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/config.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 #if __GNUC__ >= 3	/* 2.x has "attribute", but only 3.0 has "pure */
42 #define attribute(x) __attribute__(x)
43 #else
44 #define attribute(x)
45 #endif
46 
47 
48 MODULE_AUTHOR("Matt Domsch <Matt_Domsch@dell.com>");
49 MODULE_DESCRIPTION("Ethernet CRC32 calculations");
50 MODULE_LICENSE("GPL");
51 
52 #if CRC_LE_BITS == 1
53 /*
54  * In fact, the table-based code will work in this case, but it can be
55  * simplified by inlining the table in ?: form.
56  */
57 
58 /**
59  * crc32_le() - Calculate bitwise little-endian Ethernet AUTODIN II CRC32
60  * @crc - seed value for computation.  ~0 for Ethernet, sometimes 0 for
61  *        other uses, or the previous crc32 value if computing incrementally.
62  * @p   - pointer to buffer over which CRC is run
63  * @len - length of buffer @p
64  *
65  */
u32(pure)66 u32 attribute((pure)) crc32_le(u32 crc, unsigned char const *p, size_t len)
67 {
68 	int i;
69 	while (len--) {
70 		crc ^= *p++;
71 		for (i = 0; i < 8; i++)
72 			crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_LE : 0);
73 	}
74 	return crc;
75 }
76 #else				/* Table-based approach */
77 
78 /**
79  * crc32_le() - Calculate bitwise little-endian Ethernet AUTODIN II CRC32
80  * @crc - seed value for computation.  ~0 for Ethernet, sometimes 0 for
81  *        other uses, or the previous crc32 value if computing incrementally.
82  * @p   - pointer to buffer over which CRC is run
83  * @len - length of buffer @p
84  *
85  */
u32(pure)86 u32 attribute((pure)) crc32_le(u32 crc, unsigned char const *p, size_t len)
87 {
88 # if CRC_LE_BITS == 8
89 	const u32      *b =(u32 *)p;
90 	const u32      *tab = crc32table_le;
91 
92 # ifdef __LITTLE_ENDIAN
93 #  define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8)
94 # else
95 #  define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8)
96 # endif
97 
98 	crc = __cpu_to_le32(crc);
99 	/* Align it */
100 	if(unlikely(((long)b)&3 && len)){
101 		do {
102 			u8 *p = (u8 *)b;
103 			DO_CRC(*p++);
104 			b = (void *)p;
105 		} while ((--len) && ((long)b)&3 );
106 	}
107 	if(likely(len >= 4)){
108 		/* load data 32 bits wide, xor data 32 bits wide. */
109 		size_t save_len = len & 3;
110 	        len = len >> 2;
111 		--b; /* use pre increment below(*++b) for speed */
112 		do {
113 			crc ^= *++b;
114 			DO_CRC(0);
115 			DO_CRC(0);
116 			DO_CRC(0);
117 			DO_CRC(0);
118 		} while (--len);
119 		b++; /* point to next byte(s) */
120 		len = save_len;
121 	}
122 	/* And the last few bytes */
123 	if(len){
124 		do {
125 			u8 *p = (u8 *)b;
126 			DO_CRC(*p++);
127 			b = (void *)p;
128 		} while (--len);
129 	}
130 
131 	return __le32_to_cpu(crc);
132 #undef ENDIAN_SHIFT
133 #undef DO_CRC
134 
135 # elif CRC_LE_BITS == 4
136 	while (len--) {
137 		crc ^= *p++;
138 		crc = (crc >> 4) ^ crc32table_le[crc & 15];
139 		crc = (crc >> 4) ^ crc32table_le[crc & 15];
140 	}
141 	return crc;
142 # elif CRC_LE_BITS == 2
143 	while (len--) {
144 		crc ^= *p++;
145 		crc = (crc >> 2) ^ crc32table_le[crc & 3];
146 		crc = (crc >> 2) ^ crc32table_le[crc & 3];
147 		crc = (crc >> 2) ^ crc32table_le[crc & 3];
148 		crc = (crc >> 2) ^ crc32table_le[crc & 3];
149 	}
150 	return crc;
151 # endif
152 }
153 #endif
154 
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 
161 /**
162  * crc32_be() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32
163  * @crc - seed value for computation.  ~0 for Ethernet, sometimes 0 for
164  *        other uses, or the previous crc32 value if computing incrementally.
165  * @p   - pointer to buffer over which CRC is run
166  * @len - length of buffer @p
167  *
168  */
u32(pure)169 u32 attribute((pure)) crc32_be(u32 crc, unsigned char const *p, size_t len)
170 {
171 	int i;
172 	while (len--) {
173 		crc ^= *p++ << 24;
174 		for (i = 0; i < 8; i++)
175 			crc =
176 			    (crc << 1) ^ ((crc & 0x80000000) ? CRCPOLY_BE :
177 					  0);
178 	}
179 	return crc;
180 }
181 
182 #else				/* Table-based approach */
183 /**
184  * crc32_be() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32
185  * @crc - seed value for computation.  ~0 for Ethernet, sometimes 0 for
186  *        other uses, or the previous crc32 value if computing incrementally.
187  * @p   - pointer to buffer over which CRC is run
188  * @len - length of buffer @p
189  *
190  */
u32(pure)191 u32 attribute((pure)) crc32_be(u32 crc, unsigned char const *p, size_t len)
192 {
193 # if CRC_BE_BITS == 8
194 	const u32      *b =(u32 *)p;
195 	const u32      *tab = crc32table_be;
196 
197 # ifdef __LITTLE_ENDIAN
198 #  define DO_CRC(x) crc = tab[ (crc ^ (x)) & 255 ] ^ (crc>>8)
199 # else
200 #  define DO_CRC(x) crc = tab[ ((crc >> 24) ^ (x)) & 255] ^ (crc<<8)
201 # endif
202 
203 	crc = __cpu_to_be32(crc);
204 	/* Align it */
205 	if(unlikely(((long)b)&3 && len)){
206 		do {
207 			u8 *p = (u8 *)b;
208 			DO_CRC(*p++);
209 			b = (u32 *)p;
210 		} while ((--len) && ((long)b)&3 );
211 	}
212 	if(likely(len >= 4)){
213 		/* load data 32 bits wide, xor data 32 bits wide. */
214 		size_t save_len = len & 3;
215 	        len = len >> 2;
216 		--b; /* use pre increment below(*++b) for speed */
217 		do {
218 			crc ^= *++b;
219 			DO_CRC(0);
220 			DO_CRC(0);
221 			DO_CRC(0);
222 			DO_CRC(0);
223 		} while (--len);
224 		b++; /* point to next byte(s) */
225 		len = save_len;
226 	}
227 	/* And the last few bytes */
228 	if(len){
229 		do {
230 			u8 *p = (u8 *)b;
231 			DO_CRC(*p++);
232 			b = (void *)p;
233 		} while (--len);
234 	}
235 	return __be32_to_cpu(crc);
236 #undef ENDIAN_SHIFT
237 #undef DO_CRC
238 
239 # elif CRC_BE_BITS == 4
240 	while (len--) {
241 		crc ^= *p++ << 24;
242 		crc = (crc << 4) ^ crc32table_be[crc >> 28];
243 		crc = (crc << 4) ^ crc32table_be[crc >> 28];
244 	}
245 	return crc;
246 # elif CRC_BE_BITS == 2
247 	while (len--) {
248 		crc ^= *p++ << 24;
249 		crc = (crc << 2) ^ crc32table_be[crc >> 30];
250 		crc = (crc << 2) ^ crc32table_be[crc >> 30];
251 		crc = (crc << 2) ^ crc32table_be[crc >> 30];
252 		crc = (crc << 2) ^ crc32table_be[crc >> 30];
253 	}
254 	return crc;
255 # endif
256 }
257 #endif
258 
bitreverse(u32 x)259 u32 bitreverse(u32 x)
260 {
261 	x = (x >> 16) | (x << 16);
262 	x = (x >> 8 & 0x00ff00ff) | (x << 8 & 0xff00ff00);
263 	x = (x >> 4 & 0x0f0f0f0f) | (x << 4 & 0xf0f0f0f0);
264 	x = (x >> 2 & 0x33333333) | (x << 2 & 0xcccccccc);
265 	x = (x >> 1 & 0x55555555) | (x << 1 & 0xaaaaaaaa);
266 	return x;
267 }
268 
269 #ifndef CONFIG_CRC32
270 	/* To ensure that this file is pulled in from lib/lib.a if it's
271 	   configured in but nothing in-kernel uses it, we export its
272 	   symbols from kernel/ksyms.c in the CONFIG_CRC32=y case.
273 	   Otherwise (either modular or pulled in by the makefile magic)
274 	   we export them from here. */
275 EXPORT_SYMBOL(crc32_le);
276 EXPORT_SYMBOL(crc32_be);
277 EXPORT_SYMBOL(bitreverse);
278 #endif
279 
280 /*
281  * A brief CRC tutorial.
282  *
283  * A CRC is a long-division remainder.  You add the CRC to the message,
284  * and the whole thing (message+CRC) is a multiple of the given
285  * CRC polynomial.  To check the CRC, you can either check that the
286  * CRC matches the recomputed value, *or* you can check that the
287  * remainder computed on the message+CRC is 0.  This latter approach
288  * is used by a lot of hardware implementations, and is why so many
289  * protocols put the end-of-frame flag after the CRC.
290  *
291  * It's actually the same long division you learned in school, except that
292  * - We're working in binary, so the digits are only 0 and 1, and
293  * - When dividing polynomials, there are no carries.  Rather than add and
294  *   subtract, we just xor.  Thus, we tend to get a bit sloppy about
295  *   the difference between adding and subtracting.
296  *
297  * A 32-bit CRC polynomial is actually 33 bits long.  But since it's
298  * 33 bits long, bit 32 is always going to be set, so usually the CRC
299  * is written in hex with the most significant bit omitted.  (If you're
300  * familiar with the IEEE 754 floating-point format, it's the same idea.)
301  *
302  * Note that a CRC is computed over a string of *bits*, so you have
303  * to decide on the endianness of the bits within each byte.  To get
304  * the best error-detecting properties, this should correspond to the
305  * order they're actually sent.  For example, standard RS-232 serial is
306  * little-endian; the most significant bit (sometimes used for parity)
307  * is sent last.  And when appending a CRC word to a message, you should
308  * do it in the right order, matching the endianness.
309  *
310  * Just like with ordinary division, the remainder is always smaller than
311  * the divisor (the CRC polynomial) you're dividing by.  Each step of the
312  * division, you take one more digit (bit) of the dividend and append it
313  * to the current remainder.  Then you figure out the appropriate multiple
314  * of the divisor to subtract to being the remainder back into range.
315  * In binary, it's easy - it has to be either 0 or 1, and to make the
316  * XOR cancel, it's just a copy of bit 32 of the remainder.
317  *
318  * When computing a CRC, we don't care about the quotient, so we can
319  * throw the quotient bit away, but subtract the appropriate multiple of
320  * the polynomial from the remainder and we're back to where we started,
321  * ready to process the next bit.
322  *
323  * A big-endian CRC written this way would be coded like:
324  * for (i = 0; i < input_bits; i++) {
325  * 	multiple = remainder & 0x80000000 ? CRCPOLY : 0;
326  * 	remainder = (remainder << 1 | next_input_bit()) ^ multiple;
327  * }
328  * Notice how, to get at bit 32 of the shifted remainder, we look
329  * at bit 31 of the remainder *before* shifting it.
330  *
331  * But also notice how the next_input_bit() bits we're shifting into
332  * the remainder don't actually affect any decision-making until
333  * 32 bits later.  Thus, the first 32 cycles of this are pretty boring.
334  * Also, to add the CRC to a message, we need a 32-bit-long hole for it at
335  * the end, so we have to add 32 extra cycles shifting in zeros at the
336  * end of every message,
337  *
338  * So the standard trick is to rearrage merging in the next_input_bit()
339  * until the moment it's needed.  Then the first 32 cycles can be precomputed,
340  * and merging in the final 32 zero bits to make room for the CRC can be
341  * skipped entirely.
342  * This changes the code to:
343  * for (i = 0; i < input_bits; i++) {
344  *      remainder ^= next_input_bit() << 31;
345  * 	multiple = (remainder & 0x80000000) ? CRCPOLY : 0;
346  * 	remainder = (remainder << 1) ^ multiple;
347  * }
348  * With this optimization, the little-endian code is simpler:
349  * for (i = 0; i < input_bits; i++) {
350  *      remainder ^= next_input_bit();
351  * 	multiple = (remainder & 1) ? CRCPOLY : 0;
352  * 	remainder = (remainder >> 1) ^ multiple;
353  * }
354  *
355  * Note that the other details of endianness have been hidden in CRCPOLY
356  * (which must be bit-reversed) and next_input_bit().
357  *
358  * However, as long as next_input_bit is returning the bits in a sensible
359  * order, we can actually do the merging 8 or more bits at a time rather
360  * than one bit at a time:
361  * for (i = 0; i < input_bytes; i++) {
362  * 	remainder ^= next_input_byte() << 24;
363  * 	for (j = 0; j < 8; j++) {
364  * 		multiple = (remainder & 0x80000000) ? CRCPOLY : 0;
365  * 		remainder = (remainder << 1) ^ multiple;
366  * 	}
367  * }
368  * Or in little-endian:
369  * for (i = 0; i < input_bytes; i++) {
370  * 	remainder ^= next_input_byte();
371  * 	for (j = 0; j < 8; j++) {
372  * 		multiple = (remainder & 1) ? CRCPOLY : 0;
373  * 		remainder = (remainder << 1) ^ multiple;
374  * 	}
375  * }
376  * If the input is a multiple of 32 bits, you can even XOR in a 32-bit
377  * word at a time and increase the inner loop count to 32.
378  *
379  * You can also mix and match the two loop styles, for example doing the
380  * bulk of a message byte-at-a-time and adding bit-at-a-time processing
381  * for any fractional bytes at the end.
382  *
383  * The only remaining optimization is to the byte-at-a-time table method.
384  * Here, rather than just shifting one bit of the remainder to decide
385  * in the correct multiple to subtract, we can shift a byte at a time.
386  * This produces a 40-bit (rather than a 33-bit) intermediate remainder,
387  * but again the multiple of the polynomial to subtract depends only on
388  * the high bits, the high 8 bits in this case.
389  *
390  * The multile we need in that case is the low 32 bits of a 40-bit
391  * value whose high 8 bits are given, and which is a multiple of the
392  * generator polynomial.  This is simply the CRC-32 of the given
393  * one-byte message.
394  *
395  * Two more details: normally, appending zero bits to a message which
396  * is already a multiple of a polynomial produces a larger multiple of that
397  * polynomial.  To enable a CRC to detect this condition, it's common to
398  * invert the CRC before appending it.  This makes the remainder of the
399  * message+crc come out not as zero, but some fixed non-zero value.
400  *
401  * The same problem applies to zero bits prepended to the message, and
402  * a similar solution is used.  Instead of starting with a remainder of
403  * 0, an initial remainder of all ones is used.  As long as you start
404  * the same way on decoding, it doesn't make a difference.
405  */
406 
407 #if UNITTEST
408 
409 #include <stdlib.h>
410 #include <stdio.h>
411 
412 #if 0				/*Not used at present */
413 static void
414 buf_dump(char const *prefix, unsigned char const *buf, size_t len)
415 {
416 	fputs(prefix, stdout);
417 	while (len--)
418 		printf(" %02x", *buf++);
419 	putchar('\n');
420 
421 }
422 #endif
423 
bytereverse(unsigned char * buf,size_t len)424 static void bytereverse(unsigned char *buf, size_t len)
425 {
426 	while (len--) {
427 		unsigned char x = *buf;
428 		x = (x >> 4) | (x << 4);
429 		x = (x >> 2 & 0x33) | (x << 2 & 0xcc);
430 		x = (x >> 1 & 0x55) | (x << 1 & 0xaa);
431 		*buf++ = x;
432 	}
433 }
434 
random_garbage(unsigned char * buf,size_t len)435 static void random_garbage(unsigned char *buf, size_t len)
436 {
437 	while (len--)
438 		*buf++ = (unsigned char) random();
439 }
440 
441 #if 0				/* Not used at present */
442 static void store_le(u32 x, unsigned char *buf)
443 {
444 	buf[0] = (unsigned char) x;
445 	buf[1] = (unsigned char) (x >> 8);
446 	buf[2] = (unsigned char) (x >> 16);
447 	buf[3] = (unsigned char) (x >> 24);
448 }
449 #endif
450 
store_be(u32 x,unsigned char * buf)451 static void store_be(u32 x, unsigned char *buf)
452 {
453 	buf[0] = (unsigned char) (x >> 24);
454 	buf[1] = (unsigned char) (x >> 16);
455 	buf[2] = (unsigned char) (x >> 8);
456 	buf[3] = (unsigned char) x;
457 }
458 
459 /*
460  * This checks that CRC(buf + CRC(buf)) = 0, and that
461  * CRC commutes with bit-reversal.  This has the side effect
462  * of bytewise bit-reversing the input buffer, and returns
463  * the CRC of the reversed buffer.
464  */
test_step(u32 init,unsigned char * buf,size_t len)465 static u32 test_step(u32 init, unsigned char *buf, size_t len)
466 {
467 	u32 crc1, crc2;
468 	size_t i;
469 
470 	crc1 = crc32_be(init, buf, len);
471 	store_be(crc1, buf + len);
472 	crc2 = crc32_be(init, buf, len + 4);
473 	if (crc2)
474 		printf("\nCRC cancellation fail: 0x%08x should be 0\n",
475 		       crc2);
476 
477 	for (i = 0; i <= len + 4; i++) {
478 		crc2 = crc32_be(init, buf, i);
479 		crc2 = crc32_be(crc2, buf + i, len + 4 - i);
480 		if (crc2)
481 			printf("\nCRC split fail: 0x%08x\n", crc2);
482 	}
483 
484 	/* Now swap it around for the other test */
485 
486 	bytereverse(buf, len + 4);
487 	init = bitreverse(init);
488 	crc2 = bitreverse(crc1);
489 	if (crc1 != bitreverse(crc2))
490 		printf("\nBit reversal fail: 0x%08x -> %0x08x -> 0x%08x\n",
491 		       crc1, crc2, bitreverse(crc2));
492 	crc1 = crc32_le(init, buf, len);
493 	if (crc1 != crc2)
494 		printf("\nCRC endianness fail: 0x%08x != 0x%08x\n", crc1,
495 		       crc2);
496 	crc2 = crc32_le(init, buf, len + 4);
497 	if (crc2)
498 		printf("\nCRC cancellation fail: 0x%08x should be 0\n",
499 		       crc2);
500 
501 	for (i = 0; i <= len + 4; i++) {
502 		crc2 = crc32_le(init, buf, i);
503 		crc2 = crc32_le(crc2, buf + i, len + 4 - i);
504 		if (crc2)
505 			printf("\nCRC split fail: 0x%08x\n", crc2);
506 	}
507 
508 	return crc1;
509 }
510 
511 #define SIZE 64
512 #define INIT1 0
513 #define INIT2 0
514 
main(void)515 int main(void)
516 {
517 	unsigned char buf1[SIZE + 4];
518 	unsigned char buf2[SIZE + 4];
519 	unsigned char buf3[SIZE + 4];
520 	int i, j;
521 	u32 crc1, crc2, crc3;
522 
523 	for (i = 0; i <= SIZE; i++) {
524 		printf("\rTesting length %d...", i);
525 		fflush(stdout);
526 		random_garbage(buf1, i);
527 		random_garbage(buf2, i);
528 		for (j = 0; j < i; j++)
529 			buf3[j] = buf1[j] ^ buf2[j];
530 
531 		crc1 = test_step(INIT1, buf1, i);
532 		crc2 = test_step(INIT2, buf2, i);
533 		/* Now check that CRC(buf1 ^ buf2) = CRC(buf1) ^ CRC(buf2) */
534 		crc3 = test_step(INIT1 ^ INIT2, buf3, i);
535 		if (crc3 != (crc1 ^ crc2))
536 			printf("CRC XOR fail: 0x%08x != 0x%08x ^ 0x%08x\n",
537 			       crc3, crc1, crc2);
538 	}
539 	printf("\nAll test complete.  No failures expected.\n");
540 	return 0;
541 }
542 
543 #endif				/* UNITTEST */
544