Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | |
4f2a9463 | 33 | # define tole(x) __constant_cpu_to_le32(x) |
1da177e4 | 34 | #else |
4f2a9463 JT |
35 | # define tole(x) (x) |
36 | #endif | |
37 | ||
38 | #if CRC_BE_BITS == 8 | |
39 | # define tobe(x) __constant_cpu_to_be32(x) | |
40 | #else | |
41 | # define tobe(x) (x) | |
1da177e4 LT |
42 | #endif |
43 | #include "crc32table.h" | |
44 | ||
45 | MODULE_AUTHOR("Matt Domsch <Matt_Domsch@dell.com>"); | |
46 | MODULE_DESCRIPTION("Ethernet CRC32 calculations"); | |
47 | MODULE_LICENSE("GPL"); | |
48 | ||
ddcaccbc JT |
49 | #if CRC_LE_BITS == 8 || CRC_BE_BITS == 8 |
50 | ||
51 | static inline u32 | |
52 | crc32_body(u32 crc, unsigned char const *buf, size_t len, const u32 *tab) | |
53 | { | |
54 | # ifdef __LITTLE_ENDIAN | |
55 | # define DO_CRC(x) crc = tab[(crc ^ (x)) & 255 ] ^ (crc >> 8) | |
56 | # else | |
57 | # define DO_CRC(x) crc = tab[((crc >> 24) ^ (x)) & 255] ^ (crc << 8) | |
58 | # endif | |
4f2a9463 | 59 | const u32 *b; |
ddcaccbc JT |
60 | size_t rem_len; |
61 | ||
62 | /* Align it */ | |
4f2a9463 | 63 | if (unlikely((long)buf & 3 && len)) { |
ddcaccbc | 64 | do { |
4f2a9463 JT |
65 | DO_CRC(*buf++); |
66 | } while ((--len) && ((long)buf)&3); | |
ddcaccbc JT |
67 | } |
68 | rem_len = len & 3; | |
69 | /* load data 32 bits wide, xor data 32 bits wide. */ | |
70 | len = len >> 2; | |
4f2a9463 | 71 | b = (const u32 *)buf; |
ddcaccbc JT |
72 | for (--b; len; --len) { |
73 | crc ^= *++b; /* use pre increment for speed */ | |
74 | DO_CRC(0); | |
75 | DO_CRC(0); | |
76 | DO_CRC(0); | |
77 | DO_CRC(0); | |
78 | } | |
79 | len = rem_len; | |
80 | /* And the last few bytes */ | |
81 | if (len) { | |
82 | u8 *p = (u8 *)(b + 1) - 1; | |
83 | do { | |
84 | DO_CRC(*++p); /* use pre increment for speed */ | |
85 | } while (--len); | |
86 | } | |
87 | return crc; | |
4f2a9463 | 88 | #undef DO_CRC |
ddcaccbc JT |
89 | } |
90 | #endif | |
2f72100c RD |
91 | /** |
92 | * crc32_le() - Calculate bitwise little-endian Ethernet AUTODIN II CRC32 | |
93 | * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for | |
94 | * other uses, or the previous crc32 value if computing incrementally. | |
95 | * @p: pointer to buffer over which CRC is run | |
96 | * @len: length of buffer @p | |
97 | */ | |
e8c44319 | 98 | u32 __pure crc32_le(u32 crc, unsigned char const *p, size_t len); |
2f72100c | 99 | |
1da177e4 LT |
100 | #if CRC_LE_BITS == 1 |
101 | /* | |
102 | * In fact, the table-based code will work in this case, but it can be | |
103 | * simplified by inlining the table in ?: form. | |
104 | */ | |
105 | ||
e8c44319 | 106 | u32 __pure crc32_le(u32 crc, unsigned char const *p, size_t len) |
1da177e4 LT |
107 | { |
108 | int i; | |
109 | while (len--) { | |
110 | crc ^= *p++; | |
111 | for (i = 0; i < 8; i++) | |
112 | crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_LE : 0); | |
113 | } | |
114 | return crc; | |
115 | } | |
116 | #else /* Table-based approach */ | |
117 | ||
e8c44319 | 118 | u32 __pure crc32_le(u32 crc, unsigned char const *p, size_t len) |
1da177e4 LT |
119 | { |
120 | # if CRC_LE_BITS == 8 | |
1da177e4 LT |
121 | const u32 *tab = crc32table_le; |
122 | ||
1da177e4 | 123 | crc = __cpu_to_le32(crc); |
ddcaccbc | 124 | crc = crc32_body(crc, p, len, tab); |
1da177e4 | 125 | return __le32_to_cpu(crc); |
1da177e4 LT |
126 | # elif CRC_LE_BITS == 4 |
127 | while (len--) { | |
128 | crc ^= *p++; | |
129 | crc = (crc >> 4) ^ crc32table_le[crc & 15]; | |
130 | crc = (crc >> 4) ^ crc32table_le[crc & 15]; | |
131 | } | |
132 | return crc; | |
133 | # elif CRC_LE_BITS == 2 | |
134 | while (len--) { | |
135 | crc ^= *p++; | |
136 | crc = (crc >> 2) ^ crc32table_le[crc & 3]; | |
137 | crc = (crc >> 2) ^ crc32table_le[crc & 3]; | |
138 | crc = (crc >> 2) ^ crc32table_le[crc & 3]; | |
139 | crc = (crc >> 2) ^ crc32table_le[crc & 3]; | |
140 | } | |
141 | return crc; | |
142 | # endif | |
143 | } | |
144 | #endif | |
145 | ||
2f72100c RD |
146 | /** |
147 | * crc32_be() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32 | |
148 | * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for | |
149 | * other uses, or the previous crc32 value if computing incrementally. | |
150 | * @p: pointer to buffer over which CRC is run | |
151 | * @len: length of buffer @p | |
152 | */ | |
e8c44319 | 153 | u32 __pure crc32_be(u32 crc, unsigned char const *p, size_t len); |
2f72100c | 154 | |
1da177e4 LT |
155 | #if CRC_BE_BITS == 1 |
156 | /* | |
157 | * In fact, the table-based code will work in this case, but it can be | |
158 | * simplified by inlining the table in ?: form. | |
159 | */ | |
160 | ||
e8c44319 | 161 | u32 __pure crc32_be(u32 crc, unsigned char const *p, size_t len) |
1da177e4 LT |
162 | { |
163 | int i; | |
164 | while (len--) { | |
165 | crc ^= *p++ << 24; | |
166 | for (i = 0; i < 8; i++) | |
167 | crc = | |
168 | (crc << 1) ^ ((crc & 0x80000000) ? CRCPOLY_BE : | |
169 | 0); | |
170 | } | |
171 | return crc; | |
172 | } | |
173 | ||
174 | #else /* Table-based approach */ | |
e8c44319 | 175 | u32 __pure crc32_be(u32 crc, unsigned char const *p, size_t len) |
1da177e4 LT |
176 | { |
177 | # if CRC_BE_BITS == 8 | |
1da177e4 LT |
178 | const u32 *tab = crc32table_be; |
179 | ||
1da177e4 | 180 | crc = __cpu_to_be32(crc); |
ddcaccbc | 181 | crc = crc32_body(crc, p, len, tab); |
1da177e4 | 182 | return __be32_to_cpu(crc); |
1da177e4 LT |
183 | # elif CRC_BE_BITS == 4 |
184 | while (len--) { | |
185 | crc ^= *p++ << 24; | |
186 | crc = (crc << 4) ^ crc32table_be[crc >> 28]; | |
187 | crc = (crc << 4) ^ crc32table_be[crc >> 28]; | |
188 | } | |
189 | return crc; | |
190 | # elif CRC_BE_BITS == 2 | |
191 | while (len--) { | |
192 | crc ^= *p++ << 24; | |
193 | crc = (crc << 2) ^ crc32table_be[crc >> 30]; | |
194 | crc = (crc << 2) ^ crc32table_be[crc >> 30]; | |
195 | crc = (crc << 2) ^ crc32table_be[crc >> 30]; | |
196 | crc = (crc << 2) ^ crc32table_be[crc >> 30]; | |
197 | } | |
198 | return crc; | |
199 | # endif | |
200 | } | |
201 | #endif | |
202 | ||
1da177e4 LT |
203 | EXPORT_SYMBOL(crc32_le); |
204 | EXPORT_SYMBOL(crc32_be); | |
1da177e4 LT |
205 | |
206 | /* | |
207 | * A brief CRC tutorial. | |
208 | * | |
209 | * A CRC is a long-division remainder. You add the CRC to the message, | |
210 | * and the whole thing (message+CRC) is a multiple of the given | |
211 | * CRC polynomial. To check the CRC, you can either check that the | |
212 | * CRC matches the recomputed value, *or* you can check that the | |
213 | * remainder computed on the message+CRC is 0. This latter approach | |
214 | * is used by a lot of hardware implementations, and is why so many | |
215 | * protocols put the end-of-frame flag after the CRC. | |
216 | * | |
217 | * It's actually the same long division you learned in school, except that | |
218 | * - We're working in binary, so the digits are only 0 and 1, and | |
219 | * - When dividing polynomials, there are no carries. Rather than add and | |
220 | * subtract, we just xor. Thus, we tend to get a bit sloppy about | |
221 | * the difference between adding and subtracting. | |
222 | * | |
223 | * A 32-bit CRC polynomial is actually 33 bits long. But since it's | |
224 | * 33 bits long, bit 32 is always going to be set, so usually the CRC | |
225 | * is written in hex with the most significant bit omitted. (If you're | |
226 | * familiar with the IEEE 754 floating-point format, it's the same idea.) | |
227 | * | |
228 | * Note that a CRC is computed over a string of *bits*, so you have | |
229 | * to decide on the endianness of the bits within each byte. To get | |
230 | * the best error-detecting properties, this should correspond to the | |
231 | * order they're actually sent. For example, standard RS-232 serial is | |
232 | * little-endian; the most significant bit (sometimes used for parity) | |
233 | * is sent last. And when appending a CRC word to a message, you should | |
234 | * do it in the right order, matching the endianness. | |
235 | * | |
236 | * Just like with ordinary division, the remainder is always smaller than | |
237 | * the divisor (the CRC polynomial) you're dividing by. Each step of the | |
238 | * division, you take one more digit (bit) of the dividend and append it | |
239 | * to the current remainder. Then you figure out the appropriate multiple | |
240 | * of the divisor to subtract to being the remainder back into range. | |
241 | * In binary, it's easy - it has to be either 0 or 1, and to make the | |
242 | * XOR cancel, it's just a copy of bit 32 of the remainder. | |
243 | * | |
244 | * When computing a CRC, we don't care about the quotient, so we can | |
245 | * throw the quotient bit away, but subtract the appropriate multiple of | |
246 | * the polynomial from the remainder and we're back to where we started, | |
247 | * ready to process the next bit. | |
248 | * | |
249 | * A big-endian CRC written this way would be coded like: | |
250 | * for (i = 0; i < input_bits; i++) { | |
251 | * multiple = remainder & 0x80000000 ? CRCPOLY : 0; | |
252 | * remainder = (remainder << 1 | next_input_bit()) ^ multiple; | |
253 | * } | |
254 | * Notice how, to get at bit 32 of the shifted remainder, we look | |
255 | * at bit 31 of the remainder *before* shifting it. | |
256 | * | |
257 | * But also notice how the next_input_bit() bits we're shifting into | |
258 | * the remainder don't actually affect any decision-making until | |
259 | * 32 bits later. Thus, the first 32 cycles of this are pretty boring. | |
260 | * Also, to add the CRC to a message, we need a 32-bit-long hole for it at | |
261 | * the end, so we have to add 32 extra cycles shifting in zeros at the | |
262 | * end of every message, | |
263 | * | |
264 | * So the standard trick is to rearrage merging in the next_input_bit() | |
265 | * until the moment it's needed. Then the first 32 cycles can be precomputed, | |
266 | * and merging in the final 32 zero bits to make room for the CRC can be | |
267 | * skipped entirely. | |
268 | * This changes the code to: | |
269 | * for (i = 0; i < input_bits; i++) { | |
270 | * remainder ^= next_input_bit() << 31; | |
271 | * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; | |
272 | * remainder = (remainder << 1) ^ multiple; | |
273 | * } | |
274 | * With this optimization, the little-endian code is simpler: | |
275 | * for (i = 0; i < input_bits; i++) { | |
276 | * remainder ^= next_input_bit(); | |
277 | * multiple = (remainder & 1) ? CRCPOLY : 0; | |
278 | * remainder = (remainder >> 1) ^ multiple; | |
279 | * } | |
280 | * | |
281 | * Note that the other details of endianness have been hidden in CRCPOLY | |
282 | * (which must be bit-reversed) and next_input_bit(). | |
283 | * | |
284 | * However, as long as next_input_bit is returning the bits in a sensible | |
285 | * order, we can actually do the merging 8 or more bits at a time rather | |
286 | * than one bit at a time: | |
287 | * for (i = 0; i < input_bytes; i++) { | |
288 | * remainder ^= next_input_byte() << 24; | |
289 | * for (j = 0; j < 8; j++) { | |
290 | * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; | |
291 | * remainder = (remainder << 1) ^ multiple; | |
292 | * } | |
293 | * } | |
294 | * Or in little-endian: | |
295 | * for (i = 0; i < input_bytes; i++) { | |
296 | * remainder ^= next_input_byte(); | |
297 | * for (j = 0; j < 8; j++) { | |
298 | * multiple = (remainder & 1) ? CRCPOLY : 0; | |
299 | * remainder = (remainder << 1) ^ multiple; | |
300 | * } | |
301 | * } | |
302 | * If the input is a multiple of 32 bits, you can even XOR in a 32-bit | |
303 | * word at a time and increase the inner loop count to 32. | |
304 | * | |
305 | * You can also mix and match the two loop styles, for example doing the | |
306 | * bulk of a message byte-at-a-time and adding bit-at-a-time processing | |
307 | * for any fractional bytes at the end. | |
308 | * | |
309 | * The only remaining optimization is to the byte-at-a-time table method. | |
310 | * Here, rather than just shifting one bit of the remainder to decide | |
311 | * in the correct multiple to subtract, we can shift a byte at a time. | |
312 | * This produces a 40-bit (rather than a 33-bit) intermediate remainder, | |
313 | * but again the multiple of the polynomial to subtract depends only on | |
314 | * the high bits, the high 8 bits in this case. | |
315 | * | |
643d1f7f | 316 | * The multiple we need in that case is the low 32 bits of a 40-bit |
1da177e4 LT |
317 | * value whose high 8 bits are given, and which is a multiple of the |
318 | * generator polynomial. This is simply the CRC-32 of the given | |
319 | * one-byte message. | |
320 | * | |
321 | * Two more details: normally, appending zero bits to a message which | |
322 | * is already a multiple of a polynomial produces a larger multiple of that | |
323 | * polynomial. To enable a CRC to detect this condition, it's common to | |
324 | * invert the CRC before appending it. This makes the remainder of the | |
325 | * message+crc come out not as zero, but some fixed non-zero value. | |
326 | * | |
327 | * The same problem applies to zero bits prepended to the message, and | |
328 | * a similar solution is used. Instead of starting with a remainder of | |
329 | * 0, an initial remainder of all ones is used. As long as you start | |
330 | * the same way on decoding, it doesn't make a difference. | |
331 | */ | |
332 | ||
333 | #ifdef UNITTEST | |
334 | ||
335 | #include <stdlib.h> | |
336 | #include <stdio.h> | |
337 | ||
338 | #if 0 /*Not used at present */ | |
339 | static void | |
340 | buf_dump(char const *prefix, unsigned char const *buf, size_t len) | |
341 | { | |
342 | fputs(prefix, stdout); | |
343 | while (len--) | |
344 | printf(" %02x", *buf++); | |
345 | putchar('\n'); | |
346 | ||
347 | } | |
348 | #endif | |
349 | ||
350 | static void bytereverse(unsigned char *buf, size_t len) | |
351 | { | |
352 | while (len--) { | |
906d66df | 353 | unsigned char x = bitrev8(*buf); |
1da177e4 LT |
354 | *buf++ = x; |
355 | } | |
356 | } | |
357 | ||
358 | static void random_garbage(unsigned char *buf, size_t len) | |
359 | { | |
360 | while (len--) | |
361 | *buf++ = (unsigned char) random(); | |
362 | } | |
363 | ||
364 | #if 0 /* Not used at present */ | |
365 | static void store_le(u32 x, unsigned char *buf) | |
366 | { | |
367 | buf[0] = (unsigned char) x; | |
368 | buf[1] = (unsigned char) (x >> 8); | |
369 | buf[2] = (unsigned char) (x >> 16); | |
370 | buf[3] = (unsigned char) (x >> 24); | |
371 | } | |
372 | #endif | |
373 | ||
374 | static void store_be(u32 x, unsigned char *buf) | |
375 | { | |
376 | buf[0] = (unsigned char) (x >> 24); | |
377 | buf[1] = (unsigned char) (x >> 16); | |
378 | buf[2] = (unsigned char) (x >> 8); | |
379 | buf[3] = (unsigned char) x; | |
380 | } | |
381 | ||
382 | /* | |
383 | * This checks that CRC(buf + CRC(buf)) = 0, and that | |
384 | * CRC commutes with bit-reversal. This has the side effect | |
385 | * of bytewise bit-reversing the input buffer, and returns | |
386 | * the CRC of the reversed buffer. | |
387 | */ | |
388 | static u32 test_step(u32 init, unsigned char *buf, size_t len) | |
389 | { | |
390 | u32 crc1, crc2; | |
391 | size_t i; | |
392 | ||
393 | crc1 = crc32_be(init, buf, len); | |
394 | store_be(crc1, buf + len); | |
395 | crc2 = crc32_be(init, buf, len + 4); | |
396 | if (crc2) | |
397 | printf("\nCRC cancellation fail: 0x%08x should be 0\n", | |
398 | crc2); | |
399 | ||
400 | for (i = 0; i <= len + 4; i++) { | |
401 | crc2 = crc32_be(init, buf, i); | |
402 | crc2 = crc32_be(crc2, buf + i, len + 4 - i); | |
403 | if (crc2) | |
404 | printf("\nCRC split fail: 0x%08x\n", crc2); | |
405 | } | |
406 | ||
407 | /* Now swap it around for the other test */ | |
408 | ||
409 | bytereverse(buf, len + 4); | |
906d66df AM |
410 | init = bitrev32(init); |
411 | crc2 = bitrev32(crc1); | |
412 | if (crc1 != bitrev32(crc2)) | |
cfc646fa | 413 | printf("\nBit reversal fail: 0x%08x -> 0x%08x -> 0x%08x\n", |
906d66df | 414 | crc1, crc2, bitrev32(crc2)); |
1da177e4 LT |
415 | crc1 = crc32_le(init, buf, len); |
416 | if (crc1 != crc2) | |
417 | printf("\nCRC endianness fail: 0x%08x != 0x%08x\n", crc1, | |
418 | crc2); | |
419 | crc2 = crc32_le(init, buf, len + 4); | |
420 | if (crc2) | |
421 | printf("\nCRC cancellation fail: 0x%08x should be 0\n", | |
422 | crc2); | |
423 | ||
424 | for (i = 0; i <= len + 4; i++) { | |
425 | crc2 = crc32_le(init, buf, i); | |
426 | crc2 = crc32_le(crc2, buf + i, len + 4 - i); | |
427 | if (crc2) | |
428 | printf("\nCRC split fail: 0x%08x\n", crc2); | |
429 | } | |
430 | ||
431 | return crc1; | |
432 | } | |
433 | ||
434 | #define SIZE 64 | |
435 | #define INIT1 0 | |
436 | #define INIT2 0 | |
437 | ||
438 | int main(void) | |
439 | { | |
440 | unsigned char buf1[SIZE + 4]; | |
441 | unsigned char buf2[SIZE + 4]; | |
442 | unsigned char buf3[SIZE + 4]; | |
443 | int i, j; | |
444 | u32 crc1, crc2, crc3; | |
445 | ||
446 | for (i = 0; i <= SIZE; i++) { | |
447 | printf("\rTesting length %d...", i); | |
448 | fflush(stdout); | |
449 | random_garbage(buf1, i); | |
450 | random_garbage(buf2, i); | |
451 | for (j = 0; j < i; j++) | |
452 | buf3[j] = buf1[j] ^ buf2[j]; | |
453 | ||
454 | crc1 = test_step(INIT1, buf1, i); | |
455 | crc2 = test_step(INIT2, buf2, i); | |
456 | /* Now check that CRC(buf1 ^ buf2) = CRC(buf1) ^ CRC(buf2) */ | |
457 | crc3 = test_step(INIT1 ^ INIT2, buf3, i); | |
458 | if (crc3 != (crc1 ^ crc2)) | |
459 | printf("CRC XOR fail: 0x%08x != 0x%08x ^ 0x%08x\n", | |
460 | crc3, crc1, crc2); | |
461 | } | |
462 | printf("\nAll test complete. No failures expected.\n"); | |
463 | return 0; | |
464 | } | |
465 | ||
466 | #endif /* UNITTEST */ |