blob: 065198f98b3f5fc3aebb16ed8b9b81749ec8f741 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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#if CRC_LE_BITS == 1
46/*
47 * In fact, the table-based code will work in this case, but it can be
48 * simplified by inlining the table in ?: form.
49 */
50
51/**
52 * crc32_le() - Calculate bitwise little-endian Ethernet AUTODIN II CRC32
53 * @crc - seed value for computation. ~0 for Ethernet, sometimes 0 for
54 * other uses, or the previous crc32 value if computing incrementally.
55 * @p - pointer to buffer over which CRC is run
56 * @len - length of buffer @p
57 *
58 */
59u32 __attribute_pure__ crc32_le(u32 crc, unsigned char const *p, size_t len)
60{
61 int i;
62 while (len--) {
63 crc ^= *p++;
64 for (i = 0; i < 8; i++)
65 crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_LE : 0);
66 }
67 return crc;
68}
69#else /* Table-based approach */
70
71/**
72 * crc32_le() - Calculate bitwise little-endian Ethernet AUTODIN II CRC32
73 * @crc - seed value for computation. ~0 for Ethernet, sometimes 0 for
74 * other uses, or the previous crc32 value if computing incrementally.
75 * @p - pointer to buffer over which CRC is run
76 * @len - length of buffer @p
77 *
78 */
79u32 __attribute_pure__ crc32_le(u32 crc, unsigned char const *p, size_t len)
80{
81# if CRC_LE_BITS == 8
82 const u32 *b =(u32 *)p;
83 const u32 *tab = crc32table_le;
84
85# ifdef __LITTLE_ENDIAN
86# define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8)
87# else
88# define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8)
89# endif
90
91 crc = __cpu_to_le32(crc);
92 /* Align it */
93 if(unlikely(((long)b)&3 && len)){
94 do {
95 u8 *p = (u8 *)b;
96 DO_CRC(*p++);
97 b = (void *)p;
98 } while ((--len) && ((long)b)&3 );
99 }
100 if(likely(len >= 4)){
101 /* load data 32 bits wide, xor data 32 bits wide. */
102 size_t save_len = len & 3;
103 len = len >> 2;
104 --b; /* use pre increment below(*++b) for speed */
105 do {
106 crc ^= *++b;
107 DO_CRC(0);
108 DO_CRC(0);
109 DO_CRC(0);
110 DO_CRC(0);
111 } while (--len);
112 b++; /* point to next byte(s) */
113 len = save_len;
114 }
115 /* And the last few bytes */
116 if(len){
117 do {
118 u8 *p = (u8 *)b;
119 DO_CRC(*p++);
120 b = (void *)p;
121 } while (--len);
122 }
123
124 return __le32_to_cpu(crc);
125#undef ENDIAN_SHIFT
126#undef DO_CRC
127
128# elif CRC_LE_BITS == 4
129 while (len--) {
130 crc ^= *p++;
131 crc = (crc >> 4) ^ crc32table_le[crc & 15];
132 crc = (crc >> 4) ^ crc32table_le[crc & 15];
133 }
134 return crc;
135# elif CRC_LE_BITS == 2
136 while (len--) {
137 crc ^= *p++;
138 crc = (crc >> 2) ^ crc32table_le[crc & 3];
139 crc = (crc >> 2) ^ crc32table_le[crc & 3];
140 crc = (crc >> 2) ^ crc32table_le[crc & 3];
141 crc = (crc >> 2) ^ crc32table_le[crc & 3];
142 }
143 return crc;
144# endif
145}
146#endif
147
148#if CRC_BE_BITS == 1
149/*
150 * In fact, the table-based code will work in this case, but it can be
151 * simplified by inlining the table in ?: form.
152 */
153
154/**
155 * crc32_be() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32
156 * @crc - seed value for computation. ~0 for Ethernet, sometimes 0 for
157 * other uses, or the previous crc32 value if computing incrementally.
158 * @p - pointer to buffer over which CRC is run
159 * @len - length of buffer @p
160 *
161 */
162u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len)
163{
164 int i;
165 while (len--) {
166 crc ^= *p++ << 24;
167 for (i = 0; i < 8; i++)
168 crc =
169 (crc << 1) ^ ((crc & 0x80000000) ? CRCPOLY_BE :
170 0);
171 }
172 return crc;
173}
174
175#else /* Table-based approach */
176/**
177 * crc32_be() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32
178 * @crc - seed value for computation. ~0 for Ethernet, sometimes 0 for
179 * other uses, or the previous crc32 value if computing incrementally.
180 * @p - pointer to buffer over which CRC is run
181 * @len - length of buffer @p
182 *
183 */
184u32 __attribute_pure__ crc32_be(u32 crc, unsigned char const *p, size_t len)
185{
186# if CRC_BE_BITS == 8
187 const u32 *b =(u32 *)p;
188 const u32 *tab = crc32table_be;
189
190# ifdef __LITTLE_ENDIAN
191# define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8)
192# else
193# define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8)
194# endif
195
196 crc = __cpu_to_be32(crc);
197 /* Align it */
198 if(unlikely(((long)b)&3 && len)){
199 do {
200 u8 *p = (u8 *)b;
201 DO_CRC(*p++);
202 b = (u32 *)p;
203 } while ((--len) && ((long)b)&3 );
204 }
205 if(likely(len >= 4)){
206 /* load data 32 bits wide, xor data 32 bits wide. */
207 size_t save_len = len & 3;
208 len = len >> 2;
209 --b; /* use pre increment below(*++b) for speed */
210 do {
211 crc ^= *++b;
212 DO_CRC(0);
213 DO_CRC(0);
214 DO_CRC(0);
215 DO_CRC(0);
216 } while (--len);
217 b++; /* point to next byte(s) */
218 len = save_len;
219 }
220 /* And the last few bytes */
221 if(len){
222 do {
223 u8 *p = (u8 *)b;
224 DO_CRC(*p++);
225 b = (void *)p;
226 } while (--len);
227 }
228 return __be32_to_cpu(crc);
229#undef ENDIAN_SHIFT
230#undef DO_CRC
231
232# elif CRC_BE_BITS == 4
233 while (len--) {
234 crc ^= *p++ << 24;
235 crc = (crc << 4) ^ crc32table_be[crc >> 28];
236 crc = (crc << 4) ^ crc32table_be[crc >> 28];
237 }
238 return crc;
239# elif CRC_BE_BITS == 2
240 while (len--) {
241 crc ^= *p++ << 24;
242 crc = (crc << 2) ^ crc32table_be[crc >> 30];
243 crc = (crc << 2) ^ crc32table_be[crc >> 30];
244 crc = (crc << 2) ^ crc32table_be[crc >> 30];
245 crc = (crc << 2) ^ crc32table_be[crc >> 30];
246 }
247 return crc;
248# endif
249}
250#endif
251
252u32 bitreverse(u32 x)
253{
254 x = (x >> 16) | (x << 16);
255 x = (x >> 8 & 0x00ff00ff) | (x << 8 & 0xff00ff00);
256 x = (x >> 4 & 0x0f0f0f0f) | (x << 4 & 0xf0f0f0f0);
257 x = (x >> 2 & 0x33333333) | (x << 2 & 0xcccccccc);
258 x = (x >> 1 & 0x55555555) | (x << 1 & 0xaaaaaaaa);
259 return x;
260}
261
262EXPORT_SYMBOL(crc32_le);
263EXPORT_SYMBOL(crc32_be);
264EXPORT_SYMBOL(bitreverse);
265
266/*
267 * A brief CRC tutorial.
268 *
269 * A CRC is a long-division remainder. You add the CRC to the message,
270 * and the whole thing (message+CRC) is a multiple of the given
271 * CRC polynomial. To check the CRC, you can either check that the
272 * CRC matches the recomputed value, *or* you can check that the
273 * remainder computed on the message+CRC is 0. This latter approach
274 * is used by a lot of hardware implementations, and is why so many
275 * protocols put the end-of-frame flag after the CRC.
276 *
277 * It's actually the same long division you learned in school, except that
278 * - We're working in binary, so the digits are only 0 and 1, and
279 * - When dividing polynomials, there are no carries. Rather than add and
280 * subtract, we just xor. Thus, we tend to get a bit sloppy about
281 * the difference between adding and subtracting.
282 *
283 * A 32-bit CRC polynomial is actually 33 bits long. But since it's
284 * 33 bits long, bit 32 is always going to be set, so usually the CRC
285 * is written in hex with the most significant bit omitted. (If you're
286 * familiar with the IEEE 754 floating-point format, it's the same idea.)
287 *
288 * Note that a CRC is computed over a string of *bits*, so you have
289 * to decide on the endianness of the bits within each byte. To get
290 * the best error-detecting properties, this should correspond to the
291 * order they're actually sent. For example, standard RS-232 serial is
292 * little-endian; the most significant bit (sometimes used for parity)
293 * is sent last. And when appending a CRC word to a message, you should
294 * do it in the right order, matching the endianness.
295 *
296 * Just like with ordinary division, the remainder is always smaller than
297 * the divisor (the CRC polynomial) you're dividing by. Each step of the
298 * division, you take one more digit (bit) of the dividend and append it
299 * to the current remainder. Then you figure out the appropriate multiple
300 * of the divisor to subtract to being the remainder back into range.
301 * In binary, it's easy - it has to be either 0 or 1, and to make the
302 * XOR cancel, it's just a copy of bit 32 of the remainder.
303 *
304 * When computing a CRC, we don't care about the quotient, so we can
305 * throw the quotient bit away, but subtract the appropriate multiple of
306 * the polynomial from the remainder and we're back to where we started,
307 * ready to process the next bit.
308 *
309 * A big-endian CRC written this way would be coded like:
310 * for (i = 0; i < input_bits; i++) {
311 * multiple = remainder & 0x80000000 ? CRCPOLY : 0;
312 * remainder = (remainder << 1 | next_input_bit()) ^ multiple;
313 * }
314 * Notice how, to get at bit 32 of the shifted remainder, we look
315 * at bit 31 of the remainder *before* shifting it.
316 *
317 * But also notice how the next_input_bit() bits we're shifting into
318 * the remainder don't actually affect any decision-making until
319 * 32 bits later. Thus, the first 32 cycles of this are pretty boring.
320 * Also, to add the CRC to a message, we need a 32-bit-long hole for it at
321 * the end, so we have to add 32 extra cycles shifting in zeros at the
322 * end of every message,
323 *
324 * So the standard trick is to rearrage merging in the next_input_bit()
325 * until the moment it's needed. Then the first 32 cycles can be precomputed,
326 * and merging in the final 32 zero bits to make room for the CRC can be
327 * skipped entirely.
328 * This changes the code to:
329 * for (i = 0; i < input_bits; i++) {
330 * remainder ^= next_input_bit() << 31;
331 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0;
332 * remainder = (remainder << 1) ^ multiple;
333 * }
334 * With this optimization, the little-endian code is simpler:
335 * for (i = 0; i < input_bits; i++) {
336 * remainder ^= next_input_bit();
337 * multiple = (remainder & 1) ? CRCPOLY : 0;
338 * remainder = (remainder >> 1) ^ multiple;
339 * }
340 *
341 * Note that the other details of endianness have been hidden in CRCPOLY
342 * (which must be bit-reversed) and next_input_bit().
343 *
344 * However, as long as next_input_bit is returning the bits in a sensible
345 * order, we can actually do the merging 8 or more bits at a time rather
346 * than one bit at a time:
347 * for (i = 0; i < input_bytes; i++) {
348 * remainder ^= next_input_byte() << 24;
349 * for (j = 0; j < 8; j++) {
350 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0;
351 * remainder = (remainder << 1) ^ multiple;
352 * }
353 * }
354 * Or in little-endian:
355 * for (i = 0; i < input_bytes; i++) {
356 * remainder ^= next_input_byte();
357 * for (j = 0; j < 8; j++) {
358 * multiple = (remainder & 1) ? CRCPOLY : 0;
359 * remainder = (remainder << 1) ^ multiple;
360 * }
361 * }
362 * If the input is a multiple of 32 bits, you can even XOR in a 32-bit
363 * word at a time and increase the inner loop count to 32.
364 *
365 * You can also mix and match the two loop styles, for example doing the
366 * bulk of a message byte-at-a-time and adding bit-at-a-time processing
367 * for any fractional bytes at the end.
368 *
369 * The only remaining optimization is to the byte-at-a-time table method.
370 * Here, rather than just shifting one bit of the remainder to decide
371 * in the correct multiple to subtract, we can shift a byte at a time.
372 * This produces a 40-bit (rather than a 33-bit) intermediate remainder,
373 * but again the multiple of the polynomial to subtract depends only on
374 * the high bits, the high 8 bits in this case.
375 *
376 * The multile we need in that case is the low 32 bits of a 40-bit
377 * value whose high 8 bits are given, and which is a multiple of the
378 * generator polynomial. This is simply the CRC-32 of the given
379 * one-byte message.
380 *
381 * Two more details: normally, appending zero bits to a message which
382 * is already a multiple of a polynomial produces a larger multiple of that
383 * polynomial. To enable a CRC to detect this condition, it's common to
384 * invert the CRC before appending it. This makes the remainder of the
385 * message+crc come out not as zero, but some fixed non-zero value.
386 *
387 * The same problem applies to zero bits prepended to the message, and
388 * a similar solution is used. Instead of starting with a remainder of
389 * 0, an initial remainder of all ones is used. As long as you start
390 * the same way on decoding, it doesn't make a difference.
391 */
392
393#ifdef UNITTEST
394
395#include <stdlib.h>
396#include <stdio.h>
397
398#if 0 /*Not used at present */
399static void
400buf_dump(char const *prefix, unsigned char const *buf, size_t len)
401{
402 fputs(prefix, stdout);
403 while (len--)
404 printf(" %02x", *buf++);
405 putchar('\n');
406
407}
408#endif
409
410static void bytereverse(unsigned char *buf, size_t len)
411{
412 while (len--) {
413 unsigned char x = *buf;
414 x = (x >> 4) | (x << 4);
415 x = (x >> 2 & 0x33) | (x << 2 & 0xcc);
416 x = (x >> 1 & 0x55) | (x << 1 & 0xaa);
417 *buf++ = x;
418 }
419}
420
421static void random_garbage(unsigned char *buf, size_t len)
422{
423 while (len--)
424 *buf++ = (unsigned char) random();
425}
426
427#if 0 /* Not used at present */
428static void store_le(u32 x, unsigned char *buf)
429{
430 buf[0] = (unsigned char) x;
431 buf[1] = (unsigned char) (x >> 8);
432 buf[2] = (unsigned char) (x >> 16);
433 buf[3] = (unsigned char) (x >> 24);
434}
435#endif
436
437static void store_be(u32 x, unsigned char *buf)
438{
439 buf[0] = (unsigned char) (x >> 24);
440 buf[1] = (unsigned char) (x >> 16);
441 buf[2] = (unsigned char) (x >> 8);
442 buf[3] = (unsigned char) x;
443}
444
445/*
446 * This checks that CRC(buf + CRC(buf)) = 0, and that
447 * CRC commutes with bit-reversal. This has the side effect
448 * of bytewise bit-reversing the input buffer, and returns
449 * the CRC of the reversed buffer.
450 */
451static u32 test_step(u32 init, unsigned char *buf, size_t len)
452{
453 u32 crc1, crc2;
454 size_t i;
455
456 crc1 = crc32_be(init, buf, len);
457 store_be(crc1, buf + len);
458 crc2 = crc32_be(init, buf, len + 4);
459 if (crc2)
460 printf("\nCRC cancellation fail: 0x%08x should be 0\n",
461 crc2);
462
463 for (i = 0; i <= len + 4; i++) {
464 crc2 = crc32_be(init, buf, i);
465 crc2 = crc32_be(crc2, buf + i, len + 4 - i);
466 if (crc2)
467 printf("\nCRC split fail: 0x%08x\n", crc2);
468 }
469
470 /* Now swap it around for the other test */
471
472 bytereverse(buf, len + 4);
473 init = bitreverse(init);
474 crc2 = bitreverse(crc1);
475 if (crc1 != bitreverse(crc2))
Dominik Hacklcfc646f2005-08-07 09:42:53 -0700476 printf("\nBit reversal fail: 0x%08x -> 0x%08x -> 0x%08x\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 crc1, crc2, bitreverse(crc2));
478 crc1 = crc32_le(init, buf, len);
479 if (crc1 != crc2)
480 printf("\nCRC endianness fail: 0x%08x != 0x%08x\n", crc1,
481 crc2);
482 crc2 = crc32_le(init, buf, len + 4);
483 if (crc2)
484 printf("\nCRC cancellation fail: 0x%08x should be 0\n",
485 crc2);
486
487 for (i = 0; i <= len + 4; i++) {
488 crc2 = crc32_le(init, buf, i);
489 crc2 = crc32_le(crc2, buf + i, len + 4 - i);
490 if (crc2)
491 printf("\nCRC split fail: 0x%08x\n", crc2);
492 }
493
494 return crc1;
495}
496
497#define SIZE 64
498#define INIT1 0
499#define INIT2 0
500
501int main(void)
502{
503 unsigned char buf1[SIZE + 4];
504 unsigned char buf2[SIZE + 4];
505 unsigned char buf3[SIZE + 4];
506 int i, j;
507 u32 crc1, crc2, crc3;
508
509 for (i = 0; i <= SIZE; i++) {
510 printf("\rTesting length %d...", i);
511 fflush(stdout);
512 random_garbage(buf1, i);
513 random_garbage(buf2, i);
514 for (j = 0; j < i; j++)
515 buf3[j] = buf1[j] ^ buf2[j];
516
517 crc1 = test_step(INIT1, buf1, i);
518 crc2 = test_step(INIT2, buf2, i);
519 /* Now check that CRC(buf1 ^ buf2) = CRC(buf1) ^ CRC(buf2) */
520 crc3 = test_step(INIT1 ^ INIT2, buf3, i);
521 if (crc3 != (crc1 ^ crc2))
522 printf("CRC XOR fail: 0x%08x != 0x%08x ^ 0x%08x\n",
523 crc3, crc1, crc2);
524 }
525 printf("\nAll test complete. No failures expected.\n");
526 return 0;
527}
528
529#endif /* UNITTEST */