xref: /linux/lib/crc32.c (revision 6ee738610f41b59733f63718f0bdbcba7d3a3f12)
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 
41 MODULE_AUTHOR("Matt Domsch <Matt_Domsch@dell.com>");
42 MODULE_DESCRIPTION("Ethernet CRC32 calculations");
43 MODULE_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  */
52 u32 __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 
60 u32 __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 
72 u32 __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  */
148 u32 __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 
156 u32 __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 */
170 u32 __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 
238 EXPORT_SYMBOL(crc32_le);
239 EXPORT_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 multiple 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 #if 0				/*Not used at present */
374 static void
375 buf_dump(char const *prefix, unsigned char const *buf, size_t len)
376 {
377 	fputs(prefix, stdout);
378 	while (len--)
379 		printf(" %02x", *buf++);
380 	putchar('\n');
381 
382 }
383 #endif
384 
385 static void bytereverse(unsigned char *buf, size_t len)
386 {
387 	while (len--) {
388 		unsigned char x = bitrev8(*buf);
389 		*buf++ = x;
390 	}
391 }
392 
393 static void random_garbage(unsigned char *buf, size_t len)
394 {
395 	while (len--)
396 		*buf++ = (unsigned char) random();
397 }
398 
399 #if 0				/* Not used at present */
400 static void store_le(u32 x, unsigned char *buf)
401 {
402 	buf[0] = (unsigned char) x;
403 	buf[1] = (unsigned char) (x >> 8);
404 	buf[2] = (unsigned char) (x >> 16);
405 	buf[3] = (unsigned char) (x >> 24);
406 }
407 #endif
408 
409 static void store_be(u32 x, unsigned char *buf)
410 {
411 	buf[0] = (unsigned char) (x >> 24);
412 	buf[1] = (unsigned char) (x >> 16);
413 	buf[2] = (unsigned char) (x >> 8);
414 	buf[3] = (unsigned char) x;
415 }
416 
417 /*
418  * This checks that CRC(buf + CRC(buf)) = 0, and that
419  * CRC commutes with bit-reversal.  This has the side effect
420  * of bytewise bit-reversing the input buffer, and returns
421  * the CRC of the reversed buffer.
422  */
423 static u32 test_step(u32 init, unsigned char *buf, size_t len)
424 {
425 	u32 crc1, crc2;
426 	size_t i;
427 
428 	crc1 = crc32_be(init, buf, len);
429 	store_be(crc1, buf + len);
430 	crc2 = crc32_be(init, buf, len + 4);
431 	if (crc2)
432 		printf("\nCRC cancellation fail: 0x%08x should be 0\n",
433 		       crc2);
434 
435 	for (i = 0; i <= len + 4; i++) {
436 		crc2 = crc32_be(init, buf, i);
437 		crc2 = crc32_be(crc2, buf + i, len + 4 - i);
438 		if (crc2)
439 			printf("\nCRC split fail: 0x%08x\n", crc2);
440 	}
441 
442 	/* Now swap it around for the other test */
443 
444 	bytereverse(buf, len + 4);
445 	init = bitrev32(init);
446 	crc2 = bitrev32(crc1);
447 	if (crc1 != bitrev32(crc2))
448 		printf("\nBit reversal fail: 0x%08x -> 0x%08x -> 0x%08x\n",
449 		       crc1, crc2, bitrev32(crc2));
450 	crc1 = crc32_le(init, buf, len);
451 	if (crc1 != crc2)
452 		printf("\nCRC endianness fail: 0x%08x != 0x%08x\n", crc1,
453 		       crc2);
454 	crc2 = crc32_le(init, buf, len + 4);
455 	if (crc2)
456 		printf("\nCRC cancellation fail: 0x%08x should be 0\n",
457 		       crc2);
458 
459 	for (i = 0; i <= len + 4; i++) {
460 		crc2 = crc32_le(init, buf, i);
461 		crc2 = crc32_le(crc2, buf + i, len + 4 - i);
462 		if (crc2)
463 			printf("\nCRC split fail: 0x%08x\n", crc2);
464 	}
465 
466 	return crc1;
467 }
468 
469 #define SIZE 64
470 #define INIT1 0
471 #define INIT2 0
472 
473 int main(void)
474 {
475 	unsigned char buf1[SIZE + 4];
476 	unsigned char buf2[SIZE + 4];
477 	unsigned char buf3[SIZE + 4];
478 	int i, j;
479 	u32 crc1, crc2, crc3;
480 
481 	for (i = 0; i <= SIZE; i++) {
482 		printf("\rTesting length %d...", i);
483 		fflush(stdout);
484 		random_garbage(buf1, i);
485 		random_garbage(buf2, i);
486 		for (j = 0; j < i; j++)
487 			buf3[j] = buf1[j] ^ buf2[j];
488 
489 		crc1 = test_step(INIT1, buf1, i);
490 		crc2 = test_step(INIT2, buf2, i);
491 		/* Now check that CRC(buf1 ^ buf2) = CRC(buf1) ^ CRC(buf2) */
492 		crc3 = test_step(INIT1 ^ INIT2, buf3, i);
493 		if (crc3 != (crc1 ^ crc2))
494 			printf("CRC XOR fail: 0x%08x != 0x%08x ^ 0x%08x\n",
495 			       crc3, crc1, crc2);
496 	}
497 	printf("\nAll test complete.  No failures expected.\n");
498 	return 0;
499 }
500 
501 #endif				/* UNITTEST */
502