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");
45 #if CRC_LE_BITS == 8 || CRC_BE_BITS == 8
48 crc32_body(u32 crc
, unsigned char const *buf
, size_t len
, const u32
*tab
)
50 # ifdef __LITTLE_ENDIAN
51 # define DO_CRC(x) crc = tab[(crc ^ (x)) & 255 ] ^ (crc >> 8)
53 # define DO_CRC(x) crc = tab[((crc >> 24) ^ (x)) & 255] ^ (crc << 8)
55 const u32
*b
= (const u32
*)buf
;
59 if (unlikely((long)b
& 3 && len
)) {
63 } while ((--len
) && ((long)p
)&3);
67 /* load data 32 bits wide, xor data 32 bits wide. */
69 for (--b
; len
; --len
) {
70 crc
^= *++b
; /* use pre increment for speed */
77 /* And the last few bytes */
79 u8
*p
= (u8
*)(b
+ 1) - 1;
81 DO_CRC(*++p
); /* use pre increment for speed */
88 * crc32_le() - Calculate bitwise little-endian Ethernet AUTODIN II CRC32
89 * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for
90 * other uses, or the previous crc32 value if computing incrementally.
91 * @p: pointer to buffer over which CRC is run
92 * @len: length of buffer @p
94 u32 __pure
crc32_le(u32 crc
, unsigned char const *p
, size_t len
);
98 * In fact, the table-based code will work in this case, but it can be
99 * simplified by inlining the table in ?: form.
102 u32 __pure
crc32_le(u32 crc
, unsigned char const *p
, size_t len
)
107 for (i
= 0; i
< 8; i
++)
108 crc
= (crc
>> 1) ^ ((crc
& 1) ? CRCPOLY_LE
: 0);
112 #else /* Table-based approach */
114 u32 __pure
crc32_le(u32 crc
, unsigned char const *p
, size_t len
)
116 # if CRC_LE_BITS == 8
117 const u32
*tab
= crc32table_le
;
119 crc
= __cpu_to_le32(crc
);
120 crc
= crc32_body(crc
, p
, len
, tab
);
121 return __le32_to_cpu(crc
);
125 # elif CRC_LE_BITS == 4
128 crc
= (crc
>> 4) ^ crc32table_le
[crc
& 15];
129 crc
= (crc
>> 4) ^ crc32table_le
[crc
& 15];
132 # elif CRC_LE_BITS == 2
135 crc
= (crc
>> 2) ^ crc32table_le
[crc
& 3];
136 crc
= (crc
>> 2) ^ crc32table_le
[crc
& 3];
137 crc
= (crc
>> 2) ^ crc32table_le
[crc
& 3];
138 crc
= (crc
>> 2) ^ crc32table_le
[crc
& 3];
146 * crc32_be() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32
147 * @crc: seed value for computation. ~0 for Ethernet, sometimes 0 for
148 * other uses, or the previous crc32 value if computing incrementally.
149 * @p: pointer to buffer over which CRC is run
150 * @len: length of buffer @p
152 u32 __pure
crc32_be(u32 crc
, unsigned char const *p
, size_t len
);
156 * In fact, the table-based code will work in this case, but it can be
157 * simplified by inlining the table in ?: form.
160 u32 __pure
crc32_be(u32 crc
, unsigned char const *p
, size_t len
)
165 for (i
= 0; i
< 8; i
++)
167 (crc
<< 1) ^ ((crc
& 0x80000000) ? CRCPOLY_BE
:
173 #else /* Table-based approach */
174 u32 __pure
crc32_be(u32 crc
, unsigned char const *p
, size_t len
)
176 # if CRC_BE_BITS == 8
177 const u32
*tab
= crc32table_be
;
179 crc
= __cpu_to_be32(crc
);
180 crc
= crc32_body(crc
, p
, len
, tab
);
181 return __be32_to_cpu(crc
);
185 # elif CRC_BE_BITS == 4
188 crc
= (crc
<< 4) ^ crc32table_be
[crc
>> 28];
189 crc
= (crc
<< 4) ^ crc32table_be
[crc
>> 28];
192 # elif CRC_BE_BITS == 2
195 crc
= (crc
<< 2) ^ crc32table_be
[crc
>> 30];
196 crc
= (crc
<< 2) ^ crc32table_be
[crc
>> 30];
197 crc
= (crc
<< 2) ^ crc32table_be
[crc
>> 30];
198 crc
= (crc
<< 2) ^ crc32table_be
[crc
>> 30];
205 EXPORT_SYMBOL(crc32_le
);
206 EXPORT_SYMBOL(crc32_be
);
209 * A brief CRC tutorial.
211 * A CRC is a long-division remainder. You add the CRC to the message,
212 * and the whole thing (message+CRC) is a multiple of the given
213 * CRC polynomial. To check the CRC, you can either check that the
214 * CRC matches the recomputed value, *or* you can check that the
215 * remainder computed on the message+CRC is 0. This latter approach
216 * is used by a lot of hardware implementations, and is why so many
217 * protocols put the end-of-frame flag after the CRC.
219 * It's actually the same long division you learned in school, except that
220 * - We're working in binary, so the digits are only 0 and 1, and
221 * - When dividing polynomials, there are no carries. Rather than add and
222 * subtract, we just xor. Thus, we tend to get a bit sloppy about
223 * the difference between adding and subtracting.
225 * A 32-bit CRC polynomial is actually 33 bits long. But since it's
226 * 33 bits long, bit 32 is always going to be set, so usually the CRC
227 * is written in hex with the most significant bit omitted. (If you're
228 * familiar with the IEEE 754 floating-point format, it's the same idea.)
230 * Note that a CRC is computed over a string of *bits*, so you have
231 * to decide on the endianness of the bits within each byte. To get
232 * the best error-detecting properties, this should correspond to the
233 * order they're actually sent. For example, standard RS-232 serial is
234 * little-endian; the most significant bit (sometimes used for parity)
235 * is sent last. And when appending a CRC word to a message, you should
236 * do it in the right order, matching the endianness.
238 * Just like with ordinary division, the remainder is always smaller than
239 * the divisor (the CRC polynomial) you're dividing by. Each step of the
240 * division, you take one more digit (bit) of the dividend and append it
241 * to the current remainder. Then you figure out the appropriate multiple
242 * of the divisor to subtract to being the remainder back into range.
243 * In binary, it's easy - it has to be either 0 or 1, and to make the
244 * XOR cancel, it's just a copy of bit 32 of the remainder.
246 * When computing a CRC, we don't care about the quotient, so we can
247 * throw the quotient bit away, but subtract the appropriate multiple of
248 * the polynomial from the remainder and we're back to where we started,
249 * ready to process the next bit.
251 * A big-endian CRC written this way would be coded like:
252 * for (i = 0; i < input_bits; i++) {
253 * multiple = remainder & 0x80000000 ? CRCPOLY : 0;
254 * remainder = (remainder << 1 | next_input_bit()) ^ multiple;
256 * Notice how, to get at bit 32 of the shifted remainder, we look
257 * at bit 31 of the remainder *before* shifting it.
259 * But also notice how the next_input_bit() bits we're shifting into
260 * the remainder don't actually affect any decision-making until
261 * 32 bits later. Thus, the first 32 cycles of this are pretty boring.
262 * Also, to add the CRC to a message, we need a 32-bit-long hole for it at
263 * the end, so we have to add 32 extra cycles shifting in zeros at the
264 * end of every message,
266 * So the standard trick is to rearrage merging in the next_input_bit()
267 * until the moment it's needed. Then the first 32 cycles can be precomputed,
268 * and merging in the final 32 zero bits to make room for the CRC can be
270 * This changes the code to:
271 * for (i = 0; i < input_bits; i++) {
272 * remainder ^= next_input_bit() << 31;
273 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0;
274 * remainder = (remainder << 1) ^ multiple;
276 * With this optimization, the little-endian code is simpler:
277 * for (i = 0; i < input_bits; i++) {
278 * remainder ^= next_input_bit();
279 * multiple = (remainder & 1) ? CRCPOLY : 0;
280 * remainder = (remainder >> 1) ^ multiple;
283 * Note that the other details of endianness have been hidden in CRCPOLY
284 * (which must be bit-reversed) and next_input_bit().
286 * However, as long as next_input_bit is returning the bits in a sensible
287 * order, we can actually do the merging 8 or more bits at a time rather
288 * than one bit at a time:
289 * for (i = 0; i < input_bytes; i++) {
290 * remainder ^= next_input_byte() << 24;
291 * for (j = 0; j < 8; j++) {
292 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0;
293 * remainder = (remainder << 1) ^ multiple;
296 * Or in little-endian:
297 * for (i = 0; i < input_bytes; i++) {
298 * remainder ^= next_input_byte();
299 * for (j = 0; j < 8; j++) {
300 * multiple = (remainder & 1) ? CRCPOLY : 0;
301 * remainder = (remainder << 1) ^ multiple;
304 * If the input is a multiple of 32 bits, you can even XOR in a 32-bit
305 * word at a time and increase the inner loop count to 32.
307 * You can also mix and match the two loop styles, for example doing the
308 * bulk of a message byte-at-a-time and adding bit-at-a-time processing
309 * for any fractional bytes at the end.
311 * The only remaining optimization is to the byte-at-a-time table method.
312 * Here, rather than just shifting one bit of the remainder to decide
313 * in the correct multiple to subtract, we can shift a byte at a time.
314 * This produces a 40-bit (rather than a 33-bit) intermediate remainder,
315 * but again the multiple of the polynomial to subtract depends only on
316 * the high bits, the high 8 bits in this case.
318 * The multiple we need in that case is the low 32 bits of a 40-bit
319 * value whose high 8 bits are given, and which is a multiple of the
320 * generator polynomial. This is simply the CRC-32 of the given
323 * Two more details: normally, appending zero bits to a message which
324 * is already a multiple of a polynomial produces a larger multiple of that
325 * polynomial. To enable a CRC to detect this condition, it's common to
326 * invert the CRC before appending it. This makes the remainder of the
327 * message+crc come out not as zero, but some fixed non-zero value.
329 * The same problem applies to zero bits prepended to the message, and
330 * a similar solution is used. Instead of starting with a remainder of
331 * 0, an initial remainder of all ones is used. As long as you start
332 * the same way on decoding, it doesn't make a difference.
340 #if 0 /*Not used at present */
342 buf_dump(char const *prefix
, unsigned char const *buf
, size_t len
)
344 fputs(prefix
, stdout
);
346 printf(" %02x", *buf
++);
352 static void bytereverse(unsigned char *buf
, size_t len
)
355 unsigned char x
= bitrev8(*buf
);
360 static void random_garbage(unsigned char *buf
, size_t len
)
363 *buf
++ = (unsigned char) random();
366 #if 0 /* Not used at present */
367 static void store_le(u32 x
, unsigned char *buf
)
369 buf
[0] = (unsigned char) x
;
370 buf
[1] = (unsigned char) (x
>> 8);
371 buf
[2] = (unsigned char) (x
>> 16);
372 buf
[3] = (unsigned char) (x
>> 24);
376 static void store_be(u32 x
, unsigned char *buf
)
378 buf
[0] = (unsigned char) (x
>> 24);
379 buf
[1] = (unsigned char) (x
>> 16);
380 buf
[2] = (unsigned char) (x
>> 8);
381 buf
[3] = (unsigned char) x
;
385 * This checks that CRC(buf + CRC(buf)) = 0, and that
386 * CRC commutes with bit-reversal. This has the side effect
387 * of bytewise bit-reversing the input buffer, and returns
388 * the CRC of the reversed buffer.
390 static u32
test_step(u32 init
, unsigned char *buf
, size_t len
)
395 crc1
= crc32_be(init
, buf
, len
);
396 store_be(crc1
, buf
+ len
);
397 crc2
= crc32_be(init
, buf
, len
+ 4);
399 printf("\nCRC cancellation fail: 0x%08x should be 0\n",
402 for (i
= 0; i
<= len
+ 4; i
++) {
403 crc2
= crc32_be(init
, buf
, i
);
404 crc2
= crc32_be(crc2
, buf
+ i
, len
+ 4 - i
);
406 printf("\nCRC split fail: 0x%08x\n", crc2
);
409 /* Now swap it around for the other test */
411 bytereverse(buf
, len
+ 4);
412 init
= bitrev32(init
);
413 crc2
= bitrev32(crc1
);
414 if (crc1
!= bitrev32(crc2
))
415 printf("\nBit reversal fail: 0x%08x -> 0x%08x -> 0x%08x\n",
416 crc1
, crc2
, bitrev32(crc2
));
417 crc1
= crc32_le(init
, buf
, len
);
419 printf("\nCRC endianness fail: 0x%08x != 0x%08x\n", crc1
,
421 crc2
= crc32_le(init
, buf
, len
+ 4);
423 printf("\nCRC cancellation fail: 0x%08x should be 0\n",
426 for (i
= 0; i
<= len
+ 4; i
++) {
427 crc2
= crc32_le(init
, buf
, i
);
428 crc2
= crc32_le(crc2
, buf
+ i
, len
+ 4 - i
);
430 printf("\nCRC split fail: 0x%08x\n", crc2
);
442 unsigned char buf1
[SIZE
+ 4];
443 unsigned char buf2
[SIZE
+ 4];
444 unsigned char buf3
[SIZE
+ 4];
446 u32 crc1
, crc2
, crc3
;
448 for (i
= 0; i
<= SIZE
; i
++) {
449 printf("\rTesting length %d...", i
);
451 random_garbage(buf1
, i
);
452 random_garbage(buf2
, i
);
453 for (j
= 0; j
< i
; j
++)
454 buf3
[j
] = buf1
[j
] ^ buf2
[j
];
456 crc1
= test_step(INIT1
, buf1
, i
);
457 crc2
= test_step(INIT2
, buf2
, i
);
458 /* Now check that CRC(buf1 ^ buf2) = CRC(buf1) ^ CRC(buf2) */
459 crc3
= test_step(INIT1
^ INIT2
, buf3
, i
);
460 if (crc3
!= (crc1
^ crc2
))
461 printf("CRC XOR fail: 0x%08x != 0x%08x ^ 0x%08x\n",
464 printf("\nAll test complete. No failures expected.\n");
468 #endif /* UNITTEST */
This page took 0.042198 seconds and 6 git commands to generate.