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#if defined(OPENSSL_ARM) && !defined(OPENSSL_NO_ASM)
21
22#include <string.h>
23
24
25typedef struct {
26  uint32_t v[12]; /* for alignment; only using 10 */
27} fe1305x2;
28
29#define addmulmod openssl_poly1305_neon2_addmulmod
30#define blocks openssl_poly1305_neon2_blocks
31
32extern void addmulmod(fe1305x2 *r, const fe1305x2 *x, const fe1305x2 *y,
33                      const fe1305x2 *c);
34
35extern int blocks(fe1305x2 *h, const fe1305x2 *precomp, const uint8_t *in,
36                  unsigned int inlen);
37
38static void freeze(fe1305x2 *r) {
39  int i;
40
41  uint32_t x0 = r->v[0];
42  uint32_t x1 = r->v[2];
43  uint32_t x2 = r->v[4];
44  uint32_t x3 = r->v[6];
45  uint32_t x4 = r->v[8];
46  uint32_t y0;
47  uint32_t y1;
48  uint32_t y2;
49  uint32_t y3;
50  uint32_t y4;
51  uint32_t swap;
52
53  for (i = 0; i < 3; ++i) {
54    x1 += x0 >> 26;
55    x0 &= 0x3ffffff;
56    x2 += x1 >> 26;
57    x1 &= 0x3ffffff;
58    x3 += x2 >> 26;
59    x2 &= 0x3ffffff;
60    x4 += x3 >> 26;
61    x3 &= 0x3ffffff;
62    x0 += 5 * (x4 >> 26);
63    x4 &= 0x3ffffff;
64  }
65
66  y0 = x0 + 5;
67  y1 = x1 + (y0 >> 26);
68  y0 &= 0x3ffffff;
69  y2 = x2 + (y1 >> 26);
70  y1 &= 0x3ffffff;
71  y3 = x3 + (y2 >> 26);
72  y2 &= 0x3ffffff;
73  y4 = x4 + (y3 >> 26);
74  y3 &= 0x3ffffff;
75  swap = -(y4 >> 26);
76  y4 &= 0x3ffffff;
77
78  y0 ^= x0;
79  y1 ^= x1;
80  y2 ^= x2;
81  y3 ^= x3;
82  y4 ^= x4;
83
84  y0 &= swap;
85  y1 &= swap;
86  y2 &= swap;
87  y3 &= swap;
88  y4 &= swap;
89
90  y0 ^= x0;
91  y1 ^= x1;
92  y2 ^= x2;
93  y3 ^= x3;
94  y4 ^= x4;
95
96  r->v[0] = y0;
97  r->v[2] = y1;
98  r->v[4] = y2;
99  r->v[6] = y3;
100  r->v[8] = y4;
101}
102
103static void fe1305x2_tobytearray(uint8_t *r, fe1305x2 *x) {
104  uint32_t x0 = x->v[0];
105  uint32_t x1 = x->v[2];
106  uint32_t x2 = x->v[4];
107  uint32_t x3 = x->v[6];
108  uint32_t x4 = x->v[8];
109
110  x1 += x0 >> 26;
111  x0 &= 0x3ffffff;
112  x2 += x1 >> 26;
113  x1 &= 0x3ffffff;
114  x3 += x2 >> 26;
115  x2 &= 0x3ffffff;
116  x4 += x3 >> 26;
117  x3 &= 0x3ffffff;
118
119  *(uint32_t *)r = x0 + (x1 << 26);
120  *(uint32_t *)(r + 4) = (x1 >> 6) + (x2 << 20);
121  *(uint32_t *)(r + 8) = (x2 >> 12) + (x3 << 14);
122  *(uint32_t *)(r + 12) = (x3 >> 18) + (x4 << 8);
123}
124
125/* load32 exists to avoid breaking strict aliasing rules in
126 * fe1305x2_frombytearray. */
127static uint32_t load32(uint8_t *t) {
128  uint32_t tmp;
129  memcpy(&tmp, t, sizeof(tmp));
130  return tmp;
131}
132
133static void fe1305x2_frombytearray(fe1305x2 *r, const uint8_t *x,
134                                   unsigned long long xlen) {
135  int i;
136  uint8_t t[17];
137
138  for (i = 0; (i < 16) && (i < xlen); i++) {
139    t[i] = x[i];
140  }
141  xlen -= i;
142  x += i;
143  t[i++] = 1;
144  for (; i < 17; i++) {
145    t[i] = 0;
146  }
147
148  r->v[0] = 0x3ffffff & load32(t);
149  r->v[2] = 0x3ffffff & (load32(t + 3) >> 2);
150  r->v[4] = 0x3ffffff & (load32(t + 6) >> 4);
151  r->v[6] = 0x3ffffff & (load32(t + 9) >> 6);
152  r->v[8] = load32(t + 13);
153
154  if (xlen) {
155    for (i = 0; (i < 16) && (i < xlen); i++) {
156      t[i] = x[i];
157    }
158    t[i++] = 1;
159    for (; i < 17; i++) {
160      t[i] = 0;
161    }
162
163    r->v[1] = 0x3ffffff & load32(t);
164    r->v[3] = 0x3ffffff & (load32(t + 3) >> 2);
165    r->v[5] = 0x3ffffff & (load32(t + 6) >> 4);
166    r->v[7] = 0x3ffffff & (load32(t + 9) >> 6);
167    r->v[9] = load32(t + 13);
168  } else {
169    r->v[1] = r->v[3] = r->v[5] = r->v[7] = r->v[9] = 0;
170  }
171}
172
173static const fe1305x2 zero __attribute__((aligned(16)));
174
175struct poly1305_state_st {
176  uint8_t data[sizeof(fe1305x2[5]) + 128];
177  uint8_t buf[32];
178  unsigned int buf_used;
179  uint8_t key[16];
180};
181
182void CRYPTO_poly1305_init_neon(poly1305_state *state, const uint8_t key[32]) {
183  struct poly1305_state_st *st = (struct poly1305_state_st *)(state);
184  fe1305x2 *const r = (fe1305x2 *)(st->data + (15 & (-(int)st->data)));
185  fe1305x2 *const h = r + 1;
186  fe1305x2 *const c = h + 1;
187  fe1305x2 *const precomp = c + 1;
188  unsigned int j;
189
190  r->v[1] = r->v[0] = 0x3ffffff & *(uint32_t *)key;
191  r->v[3] = r->v[2] = 0x3ffff03 & ((*(uint32_t *)(key + 3)) >> 2);
192  r->v[5] = r->v[4] = 0x3ffc0ff & ((*(uint32_t *)(key + 6)) >> 4);
193  r->v[7] = r->v[6] = 0x3f03fff & ((*(uint32_t *)(key + 9)) >> 6);
194  r->v[9] = r->v[8] = 0x00fffff & ((*(uint32_t *)(key + 12)) >> 8);
195
196  for (j = 0; j < 10; j++) {
197    h->v[j] = 0; /* XXX: should fast-forward a bit */
198  }
199
200  addmulmod(precomp, r, r, &zero);                 /* precompute r^2 */
201  addmulmod(precomp + 1, precomp, precomp, &zero); /* precompute r^4 */
202
203  memcpy(st->key, key + 16, 16);
204  st->buf_used = 0;
205}
206
207void CRYPTO_poly1305_update_neon(poly1305_state *state, const uint8_t *in,
208                                 size_t in_len) {
209  struct poly1305_state_st *st = (struct poly1305_state_st *)(state);
210  fe1305x2 *const r = (fe1305x2 *)(st->data + (15 & (-(int)st->data)));
211  fe1305x2 *const h = r + 1;
212  fe1305x2 *const c = h + 1;
213  fe1305x2 *const precomp = c + 1;
214  unsigned int i;
215
216  if (st->buf_used) {
217    unsigned int todo = 32 - st->buf_used;
218    if (todo > in_len) {
219      todo = in_len;
220    }
221    for (i = 0; i < todo; i++) {
222      st->buf[st->buf_used + i] = in[i];
223    }
224    st->buf_used += todo;
225    in_len -= todo;
226    in += todo;
227
228    if (st->buf_used == sizeof(st->buf) && in_len) {
229      addmulmod(h, h, precomp, &zero);
230      fe1305x2_frombytearray(c, st->buf, sizeof(st->buf));
231      for (i = 0; i < 10; i++) {
232        h->v[i] += c->v[i];
233      }
234      st->buf_used = 0;
235    }
236  }
237
238  while (in_len > 32) {
239    unsigned int tlen = 1048576;
240    if (in_len < tlen) {
241      tlen = in_len;
242    }
243    tlen -= blocks(h, precomp, in, tlen);
244    in_len -= tlen;
245    in += tlen;
246  }
247
248  if (in_len) {
249    for (i = 0; i < in_len; i++) {
250      st->buf[i] = in[i];
251    }
252    st->buf_used = in_len;
253  }
254}
255
256void CRYPTO_poly1305_finish_neon(poly1305_state *state, uint8_t mac[16]) {
257  struct poly1305_state_st *st = (struct poly1305_state_st *)(state);
258  fe1305x2 *const r = (fe1305x2 *)(st->data + (15 & (-(int)st->data)));
259  fe1305x2 *const h = r + 1;
260  fe1305x2 *const c = h + 1;
261  fe1305x2 *const precomp = c + 1;
262
263  addmulmod(h, h, precomp, &zero);
264
265  if (st->buf_used > 16) {
266    fe1305x2_frombytearray(c, st->buf, st->buf_used);
267    precomp->v[1] = r->v[1];
268    precomp->v[3] = r->v[3];
269    precomp->v[5] = r->v[5];
270    precomp->v[7] = r->v[7];
271    precomp->v[9] = r->v[9];
272    addmulmod(h, h, precomp, c);
273  } else if (st->buf_used > 0) {
274    fe1305x2_frombytearray(c, st->buf, st->buf_used);
275    r->v[1] = 1;
276    r->v[3] = 0;
277    r->v[5] = 0;
278    r->v[7] = 0;
279    r->v[9] = 0;
280    addmulmod(h, h, r, c);
281  }
282
283  h->v[0] += h->v[1];
284  h->v[2] += h->v[3];
285  h->v[4] += h->v[5];
286  h->v[6] += h->v[7];
287  h->v[8] += h->v[9];
288  freeze(h);
289
290  fe1305x2_frombytearray(c, st->key, 16);
291  c->v[8] ^= (1 << 24);
292
293  h->v[0] += c->v[0];
294  h->v[2] += c->v[2];
295  h->v[4] += c->v[4];
296  h->v[6] += c->v[6];
297  h->v[8] += c->v[8];
298  fe1305x2_tobytearray(mac, h);
299}
300
301#endif  /* OPENSSL_ARM && !OPENSSL_NO_ASM */
302