1 /* Copyright (c) 2014, Google Inc.
2 *
3 * Permission to use, copy, modify, and/or distribute this software for any
4 * purpose with or without fee is hereby granted, provided that the above
5 * copyright notice and this permission notice appear in all copies.
6 *
7 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
8 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
9 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
10 * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
11 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
12 * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
13 * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
14
15 // This implementation was taken from the public domain, neon2 version in
16 // SUPERCOP by D. J. Bernstein and Peter Schwabe.
17
18 #include <openssl/poly1305.h>
19
20 #include <string.h>
21
22 #include "../internal.h"
23 #include "internal.h"
24
25
26 #if defined(OPENSSL_POLY1305_NEON)
27
28 typedef struct {
29 uint32_t v[12]; // for alignment; only using 10
30 } fe1305x2;
31
32 #define addmulmod openssl_poly1305_neon2_addmulmod
33 #define blocks openssl_poly1305_neon2_blocks
34
35 extern void addmulmod(fe1305x2 *r, const fe1305x2 *x, const fe1305x2 *y,
36 const fe1305x2 *c);
37
38 extern int blocks(fe1305x2 *h, const fe1305x2 *precomp, const uint8_t *in,
39 unsigned int inlen);
40
freeze(fe1305x2 * r)41 static void freeze(fe1305x2 *r) {
42 int i;
43
44 uint32_t x0 = r->v[0];
45 uint32_t x1 = r->v[2];
46 uint32_t x2 = r->v[4];
47 uint32_t x3 = r->v[6];
48 uint32_t x4 = r->v[8];
49 uint32_t y0;
50 uint32_t y1;
51 uint32_t y2;
52 uint32_t y3;
53 uint32_t y4;
54 uint32_t swap;
55
56 for (i = 0; i < 3; ++i) {
57 x1 += x0 >> 26;
58 x0 &= 0x3ffffff;
59 x2 += x1 >> 26;
60 x1 &= 0x3ffffff;
61 x3 += x2 >> 26;
62 x2 &= 0x3ffffff;
63 x4 += x3 >> 26;
64 x3 &= 0x3ffffff;
65 x0 += 5 * (x4 >> 26);
66 x4 &= 0x3ffffff;
67 }
68
69 y0 = x0 + 5;
70 y1 = x1 + (y0 >> 26);
71 y0 &= 0x3ffffff;
72 y2 = x2 + (y1 >> 26);
73 y1 &= 0x3ffffff;
74 y3 = x3 + (y2 >> 26);
75 y2 &= 0x3ffffff;
76 y4 = x4 + (y3 >> 26);
77 y3 &= 0x3ffffff;
78 swap = -(y4 >> 26);
79 y4 &= 0x3ffffff;
80
81 y0 ^= x0;
82 y1 ^= x1;
83 y2 ^= x2;
84 y3 ^= x3;
85 y4 ^= x4;
86
87 y0 &= swap;
88 y1 &= swap;
89 y2 &= swap;
90 y3 &= swap;
91 y4 &= swap;
92
93 y0 ^= x0;
94 y1 ^= x1;
95 y2 ^= x2;
96 y3 ^= x3;
97 y4 ^= x4;
98
99 r->v[0] = y0;
100 r->v[2] = y1;
101 r->v[4] = y2;
102 r->v[6] = y3;
103 r->v[8] = y4;
104 }
105
fe1305x2_tobytearray(uint8_t * r,fe1305x2 * x)106 static void fe1305x2_tobytearray(uint8_t *r, fe1305x2 *x) {
107 uint32_t x0 = x->v[0];
108 uint32_t x1 = x->v[2];
109 uint32_t x2 = x->v[4];
110 uint32_t x3 = x->v[6];
111 uint32_t x4 = x->v[8];
112
113 x1 += x0 >> 26;
114 x0 &= 0x3ffffff;
115 x2 += x1 >> 26;
116 x1 &= 0x3ffffff;
117 x3 += x2 >> 26;
118 x2 &= 0x3ffffff;
119 x4 += x3 >> 26;
120 x3 &= 0x3ffffff;
121
122 *(uint32_t *)r = x0 + (x1 << 26);
123 *(uint32_t *)(r + 4) = (x1 >> 6) + (x2 << 20);
124 *(uint32_t *)(r + 8) = (x2 >> 12) + (x3 << 14);
125 *(uint32_t *)(r + 12) = (x3 >> 18) + (x4 << 8);
126 }
127
128 // load32 exists to avoid breaking strict aliasing rules in
129 // fe1305x2_frombytearray.
load32(uint8_t * t)130 static uint32_t load32(uint8_t *t) {
131 uint32_t tmp;
132 OPENSSL_memcpy(&tmp, t, sizeof(tmp));
133 return tmp;
134 }
135
fe1305x2_frombytearray(fe1305x2 * r,const uint8_t * x,unsigned long long xlen)136 static void fe1305x2_frombytearray(fe1305x2 *r, const uint8_t *x,
137 unsigned long long xlen) {
138 unsigned i;
139 uint8_t t[17];
140
141 for (i = 0; (i < 16) && (i < xlen); i++) {
142 t[i] = x[i];
143 }
144 xlen -= i;
145 x += i;
146 t[i++] = 1;
147 for (; i < 17; i++) {
148 t[i] = 0;
149 }
150
151 r->v[0] = 0x3ffffff & load32(t);
152 r->v[2] = 0x3ffffff & (load32(t + 3) >> 2);
153 r->v[4] = 0x3ffffff & (load32(t + 6) >> 4);
154 r->v[6] = 0x3ffffff & (load32(t + 9) >> 6);
155 r->v[8] = load32(t + 13);
156
157 if (xlen) {
158 for (i = 0; (i < 16) && (i < xlen); i++) {
159 t[i] = x[i];
160 }
161 t[i++] = 1;
162 for (; i < 17; i++) {
163 t[i] = 0;
164 }
165
166 r->v[1] = 0x3ffffff & load32(t);
167 r->v[3] = 0x3ffffff & (load32(t + 3) >> 2);
168 r->v[5] = 0x3ffffff & (load32(t + 6) >> 4);
169 r->v[7] = 0x3ffffff & (load32(t + 9) >> 6);
170 r->v[9] = load32(t + 13);
171 } else {
172 r->v[1] = r->v[3] = r->v[5] = r->v[7] = r->v[9] = 0;
173 }
174 }
175
176 static const alignas(16) fe1305x2 zero;
177
178 struct poly1305_state_st {
179 uint8_t data[sizeof(fe1305x2[5]) + 128];
180 uint8_t buf[32];
181 unsigned int buf_used;
182 uint8_t key[16];
183 };
184
CRYPTO_poly1305_init_neon(poly1305_state * state,const uint8_t key[32])185 void CRYPTO_poly1305_init_neon(poly1305_state *state, const uint8_t key[32]) {
186 struct poly1305_state_st *st = (struct poly1305_state_st *)(state);
187 fe1305x2 *const r = (fe1305x2 *)(st->data + (15 & (-(int)st->data)));
188 fe1305x2 *const h = r + 1;
189 fe1305x2 *const c = h + 1;
190 fe1305x2 *const precomp = c + 1;
191 unsigned int j;
192
193 r->v[1] = r->v[0] = 0x3ffffff & *(uint32_t *)key;
194 r->v[3] = r->v[2] = 0x3ffff03 & ((*(uint32_t *)(key + 3)) >> 2);
195 r->v[5] = r->v[4] = 0x3ffc0ff & ((*(uint32_t *)(key + 6)) >> 4);
196 r->v[7] = r->v[6] = 0x3f03fff & ((*(uint32_t *)(key + 9)) >> 6);
197 r->v[9] = r->v[8] = 0x00fffff & ((*(uint32_t *)(key + 12)) >> 8);
198
199 for (j = 0; j < 10; j++) {
200 h->v[j] = 0; // XXX: should fast-forward a bit
201 }
202
203 addmulmod(precomp, r, r, &zero); // precompute r^2
204 addmulmod(precomp + 1, precomp, precomp, &zero); // precompute r^4
205
206 OPENSSL_memcpy(st->key, key + 16, 16);
207 st->buf_used = 0;
208 }
209
CRYPTO_poly1305_update_neon(poly1305_state * state,const uint8_t * in,size_t in_len)210 void CRYPTO_poly1305_update_neon(poly1305_state *state, const uint8_t *in,
211 size_t in_len) {
212 struct poly1305_state_st *st = (struct poly1305_state_st *)(state);
213 fe1305x2 *const r = (fe1305x2 *)(st->data + (15 & (-(int)st->data)));
214 fe1305x2 *const h = r + 1;
215 fe1305x2 *const c = h + 1;
216 fe1305x2 *const precomp = c + 1;
217 unsigned int i;
218
219 if (st->buf_used) {
220 unsigned int todo = 32 - st->buf_used;
221 if (todo > in_len) {
222 todo = in_len;
223 }
224 for (i = 0; i < todo; i++) {
225 st->buf[st->buf_used + i] = in[i];
226 }
227 st->buf_used += todo;
228 in_len -= todo;
229 in += todo;
230
231 if (st->buf_used == sizeof(st->buf) && in_len) {
232 addmulmod(h, h, precomp, &zero);
233 fe1305x2_frombytearray(c, st->buf, sizeof(st->buf));
234 for (i = 0; i < 10; i++) {
235 h->v[i] += c->v[i];
236 }
237 st->buf_used = 0;
238 }
239 }
240
241 while (in_len > 32) {
242 unsigned int tlen = 1048576;
243 if (in_len < tlen) {
244 tlen = in_len;
245 }
246 tlen -= blocks(h, precomp, in, tlen);
247 in_len -= tlen;
248 in += tlen;
249 }
250
251 if (in_len) {
252 for (i = 0; i < in_len; i++) {
253 st->buf[i] = in[i];
254 }
255 st->buf_used = in_len;
256 }
257 }
258
CRYPTO_poly1305_finish_neon(poly1305_state * state,uint8_t mac[16])259 void CRYPTO_poly1305_finish_neon(poly1305_state *state, uint8_t mac[16]) {
260 struct poly1305_state_st *st = (struct poly1305_state_st *)(state);
261 fe1305x2 *const r = (fe1305x2 *)(st->data + (15 & (-(int)st->data)));
262 fe1305x2 *const h = r + 1;
263 fe1305x2 *const c = h + 1;
264 fe1305x2 *const precomp = c + 1;
265
266 addmulmod(h, h, precomp, &zero);
267
268 if (st->buf_used > 16) {
269 fe1305x2_frombytearray(c, st->buf, st->buf_used);
270 precomp->v[1] = r->v[1];
271 precomp->v[3] = r->v[3];
272 precomp->v[5] = r->v[5];
273 precomp->v[7] = r->v[7];
274 precomp->v[9] = r->v[9];
275 addmulmod(h, h, precomp, c);
276 } else if (st->buf_used > 0) {
277 fe1305x2_frombytearray(c, st->buf, st->buf_used);
278 r->v[1] = 1;
279 r->v[3] = 0;
280 r->v[5] = 0;
281 r->v[7] = 0;
282 r->v[9] = 0;
283 addmulmod(h, h, r, c);
284 }
285
286 h->v[0] += h->v[1];
287 h->v[2] += h->v[3];
288 h->v[4] += h->v[5];
289 h->v[6] += h->v[7];
290 h->v[8] += h->v[9];
291 freeze(h);
292
293 fe1305x2_frombytearray(c, st->key, 16);
294 c->v[8] ^= (1 << 24);
295
296 h->v[0] += c->v[0];
297 h->v[2] += c->v[2];
298 h->v[4] += c->v[4];
299 h->v[6] += c->v[6];
300 h->v[8] += c->v[8];
301 fe1305x2_tobytearray(mac, h);
302 }
303
304 #endif // OPENSSL_POLY1305_NEON
305