OLD | NEW |
| (Empty) |
1 /* ==================================================================== | |
2 * Copyright (c) 2011-2013 The OpenSSL Project. All rights reserved. | |
3 * | |
4 * Redistribution and use in source and binary forms, with or without | |
5 * modification, are permitted provided that the following conditions | |
6 * are met: | |
7 * | |
8 * 1. Redistributions of source code must retain the above copyright | |
9 * notice, this list of conditions and the following disclaimer. | |
10 * | |
11 * 2. Redistributions in binary form must reproduce the above copyright | |
12 * notice, this list of conditions and the following disclaimer in | |
13 * the documentation and/or other materials provided with the | |
14 * distribution. | |
15 * | |
16 * 3. All advertising materials mentioning features or use of this | |
17 * software must display the following acknowledgment: | |
18 * "This product includes software developed by the OpenSSL Project | |
19 * for use in the OpenSSL Toolkit. (http://www.OpenSSL.org/)" | |
20 * | |
21 * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to | |
22 * endorse or promote products derived from this software without | |
23 * prior written permission. For written permission, please contact | |
24 * licensing@OpenSSL.org. | |
25 * | |
26 * 5. Products derived from this software may not be called "OpenSSL" | |
27 * nor may "OpenSSL" appear in their names without prior written | |
28 * permission of the OpenSSL Project. | |
29 * | |
30 * 6. Redistributions of any form whatsoever must retain the following | |
31 * acknowledgment: | |
32 * "This product includes software developed by the OpenSSL Project | |
33 * for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)" | |
34 * | |
35 * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY | |
36 * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | |
38 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR | |
39 * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
40 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT | |
41 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; | |
42 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
43 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, | |
44 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | |
45 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED | |
46 * OF THE POSSIBILITY OF SUCH DAMAGE. | |
47 * ==================================================================== | |
48 */ | |
49 | |
50 /* This implementation is by Ted Krovetz and was submitted to SUPERCOP and | |
51 * marked as public domain. It was been altered to allow for non-aligned inputs | |
52 * and to allow the block counter to be passed in specifically. */ | |
53 | |
54 #include <string.h> | |
55 #include <stdint.h> | |
56 #include <openssl/opensslconf.h> | |
57 | |
58 #if !defined(OPENSSL_NO_CHACHA) | |
59 | |
60 #include <openssl/chacha.h> | |
61 | |
62 #ifndef CHACHA_RNDS | |
63 #define CHACHA_RNDS 20 /* 8 (high speed), 20 (conservative), 12 (middle) */ | |
64 #endif | |
65 | |
66 /* Architecture-neutral way to specify 16-byte vector of ints */ | |
67 typedef unsigned vec __attribute__ ((vector_size (16))); | |
68 | |
69 /* This implementation is designed for Neon, SSE and AltiVec machines. The | |
70 * following specify how to do certain vector operations efficiently on | |
71 * each architecture, using intrinsics. | |
72 * This implementation supports parallel processing of multiple blocks, | |
73 * including potentially using general-purpose registers. | |
74 */ | |
75 #if __ARM_NEON__ | |
76 #include <arm_neon.h> | |
77 #define GPR_TOO 1 | |
78 #define VBPI 2 | |
79 #define ONE (vec)vsetq_lane_u32(1,vdupq_n_u32(0),0) | |
80 #define LOAD(m) (vec)(*((vec*)(m))) | |
81 #define STORE(m,r) (*((vec*)(m))) = (r) | |
82 #define ROTV1(x) (vec)vextq_u32((uint32x4_t)x,(uint32x4_t)x,1) | |
83 #define ROTV2(x) (vec)vextq_u32((uint32x4_t)x,(uint32x4_t)x,2) | |
84 #define ROTV3(x) (vec)vextq_u32((uint32x4_t)x,(uint32x4_t)x,3) | |
85 #define ROTW16(x) (vec)vrev32q_u16((uint16x8_t)x) | |
86 #if __clang__ | |
87 #define ROTW7(x) (x << ((vec){ 7, 7, 7, 7})) ^ (x >> ((vec){25,25,25,25})) | |
88 #define ROTW8(x) (x << ((vec){ 8, 8, 8, 8})) ^ (x >> ((vec){24,24,24,24})) | |
89 #define ROTW12(x) (x << ((vec){12,12,12,12})) ^ (x >> ((vec){20,20,20,20})) | |
90 #else | |
91 #define ROTW7(x) (vec)vsriq_n_u32(vshlq_n_u32((uint32x4_t)x,7),(uint32x4_t)x,25
) | |
92 #define ROTW8(x) (vec)vsriq_n_u32(vshlq_n_u32((uint32x4_t)x,8),(uint32x4_t)x,24
) | |
93 #define ROTW12(x) (vec)vsriq_n_u32(vshlq_n_u32((uint32x4_t)x,12),(uint32x4_t)x,2
0) | |
94 #endif | |
95 #elif __SSE2__ | |
96 #include <emmintrin.h> | |
97 #define GPR_TOO 0 | |
98 #if __clang__ | |
99 #define VBPI 4 | |
100 #else | |
101 #define VBPI 3 | |
102 #endif | |
103 #define ONE (vec)_mm_set_epi32(0,0,0,1) | |
104 #define LOAD(m) (vec)_mm_loadu_si128((__m128i*)(m)) | |
105 #define STORE(m,r) _mm_storeu_si128((__m128i*)(m), (__m128i) (r)) | |
106 #define ROTV1(x) (vec)_mm_shuffle_epi32((__m128i)x,_MM_SHUFFLE(0,3,2,1)) | |
107 #define ROTV2(x) (vec)_mm_shuffle_epi32((__m128i)x,_MM_SHUFFLE(1,0,3,2)) | |
108 #define ROTV3(x) (vec)_mm_shuffle_epi32((__m128i)x,_MM_SHUFFLE(2,1,0,3)) | |
109 #define ROTW7(x) (vec)(_mm_slli_epi32((__m128i)x, 7) ^ _mm_srli_epi32((__m128i)
x,25)) | |
110 #define ROTW12(x) (vec)(_mm_slli_epi32((__m128i)x,12) ^ _mm_srli_epi32((__m128i)
x,20)) | |
111 #if __SSSE3__ | |
112 #include <tmmintrin.h> | |
113 #define ROTW8(x) (vec)_mm_shuffle_epi8((__m128i)x,_mm_set_epi8(14,13,12,15,10,9
,8,11,6,5,4,7,2,1,0,3)) | |
114 #define ROTW16(x) (vec)_mm_shuffle_epi8((__m128i)x,_mm_set_epi8(13,12,15,14,9,8,
11,10,5,4,7,6,1,0,3,2)) | |
115 #else | |
116 #define ROTW8(x) (vec)(_mm_slli_epi32((__m128i)x, 8) ^ _mm_srli_epi32((__m128i)
x,24)) | |
117 #define ROTW16(x) (vec)(_mm_slli_epi32((__m128i)x,16) ^ _mm_srli_epi32((__m128i)
x,16)) | |
118 #endif | |
119 #else | |
120 #error -- Implementation supports only machines with neon or SSE2 | |
121 #endif | |
122 | |
123 #ifndef REVV_BE | |
124 #define REVV_BE(x) (x) | |
125 #endif | |
126 | |
127 #ifndef REVW_BE | |
128 #define REVW_BE(x) (x) | |
129 #endif | |
130 | |
131 #define BPI (VBPI + GPR_TOO) /* Blocks computed per loop iteration */ | |
132 | |
133 #define DQROUND_VECTORS(a,b,c,d) \ | |
134 a += b; d ^= a; d = ROTW16(d); \ | |
135 c += d; b ^= c; b = ROTW12(b); \ | |
136 a += b; d ^= a; d = ROTW8(d); \ | |
137 c += d; b ^= c; b = ROTW7(b); \ | |
138 b = ROTV1(b); c = ROTV2(c); d = ROTV3(d); \ | |
139 a += b; d ^= a; d = ROTW16(d); \ | |
140 c += d; b ^= c; b = ROTW12(b); \ | |
141 a += b; d ^= a; d = ROTW8(d); \ | |
142 c += d; b ^= c; b = ROTW7(b); \ | |
143 b = ROTV3(b); c = ROTV2(c); d = ROTV1(d); | |
144 | |
145 #define QROUND_WORDS(a,b,c,d) \ | |
146 a = a+b; d ^= a; d = d<<16 | d>>16; \ | |
147 c = c+d; b ^= c; b = b<<12 | b>>20; \ | |
148 a = a+b; d ^= a; d = d<< 8 | d>>24; \ | |
149 c = c+d; b ^= c; b = b<< 7 | b>>25; | |
150 | |
151 #define WRITE_XOR(in, op, d, v0, v1, v2, v3) \ | |
152 STORE(op + d + 0, LOAD(in + d + 0) ^ REVV_BE(v0)); \ | |
153 STORE(op + d + 4, LOAD(in + d + 4) ^ REVV_BE(v1)); \ | |
154 STORE(op + d + 8, LOAD(in + d + 8) ^ REVV_BE(v2)); \ | |
155 STORE(op + d +12, LOAD(in + d +12) ^ REVV_BE(v3)); | |
156 | |
157 #if __ARM_NEON__ | |
158 /* For ARM, we can't depend on NEON support, so this function is compiled with | |
159 * a different name, along with the generic code, and can be enabled at | |
160 * run-time. */ | |
161 void CRYPTO_chacha_20_neon( | |
162 #else | |
163 void CRYPTO_chacha_20( | |
164 #endif | |
165 unsigned char *out, | |
166 const unsigned char *in, | |
167 size_t inlen, | |
168 const unsigned char key[32], | |
169 const unsigned char nonce[8], | |
170 size_t counter) | |
171 { | |
172 unsigned iters, i, *op=(unsigned *)out, *ip=(unsigned *)in, *kp; | |
173 #if defined(__ARM_NEON__) | |
174 unsigned *np; | |
175 #endif | |
176 vec s0, s1, s2, s3; | |
177 #if !defined(__ARM_NEON__) && !defined(__SSE2__) | |
178 __attribute__ ((aligned (16))) unsigned key[8], nonce[4]; | |
179 #endif | |
180 __attribute__ ((aligned (16))) unsigned chacha_const[] = | |
181 {0x61707865,0x3320646E,0x79622D32,0x6B206574}; | |
182 #if defined(__ARM_NEON__) || defined(__SSE2__) | |
183 kp = (unsigned *)key; | |
184 #else | |
185 ((vec *)key)[0] = REVV_BE(((vec *)key)[0]); | |
186 ((vec *)key)[1] = REVV_BE(((vec *)key)[1]); | |
187 nonce[0] = REVW_BE(((unsigned *)nonce)[0]); | |
188 nonce[1] = REVW_BE(((unsigned *)nonce)[1]); | |
189 nonce[2] = REVW_BE(((unsigned *)nonce)[2]); | |
190 nonce[3] = REVW_BE(((unsigned *)nonce)[3]); | |
191 kp = (unsigned *)key; | |
192 np = (unsigned *)nonce; | |
193 #endif | |
194 #if defined(__ARM_NEON__) | |
195 np = (unsigned*) nonce; | |
196 #endif | |
197 s0 = LOAD(chacha_const); | |
198 s1 = LOAD(&((vec*)kp)[0]); | |
199 s2 = LOAD(&((vec*)kp)[1]); | |
200 s3 = (vec){ | |
201 counter & 0xffffffff, | |
202 #if __ARM_NEON__ | |
203 0, /* can't right-shift 32 bits on a 32-bit system. */ | |
204 #else | |
205 counter >> 32, | |
206 #endif | |
207 ((uint32_t*)nonce)[0], | |
208 ((uint32_t*)nonce)[1] | |
209 }; | |
210 | |
211 for (iters = 0; iters < inlen/(BPI*64); iters++) | |
212 { | |
213 #if GPR_TOO | |
214 register unsigned x0, x1, x2, x3, x4, x5, x6, x7, x8, | |
215 x9, x10, x11, x12, x13, x14, x15; | |
216 #endif | |
217 #if VBPI > 2 | |
218 vec v8,v9,v10,v11; | |
219 #endif | |
220 #if VBPI > 3 | |
221 vec v12,v13,v14,v15; | |
222 #endif | |
223 | |
224 vec v0,v1,v2,v3,v4,v5,v6,v7; | |
225 v4 = v0 = s0; v5 = v1 = s1; v6 = v2 = s2; v3 = s3; | |
226 v7 = v3 + ONE; | |
227 #if VBPI > 2 | |
228 v8 = v4; v9 = v5; v10 = v6; | |
229 v11 = v7 + ONE; | |
230 #endif | |
231 #if VBPI > 3 | |
232 v12 = v8; v13 = v9; v14 = v10; | |
233 v15 = v11 + ONE; | |
234 #endif | |
235 #if GPR_TOO | |
236 x0 = chacha_const[0]; x1 = chacha_const[1]; | |
237 x2 = chacha_const[2]; x3 = chacha_const[3]; | |
238 x4 = kp[0]; x5 = kp[1]; x6 = kp[2]; x7 = kp[3]; | |
239 x8 = kp[4]; x9 = kp[5]; x10 = kp[6]; x11 = kp[7]; | |
240 x12 = counter+BPI*iters+(BPI-1); x13 = 0; | |
241 x14 = np[0]; x15 = np[1]; | |
242 #endif | |
243 for (i = CHACHA_RNDS/2; i; i--) | |
244 { | |
245 DQROUND_VECTORS(v0,v1,v2,v3) | |
246 DQROUND_VECTORS(v4,v5,v6,v7) | |
247 #if VBPI > 2 | |
248 DQROUND_VECTORS(v8,v9,v10,v11) | |
249 #endif | |
250 #if VBPI > 3 | |
251 DQROUND_VECTORS(v12,v13,v14,v15) | |
252 #endif | |
253 #if GPR_TOO | |
254 QROUND_WORDS( x0, x4, x8,x12) | |
255 QROUND_WORDS( x1, x5, x9,x13) | |
256 QROUND_WORDS( x2, x6,x10,x14) | |
257 QROUND_WORDS( x3, x7,x11,x15) | |
258 QROUND_WORDS( x0, x5,x10,x15) | |
259 QROUND_WORDS( x1, x6,x11,x12) | |
260 QROUND_WORDS( x2, x7, x8,x13) | |
261 QROUND_WORDS( x3, x4, x9,x14) | |
262 #endif | |
263 } | |
264 | |
265 WRITE_XOR(ip, op, 0, v0+s0, v1+s1, v2+s2, v3+s3) | |
266 s3 += ONE; | |
267 WRITE_XOR(ip, op, 16, v4+s0, v5+s1, v6+s2, v7+s3) | |
268 s3 += ONE; | |
269 #if VBPI > 2 | |
270 WRITE_XOR(ip, op, 32, v8+s0, v9+s1, v10+s2, v11+s3) | |
271 s3 += ONE; | |
272 #endif | |
273 #if VBPI > 3 | |
274 WRITE_XOR(ip, op, 48, v12+s0, v13+s1, v14+s2, v15+s3) | |
275 s3 += ONE; | |
276 #endif | |
277 ip += VBPI*16; | |
278 op += VBPI*16; | |
279 #if GPR_TOO | |
280 op[0] = REVW_BE(REVW_BE(ip[0]) ^ (x0 + chacha_const[0])); | |
281 op[1] = REVW_BE(REVW_BE(ip[1]) ^ (x1 + chacha_const[1])); | |
282 op[2] = REVW_BE(REVW_BE(ip[2]) ^ (x2 + chacha_const[2])); | |
283 op[3] = REVW_BE(REVW_BE(ip[3]) ^ (x3 + chacha_const[3])); | |
284 op[4] = REVW_BE(REVW_BE(ip[4]) ^ (x4 + kp[0])); | |
285 op[5] = REVW_BE(REVW_BE(ip[5]) ^ (x5 + kp[1])); | |
286 op[6] = REVW_BE(REVW_BE(ip[6]) ^ (x6 + kp[2])); | |
287 op[7] = REVW_BE(REVW_BE(ip[7]) ^ (x7 + kp[3])); | |
288 op[8] = REVW_BE(REVW_BE(ip[8]) ^ (x8 + kp[4])); | |
289 op[9] = REVW_BE(REVW_BE(ip[9]) ^ (x9 + kp[5])); | |
290 op[10] = REVW_BE(REVW_BE(ip[10]) ^ (x10 + kp[6])); | |
291 op[11] = REVW_BE(REVW_BE(ip[11]) ^ (x11 + kp[7])); | |
292 op[12] = REVW_BE(REVW_BE(ip[12]) ^ (x12 + counter+BPI*iters+(BPI
-1))); | |
293 op[13] = REVW_BE(REVW_BE(ip[13]) ^ (x13)); | |
294 op[14] = REVW_BE(REVW_BE(ip[14]) ^ (x14 + np[0])); | |
295 op[15] = REVW_BE(REVW_BE(ip[15]) ^ (x15 + np[1])); | |
296 s3 += ONE; | |
297 ip += 16; | |
298 op += 16; | |
299 #endif | |
300 } | |
301 | |
302 for (iters = inlen%(BPI*64)/64; iters != 0; iters--) | |
303 { | |
304 vec v0 = s0, v1 = s1, v2 = s2, v3 = s3; | |
305 for (i = CHACHA_RNDS/2; i; i--) | |
306 { | |
307 DQROUND_VECTORS(v0,v1,v2,v3); | |
308 } | |
309 WRITE_XOR(ip, op, 0, v0+s0, v1+s1, v2+s2, v3+s3) | |
310 s3 += ONE; | |
311 ip += 16; | |
312 op += 16; | |
313 } | |
314 | |
315 inlen = inlen % 64; | |
316 if (inlen) | |
317 { | |
318 __attribute__ ((aligned (16))) vec buf[4]; | |
319 vec v0,v1,v2,v3; | |
320 v0 = s0; v1 = s1; v2 = s2; v3 = s3; | |
321 for (i = CHACHA_RNDS/2; i; i--) | |
322 { | |
323 DQROUND_VECTORS(v0,v1,v2,v3); | |
324 } | |
325 | |
326 if (inlen >= 16) | |
327 { | |
328 STORE(op + 0, LOAD(ip + 0) ^ REVV_BE(v0 + s0)); | |
329 if (inlen >= 32) | |
330 { | |
331 STORE(op + 4, LOAD(ip + 4) ^ REVV_BE(v1 + s1)); | |
332 if (inlen >= 48) | |
333 { | |
334 STORE(op + 8, LOAD(ip + 8) ^ | |
335 REVV_BE(v2 + s2)); | |
336 buf[3] = REVV_BE(v3 + s3); | |
337 } | |
338 else | |
339 buf[2] = REVV_BE(v2 + s2); | |
340 } | |
341 else | |
342 buf[1] = REVV_BE(v1 + s1); | |
343 } | |
344 else | |
345 buf[0] = REVV_BE(v0 + s0); | |
346 | |
347 for (i=inlen & ~15; i<inlen; i++) | |
348 ((char *)op)[i] = ((char *)ip)[i] ^ ((char *)buf)[i]; | |
349 } | |
350 } | |
351 | |
352 #endif /* !OPENSSL_NO_CHACHA */ | |
OLD | NEW |