wide. */
193 size_t save_len = len & 3;
195 --b; /* use pre increment below(*++b) for speed */
203 b++; /* point to next byte(s) */
206 /* And the last few bytes */
214 return __be32_to_cpu(crc);
218 # elif CRC_BE_BITS == 4
221 crc = (crc << 4) ^ crc32table_be[crc >> 28];
222 crc = (crc << 4) ^ crc32table_be[crc >> 28];
225 # elif CRC_BE_BITS == 2
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];
239 * bitreverse - reverse the order of bits in a u32 value
240 * @x: value to be bit-reversed
242 u32 bitreverse(u32 x)
244 x = (x >> 16) | (x << 16);
245 x = (x >> 8 & 0x00ff00ff) | (x << 8 & 0xff00ff00);
246 x = (x >> 4 & 0x0f0f0f0f) | (x << 4 & 0xf0f0f0f0);
247 x = (x >> 2 & 0x33333333) | (x << 2 & 0xcccccccc);
248 x = (x >> 1 & 0x55555555) | (x << 1 & 0xaaaaaaaa);
252 EXPORT_SYMBOL(crc32_le);
253 EXPORT_SYMBOL(crc32_be);
254 EXPORT_SYMBOL(bitreverse);
257 * A brief CRC tutorial.
259 * A CRC is a long-division remainder. You add the CRC to the message,
260 * and the whole thing (message+CRC) is a multiple of the given
261 * CRC polynomial. To check the CRC, you can either check that the
262 * CRC matches the recomputed value, *or* you can check that the
263 * remainder computed on the message+CRC is 0. This latter approach
264 * is used by a lot of hardware implementations, and is why so many
265 * protocols put the end-of-frame flag after the CRC.
267 * It's actually the same long division you learned in school, except that
268 * - We're working in binary, so the digits are only 0 and 1, and
269 * - When dividing polynomials, there are no carries. Rather than add and
270 * subtract, we just xor. Thus, we tend to get a bit sloppy about
271 * the difference between adding and subtracting.
273 * A 32-bit CRC polynomial is actually 33 bits long. But since it's
274 * 33 bits long, bit 32 is always going to be set, so usually the CRC
275 * is written in hex with the most significant bit omitted. (If you're
276 * familiar with the IEEE 754 floating-point format, it's the same idea.)
278 * Note that a CRC is computed over a string of *bits*, so you have
279 * to decide on the endianness of the bits within each byte. To get
280 * the best error-detecting properties, this should correspond to the
281 * order they're actually sent. For example, standard RS-232 serial is
282 * little-endian; the most significant bit (sometimes used for parity)
283 * is sent last. And when appending a CRC word to a message, you should
284 * do it in the right order, matching the endianness.
286 * Just like with ordinary division, the remainder is always smaller than
287 * the divisor (the CRC polynomial) you're dividing by. Each step of the
288 * division, you take one more digit (bit) of the dividend and append it
289 * to the current remainder. Then you figure out the appropriate multiple
290 * of the divisor to subtract to being the remainder back into range.
291 * In binary, it's easy - it has to be either 0 or 1, and to make the
292 * XOR cancel, it's just a copy of bit 32 of the remainder.
294 * When computing a CRC, we don't care about the quotient, so we can
295 * throw the quotient bit away, but subtract the appropriate multiple of
296 * the polynomial from the remainder and we're back to where we started,
297 * ready to process the next bit.
299 * A big-endian CRC written this way would be coded like:
300 * for (i = 0; i < input_bits; i++) {
301 * multiple = remainder & 0x80000000 ? CRCPOLY : 0;
302 * remainder = (remainder << 1 | next_input_bit()) ^ multiple;
304 * Notice how, to get at bit 32 of the shifted remainder, we look
305 * at bit 31 of the remainder *before* shifting it.
307 * But also notice how the next_input_bit() bits we're shifting into
308 * the remainder don't actually affect any decision-making until
309 * 32 bits later. Thus, the first 32 cycles of this are pretty boring.
310 * Also, to add the CRC to a message, we need a 32-bit-long hole for it at
311 * the end, so we have to add 32 extra cycles shifting in zeros at the
312 * end of every message,
314 * So the standard trick is to rearrage merging in the next_input_bit()
315 * until the moment it's needed. Then the first 32 cycles can be precomputed,
316 * and merging in the final 32 zero bits to make room for the CRC can be
318 * This changes the code to:
319 * for (i = 0; i < input_bits; i++) {
320 * remainder ^= next_input_bit() << 31;
321 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0;
322 * remainder = (remainder << 1) ^ multiple;
324 * With this optimization, the little-endian code is simpler:
325 * for (i = 0; i < input_bits; i++) {
326 * remainder ^= next_input_bit();
327 * multiple = (remainder & 1) ? CRCPOLY : 0;
328 * remainder = (remainder >> 1) ^ multiple;
331 * Note that the other details of endianness have been hidden in CRCPOLY
332 * (which must be bit-reversed) and next_input_bit().
334 * However, as long as next_input_bit is returning the bits in a sensible
335 * order, we can actually do the merging 8 or more bits at a time rather
336 * than one bit at a time:
337 * for (i = 0; i < input_bytes; i++) {
338 * remainder ^= next_input_byte() << 24;
339 * for (j = 0; j < 8; j++) {
340 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0;
341 * remainder = (remainder << 1) ^ multiple;
344 * Or in little-endian:
345 * for (i = 0; i < input_bytes; i++) {
346 * remainder ^= next_input_byte();
347 * for (j = 0; j < 8; j++) {
348 * multiple = (remainder & 1) ? CRCPOLY : 0;
349 * remainder = (remainder << 1) ^ multiple;
352 * If the input is a multiple of 32 bits, you can even XOR in a 32-bit
353 * word at a time and increase the inner loop count to 32.
355 * You can also mix and match the two loop styles, for example doing the
356 * bulk of a message byte-at-a-time and adding bit-at-a-time processing
357 * for any fractional bytes at the end.
359 * The only remaining optimization is to the byte-at-a-time table method.
360 * Here, rather than just shifting one bit of the remainder to decide
361 * in the correct multiple to subtract, we can shift a byte at a time.
362 * This produces a 40-bit (rather than a 33-bit) intermediate remainder,
363 * but again the multiple of the polynomial to subtract depends only on
364 * the high bits, the high 8 bits in this case.
366 * The multile we need in that case is the low 32 bits of a 40-bit
367 * value whose high 8 bits are given, and which is a multiple of the
368 * generator polynomial. This is simply the CRC-32 of the given
371 * Two more details: normally, appending zero bits to a message which
372 * is already a multiple of a polynomial produces a larger multiple of that
373 * polynomial. To enable a CRC to detect this condition, it's common to
374 * invert the CRC before appending it. This makes the remainder of the
375 * message+crc come out not as zero, but some fixed non-zero value.
377 * The same problem applies to zero bits prepended to the message, and
378 * a similar solution is used. Instead of starting with a remainder of
379 * 0, an initial remainder of all ones is used. As long as you start
380 * the same way on decoding, it doesn't make a difference.
388 #if 0 /*Not used at present */
390 buf_dump(char const *prefix, unsigned char const *buf, size_t len)
392 fputs(prefix, stdout);
394 printf(" %02x", *buf++);
400 static void bytereverse(unsigned char *buf, size_t len)
403 unsigned char x = *buf;
404 x = (x >> 4) | (x << 4);
405 x = (x >> 2 & 0x33) | (x << 2 & 0xcc);
406 x = (x >> 1 & 0x55) | (x << 1 & 0xaa);
411 static void random_garbage(unsigned char *buf, size_t len)
414 *buf++ = (unsigned char) random();
417 #if 0 /* Not used at present */
418 static void store_le(u32 x, unsigned char *buf)
420 buf[0] = (unsigned char) x;
421 buf[1] = (unsigned char) (x >> 8);
422 buf[2] = (unsigned char) (x >> 16);
423 buf[3] = (unsigned char) (x >> 24);
427 static void store_be(u32 x, unsigned char *buf)
429 buf[0] = (unsigned char) (x >> 24);
430 buf[1] = (unsigned char) (x >> 16);
431 buf[2] = (unsigned char) (x >> 8);
432 buf[3] = (unsigned char) x;
436 * This checks that CRC(buf + CRC(buf)) = 0, and that
437 * CRC commutes with bit-reversal. This has the side effect
438 * of bytewise bit-reversing the input buffer, and returns
439 * the CRC of the reversed buffer.
441 static u32 test_step(u32 init, unsigned char *buf, size_t len)
446 crc1 = crc32_be(init, buf, len);
447 store_be(crc1, buf + len);
448 crc2 = crc32_be(init, buf, len + 4);
450 printf("\nCRC cancellation fail: 0x%08x should be 0\n",
453 for (i = 0; i <= len + 4; i++) {
454 crc2 = crc32_be(init, buf, i);
455 crc2 = crc32_be(crc2, buf + i, len + 4 - i);
457 printf("\nCRC split fail: 0x%08x\n", crc2);
460 /* Now swap it around for the other test */
462 bytereverse(buf, len + 4);
463 init = bitreverse(init);
464 crc2 = bitreverse(crc1);
465 if (crc1 != bitreverse(crc2))
466 printf("\nBit reversal fail: 0x%08x -> 0x%08x -> 0x%08x\n",
467 crc1, crc2, bitreverse(crc2));
468 crc1 = crc32_le(init, buf, len);
470 printf("\nCRC endianness fail: 0x%08x != 0x%08x\n", crc1,
472 crc2 = crc32_le(init, buf, len + 4);
474 printf("\nCRC cancellation fail: 0x%08x should be 0\n",
477 for (i = 0; i <= len + 4; i++) {
478 crc2 = crc32_le(init, buf, i);
479 crc2 = crc32_le(crc2, buf + i, len + 4 - i);
481 printf("\nCRC split fail: 0x%08x\n", crc2);
493 unsigned char buf1[SIZE + 4];
494 unsigned char buf2[SIZE + 4];
495 unsigned char buf3[SIZE + 4];
497 u32 crc1, crc2, crc3;
499 for (i = 0; i <= SIZE; i++) {
500 printf("\rTesting length %d...", i);
502 random_garbage(buf1, i);
503 random_garbage(buf2, i);
504 for (j = 0; j < i; j++)
505 buf3[j] = buf1[j] ^ buf2[j];
507 crc1 = test_step(INIT1, buf1, i);
508 crc2 = test_step(INIT2, buf2, i);
509 /* Now check that CRC(buf1 ^ buf2) = CRC(buf1) ^ CRC(buf2) */
510 crc3 = test_step(INIT1 ^ INIT2, buf3, i);
511 if (crc3 != (crc1 ^ crc2))
512 printf("CRC XOR fail: 0x%08x != 0x%08x ^ 0x%08x\n",
515 printf("\nAll test complete. No failures expected.\n");
519 #endif /* UNITTEST */