diff options
author | David S. Miller <davem@davemloft.net> | 2014-06-25 19:04:09 -0400 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2014-06-25 19:04:09 -0400 |
commit | cb8eb77663dda1502ec419962980053cd0092d48 (patch) | |
tree | 734ab1bfada593301e315a18b01ad39e68a9730e | |
parent | 5433ba365f6dd9f30899188755eb4b093314732c (diff) | |
parent | d8f1c4778e957273c3b5b6e045d8d3af38484ca8 (diff) |
Merge branch 'crc32'
Daniel Borkmann says:
====================
crc32 combine improvements
So almost a month passed, and I don't want this to get lost
somewhere. I have applied the feedback given at that time to
this set, rebased plus tested it against latest net-next. I
decided to route this via netdev as it improves performance
upon library code that provides library bits for SCTP, i.e.
for non-linear skb csum handling in IPVS. Thus, resending
this for George before it gets lost.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
-rw-r--r-- | include/linux/crc32.h | 20 | ||||
-rw-r--r-- | lib/crc32.c | 153 |
2 files changed, 88 insertions, 85 deletions
diff --git a/include/linux/crc32.h b/include/linux/crc32.h index 7d275c4fc011..9e8a032c1788 100644 --- a/include/linux/crc32.h +++ b/include/linux/crc32.h | |||
@@ -8,8 +8,8 @@ | |||
8 | #include <linux/types.h> | 8 | #include <linux/types.h> |
9 | #include <linux/bitrev.h> | 9 | #include <linux/bitrev.h> |
10 | 10 | ||
11 | extern u32 crc32_le(u32 crc, unsigned char const *p, size_t len); | 11 | u32 __pure crc32_le(u32 crc, unsigned char const *p, size_t len); |
12 | extern u32 crc32_be(u32 crc, unsigned char const *p, size_t len); | 12 | u32 __pure crc32_be(u32 crc, unsigned char const *p, size_t len); |
13 | 13 | ||
14 | /** | 14 | /** |
15 | * crc32_le_combine - Combine two crc32 check values into one. For two | 15 | * crc32_le_combine - Combine two crc32 check values into one. For two |
@@ -29,9 +29,14 @@ extern u32 crc32_be(u32 crc, unsigned char const *p, size_t len); | |||
29 | * with the same initializer as crc1, and crc2 seed was 0. See | 29 | * with the same initializer as crc1, and crc2 seed was 0. See |
30 | * also crc32_combine_test(). | 30 | * also crc32_combine_test(). |
31 | */ | 31 | */ |
32 | extern u32 crc32_le_combine(u32 crc1, u32 crc2, size_t len2); | 32 | u32 __attribute_const__ crc32_le_shift(u32 crc, size_t len); |
33 | 33 | ||
34 | extern u32 __crc32c_le(u32 crc, unsigned char const *p, size_t len); | 34 | static inline u32 crc32_le_combine(u32 crc1, u32 crc2, size_t len2) |
35 | { | ||
36 | return crc32_le_shift(crc1, len2) ^ crc2; | ||
37 | } | ||
38 | |||
39 | u32 __pure __crc32c_le(u32 crc, unsigned char const *p, size_t len); | ||
35 | 40 | ||
36 | /** | 41 | /** |
37 | * __crc32c_le_combine - Combine two crc32c check values into one. For two | 42 | * __crc32c_le_combine - Combine two crc32c check values into one. For two |
@@ -51,7 +56,12 @@ extern u32 __crc32c_le(u32 crc, unsigned char const *p, size_t len); | |||
51 | * seeded with the same initializer as crc1, and crc2 seed | 56 | * seeded with the same initializer as crc1, and crc2 seed |
52 | * was 0. See also crc32c_combine_test(). | 57 | * was 0. See also crc32c_combine_test(). |
53 | */ | 58 | */ |
54 | extern u32 __crc32c_le_combine(u32 crc1, u32 crc2, size_t len2); | 59 | u32 __attribute_const__ __crc32c_le_shift(u32 crc, size_t len); |
60 | |||
61 | static inline u32 __crc32c_le_combine(u32 crc1, u32 crc2, size_t len2) | ||
62 | { | ||
63 | return __crc32c_le_shift(crc1, len2) ^ crc2; | ||
64 | } | ||
55 | 65 | ||
56 | #define crc32(seed, data, length) crc32_le(seed, (unsigned char const *)(data), length) | 66 | #define crc32(seed, data, length) crc32_le(seed, (unsigned char const *)(data), length) |
57 | 67 | ||
diff --git a/lib/crc32.c b/lib/crc32.c index 21a7b2135af6..9a907d489d95 100644 --- a/lib/crc32.c +++ b/lib/crc32.c | |||
@@ -50,34 +50,10 @@ MODULE_AUTHOR("Matt Domsch <Matt_Domsch@dell.com>"); | |||
50 | MODULE_DESCRIPTION("Various CRC32 calculations"); | 50 | MODULE_DESCRIPTION("Various CRC32 calculations"); |
51 | MODULE_LICENSE("GPL"); | 51 | MODULE_LICENSE("GPL"); |
52 | 52 | ||
53 | #define GF2_DIM 32 | ||
54 | |||
55 | static u32 gf2_matrix_times(u32 *mat, u32 vec) | ||
56 | { | ||
57 | u32 sum = 0; | ||
58 | |||
59 | while (vec) { | ||
60 | if (vec & 1) | ||
61 | sum ^= *mat; | ||
62 | vec >>= 1; | ||
63 | mat++; | ||
64 | } | ||
65 | |||
66 | return sum; | ||
67 | } | ||
68 | |||
69 | static void gf2_matrix_square(u32 *square, u32 *mat) | ||
70 | { | ||
71 | int i; | ||
72 | |||
73 | for (i = 0; i < GF2_DIM; i++) | ||
74 | square[i] = gf2_matrix_times(mat, mat[i]); | ||
75 | } | ||
76 | |||
77 | #if CRC_LE_BITS > 8 || CRC_BE_BITS > 8 | 53 | #if CRC_LE_BITS > 8 || CRC_BE_BITS > 8 |
78 | 54 | ||
79 | /* implements slicing-by-4 or slicing-by-8 algorithm */ | 55 | /* implements slicing-by-4 or slicing-by-8 algorithm */ |
80 | static inline u32 | 56 | static inline u32 __pure |
81 | crc32_body(u32 crc, unsigned char const *buf, size_t len, const u32 (*tab)[256]) | 57 | crc32_body(u32 crc, unsigned char const *buf, size_t len, const u32 (*tab)[256]) |
82 | { | 58 | { |
83 | # ifdef __LITTLE_ENDIAN | 59 | # ifdef __LITTLE_ENDIAN |
@@ -155,51 +131,6 @@ crc32_body(u32 crc, unsigned char const *buf, size_t len, const u32 (*tab)[256]) | |||
155 | } | 131 | } |
156 | #endif | 132 | #endif |
157 | 133 | ||
158 | /* For conditions of distribution and use, see copyright notice in zlib.h */ | ||
159 | static u32 crc32_generic_combine(u32 crc1, u32 crc2, size_t len2, | ||
160 | u32 polynomial) | ||
161 | { | ||
162 | u32 even[GF2_DIM]; /* Even-power-of-two zeros operator */ | ||
163 | u32 odd[GF2_DIM]; /* Odd-power-of-two zeros operator */ | ||
164 | u32 row; | ||
165 | int i; | ||
166 | |||
167 | if (len2 <= 0) | ||
168 | return crc1; | ||
169 | |||
170 | /* Put operator for one zero bit in odd */ | ||
171 | odd[0] = polynomial; | ||
172 | row = 1; | ||
173 | for (i = 1; i < GF2_DIM; i++) { | ||
174 | odd[i] = row; | ||
175 | row <<= 1; | ||
176 | } | ||
177 | |||
178 | gf2_matrix_square(even, odd); /* Put operator for two zero bits in even */ | ||
179 | gf2_matrix_square(odd, even); /* Put operator for four zero bits in odd */ | ||
180 | |||
181 | /* Apply len2 zeros to crc1 (first square will put the operator for one | ||
182 | * zero byte, eight zero bits, in even). | ||
183 | */ | ||
184 | do { | ||
185 | /* Apply zeros operator for this bit of len2 */ | ||
186 | gf2_matrix_square(even, odd); | ||
187 | if (len2 & 1) | ||
188 | crc1 = gf2_matrix_times(even, crc1); | ||
189 | len2 >>= 1; | ||
190 | /* If no more bits set, then done */ | ||
191 | if (len2 == 0) | ||
192 | break; | ||
193 | /* Another iteration of the loop with odd and even swapped */ | ||
194 | gf2_matrix_square(odd, even); | ||
195 | if (len2 & 1) | ||
196 | crc1 = gf2_matrix_times(odd, crc1); | ||
197 | len2 >>= 1; | ||
198 | } while (len2 != 0); | ||
199 | |||
200 | crc1 ^= crc2; | ||
201 | return crc1; | ||
202 | } | ||
203 | 134 | ||
204 | /** | 135 | /** |
205 | * crc32_le_generic() - Calculate bitwise little-endian Ethernet AUTODIN II | 136 | * crc32_le_generic() - Calculate bitwise little-endian Ethernet AUTODIN II |
@@ -271,19 +202,81 @@ u32 __pure __crc32c_le(u32 crc, unsigned char const *p, size_t len) | |||
271 | (const u32 (*)[256])crc32ctable_le, CRC32C_POLY_LE); | 202 | (const u32 (*)[256])crc32ctable_le, CRC32C_POLY_LE); |
272 | } | 203 | } |
273 | #endif | 204 | #endif |
274 | u32 __pure crc32_le_combine(u32 crc1, u32 crc2, size_t len2) | 205 | EXPORT_SYMBOL(crc32_le); |
206 | EXPORT_SYMBOL(__crc32c_le); | ||
207 | |||
208 | /* | ||
209 | * This multiplies the polynomials x and y modulo the given modulus. | ||
210 | * This follows the "little-endian" CRC convention that the lsbit | ||
211 | * represents the highest power of x, and the msbit represents x^0. | ||
212 | */ | ||
213 | static u32 __attribute_const__ gf2_multiply(u32 x, u32 y, u32 modulus) | ||
275 | { | 214 | { |
276 | return crc32_generic_combine(crc1, crc2, len2, CRCPOLY_LE); | 215 | u32 product = x & 1 ? y : 0; |
216 | int i; | ||
217 | |||
218 | for (i = 0; i < 31; i++) { | ||
219 | product = (product >> 1) ^ (product & 1 ? modulus : 0); | ||
220 | x >>= 1; | ||
221 | product ^= x & 1 ? y : 0; | ||
222 | } | ||
223 | |||
224 | return product; | ||
277 | } | 225 | } |
278 | 226 | ||
279 | u32 __pure __crc32c_le_combine(u32 crc1, u32 crc2, size_t len2) | 227 | /** |
228 | * crc32_generic_shift - Append len 0 bytes to crc, in logarithmic time | ||
229 | * @crc: The original little-endian CRC (i.e. lsbit is x^31 coefficient) | ||
230 | * @len: The number of bytes. @crc is multiplied by x^(8*@len) | ||
231 | * @polynomial: The modulus used to reduce the result to 32 bits. | ||
232 | * | ||
233 | * It's possible to parallelize CRC computations by computing a CRC | ||
234 | * over separate ranges of a buffer, then summing them. | ||
235 | * This shifts the given CRC by 8*len bits (i.e. produces the same effect | ||
236 | * as appending len bytes of zero to the data), in time proportional | ||
237 | * to log(len). | ||
238 | */ | ||
239 | static u32 __attribute_const__ crc32_generic_shift(u32 crc, size_t len, | ||
240 | u32 polynomial) | ||
280 | { | 241 | { |
281 | return crc32_generic_combine(crc1, crc2, len2, CRC32C_POLY_LE); | 242 | u32 power = polynomial; /* CRC of x^32 */ |
243 | int i; | ||
244 | |||
245 | /* Shift up to 32 bits in the simple linear way */ | ||
246 | for (i = 0; i < 8 * (int)(len & 3); i++) | ||
247 | crc = (crc >> 1) ^ (crc & 1 ? polynomial : 0); | ||
248 | |||
249 | len >>= 2; | ||
250 | if (!len) | ||
251 | return crc; | ||
252 | |||
253 | for (;;) { | ||
254 | /* "power" is x^(2^i), modulo the polynomial */ | ||
255 | if (len & 1) | ||
256 | crc = gf2_multiply(crc, power, polynomial); | ||
257 | |||
258 | len >>= 1; | ||
259 | if (!len) | ||
260 | break; | ||
261 | |||
262 | /* Square power, advancing to x^(2^(i+1)) */ | ||
263 | power = gf2_multiply(power, power, polynomial); | ||
264 | } | ||
265 | |||
266 | return crc; | ||
282 | } | 267 | } |
283 | EXPORT_SYMBOL(crc32_le); | 268 | |
284 | EXPORT_SYMBOL(crc32_le_combine); | 269 | u32 __attribute_const__ crc32_le_shift(u32 crc, size_t len) |
285 | EXPORT_SYMBOL(__crc32c_le); | 270 | { |
286 | EXPORT_SYMBOL(__crc32c_le_combine); | 271 | return crc32_generic_shift(crc, len, CRCPOLY_LE); |
272 | } | ||
273 | |||
274 | u32 __attribute_const__ __crc32c_le_shift(u32 crc, size_t len) | ||
275 | { | ||
276 | return crc32_generic_shift(crc, len, CRC32C_POLY_LE); | ||
277 | } | ||
278 | EXPORT_SYMBOL(crc32_le_shift); | ||
279 | EXPORT_SYMBOL(__crc32c_le_shift); | ||
287 | 280 | ||
288 | /** | 281 | /** |
289 | * crc32_be_generic() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32 | 282 | * crc32_be_generic() - Calculate bitwise big-endian Ethernet AUTODIN II CRC32 |
@@ -351,7 +344,7 @@ EXPORT_SYMBOL(crc32_be); | |||
351 | #ifdef CONFIG_CRC32_SELFTEST | 344 | #ifdef CONFIG_CRC32_SELFTEST |
352 | 345 | ||
353 | /* 4096 random bytes */ | 346 | /* 4096 random bytes */ |
354 | static u8 __attribute__((__aligned__(8))) test_buf[] = | 347 | static u8 const __aligned(8) test_buf[] __initconst = |
355 | { | 348 | { |
356 | 0x5b, 0x85, 0x21, 0xcb, 0x09, 0x68, 0x7d, 0x30, | 349 | 0x5b, 0x85, 0x21, 0xcb, 0x09, 0x68, 0x7d, 0x30, |
357 | 0xc7, 0x69, 0xd7, 0x30, 0x92, 0xde, 0x59, 0xe4, | 350 | 0xc7, 0x69, 0xd7, 0x30, 0x92, 0xde, 0x59, 0xe4, |
@@ -875,7 +868,7 @@ static struct crc_test { | |||
875 | u32 crc_le; /* expected crc32_le result */ | 868 | u32 crc_le; /* expected crc32_le result */ |
876 | u32 crc_be; /* expected crc32_be result */ | 869 | u32 crc_be; /* expected crc32_be result */ |
877 | u32 crc32c_le; /* expected crc32c_le result */ | 870 | u32 crc32c_le; /* expected crc32c_le result */ |
878 | } test[] = | 871 | } const test[] __initconst = |
879 | { | 872 | { |
880 | {0x674bf11d, 0x00000038, 0x00000542, 0x0af6d466, 0xd8b6e4c1, 0xf6e93d6c}, | 873 | {0x674bf11d, 0x00000038, 0x00000542, 0x0af6d466, 0xd8b6e4c1, 0xf6e93d6c}, |
881 | {0x35c672c6, 0x0000003a, 0x000001aa, 0xc6d3dfba, 0x28aaf3ad, 0x0fe92aca}, | 874 | {0x35c672c6, 0x0000003a, 0x000001aa, 0xc6d3dfba, 0x28aaf3ad, 0x0fe92aca}, |