1 /* Copyright 2014 The BoringSSL Authors
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 of poly1305 is by Andrew Moon
16 // (https://github.com/floodyberry/poly1305-donna) and released as public
17 // domain. It implements SIMD vectorization based on the algorithm described in
18 // http://cr.yp.to/papers.html#neoncrypto. Unrolled to 2 powers, i.e. 64 byte
19 // block size
20
21 #include <openssl/poly1305.h>
22
23 #include <assert.h>
24
25 #include "../internal.h"
26
27
28 #if defined(BORINGSSL_HAS_UINT128) && defined(OPENSSL_X86_64)
29
30 #include <emmintrin.h>
31
32 typedef __m128i xmmi;
33
34 alignas(16) static const uint32_t poly1305_x64_sse2_message_mask[4] = {
35 (1 << 26) - 1, 0, (1 << 26) - 1, 0};
36 alignas(16) static const uint32_t poly1305_x64_sse2_5[4] = {5, 0, 5, 0};
37 alignas(16) static const uint32_t poly1305_x64_sse2_1shl128[4] = {(1 << 24), 0,
38 (1 << 24), 0};
39
add128(uint128_t a,uint128_t b)40 static inline uint128_t add128(uint128_t a, uint128_t b) { return a + b; }
41
add128_64(uint128_t a,uint64_t b)42 static inline uint128_t add128_64(uint128_t a, uint64_t b) { return a + b; }
43
mul64x64_128(uint64_t a,uint64_t b)44 static inline uint128_t mul64x64_128(uint64_t a, uint64_t b) {
45 return (uint128_t)a * b;
46 }
47
lo128(uint128_t a)48 static inline uint64_t lo128(uint128_t a) { return (uint64_t)a; }
49
shr128(uint128_t v,const int shift)50 static inline uint64_t shr128(uint128_t v, const int shift) {
51 return (uint64_t)(v >> shift);
52 }
53
shr128_pair(uint64_t hi,uint64_t lo,const int shift)54 static inline uint64_t shr128_pair(uint64_t hi, uint64_t lo, const int shift) {
55 return (uint64_t)((((uint128_t)hi << 64) | lo) >> shift);
56 }
57
58 typedef struct poly1305_power_t {
59 union {
60 xmmi v;
61 uint64_t u[2];
62 uint32_t d[4];
63 } R20, R21, R22, R23, R24, S21, S22, S23, S24;
64 } poly1305_power;
65
66 typedef struct poly1305_state_internal_t {
67 poly1305_power P[2]; /* 288 bytes, top 32 bit halves unused = 144
68 bytes of free storage */
69 union {
70 xmmi H[5]; // 80 bytes
71 uint64_t HH[10];
72 };
73 // uint64_t r0,r1,r2; [24 bytes]
74 // uint64_t pad0,pad1; [16 bytes]
75 uint64_t started; // 8 bytes
76 uint64_t leftover; // 8 bytes
77 uint8_t buffer[64]; // 64 bytes
78 } poly1305_state_internal; /* 448 bytes total + 63 bytes for
79 alignment = 511 bytes raw */
80
81 static_assert(sizeof(struct poly1305_state_internal_t) + 63 <=
82 sizeof(poly1305_state),
83 "poly1305_state isn't large enough to hold aligned "
84 "poly1305_state_internal_t");
85
poly1305_aligned_state(poly1305_state * state)86 static inline poly1305_state_internal *poly1305_aligned_state(
87 poly1305_state *state) {
88 return (poly1305_state_internal *)(((uint64_t)state + 63) & ~63);
89 }
90
poly1305_min(size_t a,size_t b)91 static inline size_t poly1305_min(size_t a, size_t b) {
92 return (a < b) ? a : b;
93 }
94
CRYPTO_poly1305_init(poly1305_state * state,const uint8_t key[32])95 void CRYPTO_poly1305_init(poly1305_state *state, const uint8_t key[32]) {
96 poly1305_state_internal *st = poly1305_aligned_state(state);
97 poly1305_power *p;
98 uint64_t r0, r1, r2;
99 uint64_t t0, t1;
100
101 // clamp key
102 t0 = CRYPTO_load_u64_le(key + 0);
103 t1 = CRYPTO_load_u64_le(key + 8);
104 r0 = t0 & 0xffc0fffffff;
105 t0 >>= 44;
106 t0 |= t1 << 20;
107 r1 = t0 & 0xfffffc0ffff;
108 t1 >>= 24;
109 r2 = t1 & 0x00ffffffc0f;
110
111 // store r in un-used space of st->P[1]
112 p = &st->P[1];
113 p->R20.d[1] = (uint32_t)(r0);
114 p->R20.d[3] = (uint32_t)(r0 >> 32);
115 p->R21.d[1] = (uint32_t)(r1);
116 p->R21.d[3] = (uint32_t)(r1 >> 32);
117 p->R22.d[1] = (uint32_t)(r2);
118 p->R22.d[3] = (uint32_t)(r2 >> 32);
119
120 // store pad
121 p->R23.d[1] = CRYPTO_load_u32_le(key + 16);
122 p->R23.d[3] = CRYPTO_load_u32_le(key + 20);
123 p->R24.d[1] = CRYPTO_load_u32_le(key + 24);
124 p->R24.d[3] = CRYPTO_load_u32_le(key + 28);
125
126 // H = 0
127 st->H[0] = _mm_setzero_si128();
128 st->H[1] = _mm_setzero_si128();
129 st->H[2] = _mm_setzero_si128();
130 st->H[3] = _mm_setzero_si128();
131 st->H[4] = _mm_setzero_si128();
132
133 st->started = 0;
134 st->leftover = 0;
135 }
136
poly1305_first_block(poly1305_state_internal * st,const uint8_t * m)137 static void poly1305_first_block(poly1305_state_internal *st,
138 const uint8_t *m) {
139 const xmmi MMASK =
140 _mm_load_si128((const xmmi *)poly1305_x64_sse2_message_mask);
141 const xmmi FIVE = _mm_load_si128((const xmmi *)poly1305_x64_sse2_5);
142 const xmmi HIBIT = _mm_load_si128((const xmmi *)poly1305_x64_sse2_1shl128);
143 xmmi T5, T6;
144 poly1305_power *p;
145 uint128_t d[3];
146 uint64_t r0, r1, r2;
147 uint64_t r20, r21, r22, s22;
148 uint64_t pad0, pad1;
149 uint64_t c;
150 uint64_t i;
151
152 // pull out stored info
153 p = &st->P[1];
154
155 r0 = ((uint64_t)p->R20.d[3] << 32) | (uint64_t)p->R20.d[1];
156 r1 = ((uint64_t)p->R21.d[3] << 32) | (uint64_t)p->R21.d[1];
157 r2 = ((uint64_t)p->R22.d[3] << 32) | (uint64_t)p->R22.d[1];
158 pad0 = ((uint64_t)p->R23.d[3] << 32) | (uint64_t)p->R23.d[1];
159 pad1 = ((uint64_t)p->R24.d[3] << 32) | (uint64_t)p->R24.d[1];
160
161 // compute powers r^2,r^4
162 r20 = r0;
163 r21 = r1;
164 r22 = r2;
165 for (i = 0; i < 2; i++) {
166 s22 = r22 * (5 << 2);
167
168 d[0] = add128(mul64x64_128(r20, r20), mul64x64_128(r21 * 2, s22));
169 d[1] = add128(mul64x64_128(r22, s22), mul64x64_128(r20 * 2, r21));
170 d[2] = add128(mul64x64_128(r21, r21), mul64x64_128(r22 * 2, r20));
171
172 r20 = lo128(d[0]) & 0xfffffffffff;
173 c = shr128(d[0], 44);
174 d[1] = add128_64(d[1], c);
175 r21 = lo128(d[1]) & 0xfffffffffff;
176 c = shr128(d[1], 44);
177 d[2] = add128_64(d[2], c);
178 r22 = lo128(d[2]) & 0x3ffffffffff;
179 c = shr128(d[2], 42);
180 r20 += c * 5;
181 c = (r20 >> 44);
182 r20 = r20 & 0xfffffffffff;
183 r21 += c;
184
185 p->R20.v = _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)(r20) & 0x3ffffff),
186 _MM_SHUFFLE(1, 0, 1, 0));
187 p->R21.v = _mm_shuffle_epi32(
188 _mm_cvtsi32_si128((uint32_t)((r20 >> 26) | (r21 << 18)) & 0x3ffffff),
189 _MM_SHUFFLE(1, 0, 1, 0));
190 p->R22.v =
191 _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)((r21 >> 8)) & 0x3ffffff),
192 _MM_SHUFFLE(1, 0, 1, 0));
193 p->R23.v = _mm_shuffle_epi32(
194 _mm_cvtsi32_si128((uint32_t)((r21 >> 34) | (r22 << 10)) & 0x3ffffff),
195 _MM_SHUFFLE(1, 0, 1, 0));
196 p->R24.v = _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)((r22 >> 16))),
197 _MM_SHUFFLE(1, 0, 1, 0));
198 p->S21.v = _mm_mul_epu32(p->R21.v, FIVE);
199 p->S22.v = _mm_mul_epu32(p->R22.v, FIVE);
200 p->S23.v = _mm_mul_epu32(p->R23.v, FIVE);
201 p->S24.v = _mm_mul_epu32(p->R24.v, FIVE);
202 p--;
203 }
204
205 // put saved info back
206 p = &st->P[1];
207 p->R20.d[1] = (uint32_t)(r0);
208 p->R20.d[3] = (uint32_t)(r0 >> 32);
209 p->R21.d[1] = (uint32_t)(r1);
210 p->R21.d[3] = (uint32_t)(r1 >> 32);
211 p->R22.d[1] = (uint32_t)(r2);
212 p->R22.d[3] = (uint32_t)(r2 >> 32);
213 p->R23.d[1] = (uint32_t)(pad0);
214 p->R23.d[3] = (uint32_t)(pad0 >> 32);
215 p->R24.d[1] = (uint32_t)(pad1);
216 p->R24.d[3] = (uint32_t)(pad1 >> 32);
217
218 // H = [Mx,My]
219 T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((const xmmi *)(m + 0)),
220 _mm_loadl_epi64((const xmmi *)(m + 16)));
221 T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((const xmmi *)(m + 8)),
222 _mm_loadl_epi64((const xmmi *)(m + 24)));
223 st->H[0] = _mm_and_si128(MMASK, T5);
224 st->H[1] = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
225 T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12));
226 st->H[2] = _mm_and_si128(MMASK, T5);
227 st->H[3] = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
228 st->H[4] = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT);
229 }
230
poly1305_blocks(poly1305_state_internal * st,const uint8_t * m,size_t bytes)231 static void poly1305_blocks(poly1305_state_internal *st, const uint8_t *m,
232 size_t bytes) {
233 const xmmi MMASK =
234 _mm_load_si128((const xmmi *)poly1305_x64_sse2_message_mask);
235 const xmmi FIVE = _mm_load_si128((const xmmi *)poly1305_x64_sse2_5);
236 const xmmi HIBIT = _mm_load_si128((const xmmi *)poly1305_x64_sse2_1shl128);
237
238 poly1305_power *p;
239 xmmi H0, H1, H2, H3, H4;
240 xmmi T0, T1, T2, T3, T4, T5, T6;
241 xmmi M0, M1, M2, M3, M4;
242 xmmi C1, C2;
243
244 H0 = st->H[0];
245 H1 = st->H[1];
246 H2 = st->H[2];
247 H3 = st->H[3];
248 H4 = st->H[4];
249
250 while (bytes >= 64) {
251 // H *= [r^4,r^4]
252 p = &st->P[0];
253 T0 = _mm_mul_epu32(H0, p->R20.v);
254 T1 = _mm_mul_epu32(H0, p->R21.v);
255 T2 = _mm_mul_epu32(H0, p->R22.v);
256 T3 = _mm_mul_epu32(H0, p->R23.v);
257 T4 = _mm_mul_epu32(H0, p->R24.v);
258 T5 = _mm_mul_epu32(H1, p->S24.v);
259 T6 = _mm_mul_epu32(H1, p->R20.v);
260 T0 = _mm_add_epi64(T0, T5);
261 T1 = _mm_add_epi64(T1, T6);
262 T5 = _mm_mul_epu32(H2, p->S23.v);
263 T6 = _mm_mul_epu32(H2, p->S24.v);
264 T0 = _mm_add_epi64(T0, T5);
265 T1 = _mm_add_epi64(T1, T6);
266 T5 = _mm_mul_epu32(H3, p->S22.v);
267 T6 = _mm_mul_epu32(H3, p->S23.v);
268 T0 = _mm_add_epi64(T0, T5);
269 T1 = _mm_add_epi64(T1, T6);
270 T5 = _mm_mul_epu32(H4, p->S21.v);
271 T6 = _mm_mul_epu32(H4, p->S22.v);
272 T0 = _mm_add_epi64(T0, T5);
273 T1 = _mm_add_epi64(T1, T6);
274 T5 = _mm_mul_epu32(H1, p->R21.v);
275 T6 = _mm_mul_epu32(H1, p->R22.v);
276 T2 = _mm_add_epi64(T2, T5);
277 T3 = _mm_add_epi64(T3, T6);
278 T5 = _mm_mul_epu32(H2, p->R20.v);
279 T6 = _mm_mul_epu32(H2, p->R21.v);
280 T2 = _mm_add_epi64(T2, T5);
281 T3 = _mm_add_epi64(T3, T6);
282 T5 = _mm_mul_epu32(H3, p->S24.v);
283 T6 = _mm_mul_epu32(H3, p->R20.v);
284 T2 = _mm_add_epi64(T2, T5);
285 T3 = _mm_add_epi64(T3, T6);
286 T5 = _mm_mul_epu32(H4, p->S23.v);
287 T6 = _mm_mul_epu32(H4, p->S24.v);
288 T2 = _mm_add_epi64(T2, T5);
289 T3 = _mm_add_epi64(T3, T6);
290 T5 = _mm_mul_epu32(H1, p->R23.v);
291 T4 = _mm_add_epi64(T4, T5);
292 T5 = _mm_mul_epu32(H2, p->R22.v);
293 T4 = _mm_add_epi64(T4, T5);
294 T5 = _mm_mul_epu32(H3, p->R21.v);
295 T4 = _mm_add_epi64(T4, T5);
296 T5 = _mm_mul_epu32(H4, p->R20.v);
297 T4 = _mm_add_epi64(T4, T5);
298
299 // H += [Mx,My]*[r^2,r^2]
300 T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((const xmmi *)(m + 0)),
301 _mm_loadl_epi64((const xmmi *)(m + 16)));
302 T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((const xmmi *)(m + 8)),
303 _mm_loadl_epi64((const xmmi *)(m + 24)));
304 M0 = _mm_and_si128(MMASK, T5);
305 M1 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
306 T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12));
307 M2 = _mm_and_si128(MMASK, T5);
308 M3 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
309 M4 = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT);
310
311 p = &st->P[1];
312 T5 = _mm_mul_epu32(M0, p->R20.v);
313 T6 = _mm_mul_epu32(M0, p->R21.v);
314 T0 = _mm_add_epi64(T0, T5);
315 T1 = _mm_add_epi64(T1, T6);
316 T5 = _mm_mul_epu32(M1, p->S24.v);
317 T6 = _mm_mul_epu32(M1, p->R20.v);
318 T0 = _mm_add_epi64(T0, T5);
319 T1 = _mm_add_epi64(T1, T6);
320 T5 = _mm_mul_epu32(M2, p->S23.v);
321 T6 = _mm_mul_epu32(M2, p->S24.v);
322 T0 = _mm_add_epi64(T0, T5);
323 T1 = _mm_add_epi64(T1, T6);
324 T5 = _mm_mul_epu32(M3, p->S22.v);
325 T6 = _mm_mul_epu32(M3, p->S23.v);
326 T0 = _mm_add_epi64(T0, T5);
327 T1 = _mm_add_epi64(T1, T6);
328 T5 = _mm_mul_epu32(M4, p->S21.v);
329 T6 = _mm_mul_epu32(M4, p->S22.v);
330 T0 = _mm_add_epi64(T0, T5);
331 T1 = _mm_add_epi64(T1, T6);
332 T5 = _mm_mul_epu32(M0, p->R22.v);
333 T6 = _mm_mul_epu32(M0, p->R23.v);
334 T2 = _mm_add_epi64(T2, T5);
335 T3 = _mm_add_epi64(T3, T6);
336 T5 = _mm_mul_epu32(M1, p->R21.v);
337 T6 = _mm_mul_epu32(M1, p->R22.v);
338 T2 = _mm_add_epi64(T2, T5);
339 T3 = _mm_add_epi64(T3, T6);
340 T5 = _mm_mul_epu32(M2, p->R20.v);
341 T6 = _mm_mul_epu32(M2, p->R21.v);
342 T2 = _mm_add_epi64(T2, T5);
343 T3 = _mm_add_epi64(T3, T6);
344 T5 = _mm_mul_epu32(M3, p->S24.v);
345 T6 = _mm_mul_epu32(M3, p->R20.v);
346 T2 = _mm_add_epi64(T2, T5);
347 T3 = _mm_add_epi64(T3, T6);
348 T5 = _mm_mul_epu32(M4, p->S23.v);
349 T6 = _mm_mul_epu32(M4, p->S24.v);
350 T2 = _mm_add_epi64(T2, T5);
351 T3 = _mm_add_epi64(T3, T6);
352 T5 = _mm_mul_epu32(M0, p->R24.v);
353 T4 = _mm_add_epi64(T4, T5);
354 T5 = _mm_mul_epu32(M1, p->R23.v);
355 T4 = _mm_add_epi64(T4, T5);
356 T5 = _mm_mul_epu32(M2, p->R22.v);
357 T4 = _mm_add_epi64(T4, T5);
358 T5 = _mm_mul_epu32(M3, p->R21.v);
359 T4 = _mm_add_epi64(T4, T5);
360 T5 = _mm_mul_epu32(M4, p->R20.v);
361 T4 = _mm_add_epi64(T4, T5);
362
363 // H += [Mx,My]
364 T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((const xmmi *)(m + 32)),
365 _mm_loadl_epi64((const xmmi *)(m + 48)));
366 T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((const xmmi *)(m + 40)),
367 _mm_loadl_epi64((const xmmi *)(m + 56)));
368 M0 = _mm_and_si128(MMASK, T5);
369 M1 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
370 T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12));
371 M2 = _mm_and_si128(MMASK, T5);
372 M3 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
373 M4 = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT);
374
375 T0 = _mm_add_epi64(T0, M0);
376 T1 = _mm_add_epi64(T1, M1);
377 T2 = _mm_add_epi64(T2, M2);
378 T3 = _mm_add_epi64(T3, M3);
379 T4 = _mm_add_epi64(T4, M4);
380
381 // reduce
382 C1 = _mm_srli_epi64(T0, 26);
383 C2 = _mm_srli_epi64(T3, 26);
384 T0 = _mm_and_si128(T0, MMASK);
385 T3 = _mm_and_si128(T3, MMASK);
386 T1 = _mm_add_epi64(T1, C1);
387 T4 = _mm_add_epi64(T4, C2);
388 C1 = _mm_srli_epi64(T1, 26);
389 C2 = _mm_srli_epi64(T4, 26);
390 T1 = _mm_and_si128(T1, MMASK);
391 T4 = _mm_and_si128(T4, MMASK);
392 T2 = _mm_add_epi64(T2, C1);
393 T0 = _mm_add_epi64(T0, _mm_mul_epu32(C2, FIVE));
394 C1 = _mm_srli_epi64(T2, 26);
395 C2 = _mm_srli_epi64(T0, 26);
396 T2 = _mm_and_si128(T2, MMASK);
397 T0 = _mm_and_si128(T0, MMASK);
398 T3 = _mm_add_epi64(T3, C1);
399 T1 = _mm_add_epi64(T1, C2);
400 C1 = _mm_srli_epi64(T3, 26);
401 T3 = _mm_and_si128(T3, MMASK);
402 T4 = _mm_add_epi64(T4, C1);
403
404 // H = (H*[r^4,r^4] + [Mx,My]*[r^2,r^2] + [Mx,My])
405 H0 = T0;
406 H1 = T1;
407 H2 = T2;
408 H3 = T3;
409 H4 = T4;
410
411 m += 64;
412 bytes -= 64;
413 }
414
415 st->H[0] = H0;
416 st->H[1] = H1;
417 st->H[2] = H2;
418 st->H[3] = H3;
419 st->H[4] = H4;
420 }
421
poly1305_combine(poly1305_state_internal * st,const uint8_t * m,size_t bytes)422 static size_t poly1305_combine(poly1305_state_internal *st, const uint8_t *m,
423 size_t bytes) {
424 const xmmi MMASK =
425 _mm_load_si128((const xmmi *)poly1305_x64_sse2_message_mask);
426 const xmmi HIBIT = _mm_load_si128((const xmmi *)poly1305_x64_sse2_1shl128);
427 const xmmi FIVE = _mm_load_si128((const xmmi *)poly1305_x64_sse2_5);
428
429 poly1305_power *p;
430 xmmi H0, H1, H2, H3, H4;
431 xmmi M0, M1, M2, M3, M4;
432 xmmi T0, T1, T2, T3, T4, T5, T6;
433 xmmi C1, C2;
434
435 uint64_t r0, r1, r2;
436 uint64_t t0, t1, t2, t3, t4;
437 uint64_t c;
438 size_t consumed = 0;
439
440 H0 = st->H[0];
441 H1 = st->H[1];
442 H2 = st->H[2];
443 H3 = st->H[3];
444 H4 = st->H[4];
445
446 // p = [r^2,r^2]
447 p = &st->P[1];
448
449 if (bytes >= 32) {
450 // H *= [r^2,r^2]
451 T0 = _mm_mul_epu32(H0, p->R20.v);
452 T1 = _mm_mul_epu32(H0, p->R21.v);
453 T2 = _mm_mul_epu32(H0, p->R22.v);
454 T3 = _mm_mul_epu32(H0, p->R23.v);
455 T4 = _mm_mul_epu32(H0, p->R24.v);
456 T5 = _mm_mul_epu32(H1, p->S24.v);
457 T6 = _mm_mul_epu32(H1, p->R20.v);
458 T0 = _mm_add_epi64(T0, T5);
459 T1 = _mm_add_epi64(T1, T6);
460 T5 = _mm_mul_epu32(H2, p->S23.v);
461 T6 = _mm_mul_epu32(H2, p->S24.v);
462 T0 = _mm_add_epi64(T0, T5);
463 T1 = _mm_add_epi64(T1, T6);
464 T5 = _mm_mul_epu32(H3, p->S22.v);
465 T6 = _mm_mul_epu32(H3, p->S23.v);
466 T0 = _mm_add_epi64(T0, T5);
467 T1 = _mm_add_epi64(T1, T6);
468 T5 = _mm_mul_epu32(H4, p->S21.v);
469 T6 = _mm_mul_epu32(H4, p->S22.v);
470 T0 = _mm_add_epi64(T0, T5);
471 T1 = _mm_add_epi64(T1, T6);
472 T5 = _mm_mul_epu32(H1, p->R21.v);
473 T6 = _mm_mul_epu32(H1, p->R22.v);
474 T2 = _mm_add_epi64(T2, T5);
475 T3 = _mm_add_epi64(T3, T6);
476 T5 = _mm_mul_epu32(H2, p->R20.v);
477 T6 = _mm_mul_epu32(H2, p->R21.v);
478 T2 = _mm_add_epi64(T2, T5);
479 T3 = _mm_add_epi64(T3, T6);
480 T5 = _mm_mul_epu32(H3, p->S24.v);
481 T6 = _mm_mul_epu32(H3, p->R20.v);
482 T2 = _mm_add_epi64(T2, T5);
483 T3 = _mm_add_epi64(T3, T6);
484 T5 = _mm_mul_epu32(H4, p->S23.v);
485 T6 = _mm_mul_epu32(H4, p->S24.v);
486 T2 = _mm_add_epi64(T2, T5);
487 T3 = _mm_add_epi64(T3, T6);
488 T5 = _mm_mul_epu32(H1, p->R23.v);
489 T4 = _mm_add_epi64(T4, T5);
490 T5 = _mm_mul_epu32(H2, p->R22.v);
491 T4 = _mm_add_epi64(T4, T5);
492 T5 = _mm_mul_epu32(H3, p->R21.v);
493 T4 = _mm_add_epi64(T4, T5);
494 T5 = _mm_mul_epu32(H4, p->R20.v);
495 T4 = _mm_add_epi64(T4, T5);
496
497 // H += [Mx,My]
498 T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((const xmmi *)(m + 0)),
499 _mm_loadl_epi64((const xmmi *)(m + 16)));
500 T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((const xmmi *)(m + 8)),
501 _mm_loadl_epi64((const xmmi *)(m + 24)));
502 M0 = _mm_and_si128(MMASK, T5);
503 M1 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
504 T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12));
505 M2 = _mm_and_si128(MMASK, T5);
506 M3 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
507 M4 = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT);
508
509 T0 = _mm_add_epi64(T0, M0);
510 T1 = _mm_add_epi64(T1, M1);
511 T2 = _mm_add_epi64(T2, M2);
512 T3 = _mm_add_epi64(T3, M3);
513 T4 = _mm_add_epi64(T4, M4);
514
515 // reduce
516 C1 = _mm_srli_epi64(T0, 26);
517 C2 = _mm_srli_epi64(T3, 26);
518 T0 = _mm_and_si128(T0, MMASK);
519 T3 = _mm_and_si128(T3, MMASK);
520 T1 = _mm_add_epi64(T1, C1);
521 T4 = _mm_add_epi64(T4, C2);
522 C1 = _mm_srli_epi64(T1, 26);
523 C2 = _mm_srli_epi64(T4, 26);
524 T1 = _mm_and_si128(T1, MMASK);
525 T4 = _mm_and_si128(T4, MMASK);
526 T2 = _mm_add_epi64(T2, C1);
527 T0 = _mm_add_epi64(T0, _mm_mul_epu32(C2, FIVE));
528 C1 = _mm_srli_epi64(T2, 26);
529 C2 = _mm_srli_epi64(T0, 26);
530 T2 = _mm_and_si128(T2, MMASK);
531 T0 = _mm_and_si128(T0, MMASK);
532 T3 = _mm_add_epi64(T3, C1);
533 T1 = _mm_add_epi64(T1, C2);
534 C1 = _mm_srli_epi64(T3, 26);
535 T3 = _mm_and_si128(T3, MMASK);
536 T4 = _mm_add_epi64(T4, C1);
537
538 // H = (H*[r^2,r^2] + [Mx,My])
539 H0 = T0;
540 H1 = T1;
541 H2 = T2;
542 H3 = T3;
543 H4 = T4;
544
545 consumed = 32;
546 }
547
548 // finalize, H *= [r^2,r]
549 r0 = ((uint64_t)p->R20.d[3] << 32) | (uint64_t)p->R20.d[1];
550 r1 = ((uint64_t)p->R21.d[3] << 32) | (uint64_t)p->R21.d[1];
551 r2 = ((uint64_t)p->R22.d[3] << 32) | (uint64_t)p->R22.d[1];
552
553 p->R20.d[2] = (uint32_t)(r0) & 0x3ffffff;
554 p->R21.d[2] = (uint32_t)((r0 >> 26) | (r1 << 18)) & 0x3ffffff;
555 p->R22.d[2] = (uint32_t)((r1 >> 8)) & 0x3ffffff;
556 p->R23.d[2] = (uint32_t)((r1 >> 34) | (r2 << 10)) & 0x3ffffff;
557 p->R24.d[2] = (uint32_t)((r2 >> 16));
558 p->S21.d[2] = p->R21.d[2] * 5;
559 p->S22.d[2] = p->R22.d[2] * 5;
560 p->S23.d[2] = p->R23.d[2] * 5;
561 p->S24.d[2] = p->R24.d[2] * 5;
562
563 // H *= [r^2,r]
564 T0 = _mm_mul_epu32(H0, p->R20.v);
565 T1 = _mm_mul_epu32(H0, p->R21.v);
566 T2 = _mm_mul_epu32(H0, p->R22.v);
567 T3 = _mm_mul_epu32(H0, p->R23.v);
568 T4 = _mm_mul_epu32(H0, p->R24.v);
569 T5 = _mm_mul_epu32(H1, p->S24.v);
570 T6 = _mm_mul_epu32(H1, p->R20.v);
571 T0 = _mm_add_epi64(T0, T5);
572 T1 = _mm_add_epi64(T1, T6);
573 T5 = _mm_mul_epu32(H2, p->S23.v);
574 T6 = _mm_mul_epu32(H2, p->S24.v);
575 T0 = _mm_add_epi64(T0, T5);
576 T1 = _mm_add_epi64(T1, T6);
577 T5 = _mm_mul_epu32(H3, p->S22.v);
578 T6 = _mm_mul_epu32(H3, p->S23.v);
579 T0 = _mm_add_epi64(T0, T5);
580 T1 = _mm_add_epi64(T1, T6);
581 T5 = _mm_mul_epu32(H4, p->S21.v);
582 T6 = _mm_mul_epu32(H4, p->S22.v);
583 T0 = _mm_add_epi64(T0, T5);
584 T1 = _mm_add_epi64(T1, T6);
585 T5 = _mm_mul_epu32(H1, p->R21.v);
586 T6 = _mm_mul_epu32(H1, p->R22.v);
587 T2 = _mm_add_epi64(T2, T5);
588 T3 = _mm_add_epi64(T3, T6);
589 T5 = _mm_mul_epu32(H2, p->R20.v);
590 T6 = _mm_mul_epu32(H2, p->R21.v);
591 T2 = _mm_add_epi64(T2, T5);
592 T3 = _mm_add_epi64(T3, T6);
593 T5 = _mm_mul_epu32(H3, p->S24.v);
594 T6 = _mm_mul_epu32(H3, p->R20.v);
595 T2 = _mm_add_epi64(T2, T5);
596 T3 = _mm_add_epi64(T3, T6);
597 T5 = _mm_mul_epu32(H4, p->S23.v);
598 T6 = _mm_mul_epu32(H4, p->S24.v);
599 T2 = _mm_add_epi64(T2, T5);
600 T3 = _mm_add_epi64(T3, T6);
601 T5 = _mm_mul_epu32(H1, p->R23.v);
602 T4 = _mm_add_epi64(T4, T5);
603 T5 = _mm_mul_epu32(H2, p->R22.v);
604 T4 = _mm_add_epi64(T4, T5);
605 T5 = _mm_mul_epu32(H3, p->R21.v);
606 T4 = _mm_add_epi64(T4, T5);
607 T5 = _mm_mul_epu32(H4, p->R20.v);
608 T4 = _mm_add_epi64(T4, T5);
609
610 C1 = _mm_srli_epi64(T0, 26);
611 C2 = _mm_srli_epi64(T3, 26);
612 T0 = _mm_and_si128(T0, MMASK);
613 T3 = _mm_and_si128(T3, MMASK);
614 T1 = _mm_add_epi64(T1, C1);
615 T4 = _mm_add_epi64(T4, C2);
616 C1 = _mm_srli_epi64(T1, 26);
617 C2 = _mm_srli_epi64(T4, 26);
618 T1 = _mm_and_si128(T1, MMASK);
619 T4 = _mm_and_si128(T4, MMASK);
620 T2 = _mm_add_epi64(T2, C1);
621 T0 = _mm_add_epi64(T0, _mm_mul_epu32(C2, FIVE));
622 C1 = _mm_srli_epi64(T2, 26);
623 C2 = _mm_srli_epi64(T0, 26);
624 T2 = _mm_and_si128(T2, MMASK);
625 T0 = _mm_and_si128(T0, MMASK);
626 T3 = _mm_add_epi64(T3, C1);
627 T1 = _mm_add_epi64(T1, C2);
628 C1 = _mm_srli_epi64(T3, 26);
629 T3 = _mm_and_si128(T3, MMASK);
630 T4 = _mm_add_epi64(T4, C1);
631
632 // H = H[0]+H[1]
633 H0 = _mm_add_epi64(T0, _mm_srli_si128(T0, 8));
634 H1 = _mm_add_epi64(T1, _mm_srli_si128(T1, 8));
635 H2 = _mm_add_epi64(T2, _mm_srli_si128(T2, 8));
636 H3 = _mm_add_epi64(T3, _mm_srli_si128(T3, 8));
637 H4 = _mm_add_epi64(T4, _mm_srli_si128(T4, 8));
638
639 t0 = _mm_cvtsi128_si32(H0);
640 c = (t0 >> 26);
641 t0 &= 0x3ffffff;
642 t1 = _mm_cvtsi128_si32(H1) + c;
643 c = (t1 >> 26);
644 t1 &= 0x3ffffff;
645 t2 = _mm_cvtsi128_si32(H2) + c;
646 c = (t2 >> 26);
647 t2 &= 0x3ffffff;
648 t3 = _mm_cvtsi128_si32(H3) + c;
649 c = (t3 >> 26);
650 t3 &= 0x3ffffff;
651 t4 = _mm_cvtsi128_si32(H4) + c;
652 c = (t4 >> 26);
653 t4 &= 0x3ffffff;
654 t0 = t0 + (c * 5);
655 c = (t0 >> 26);
656 t0 &= 0x3ffffff;
657 t1 = t1 + c;
658
659 st->HH[0] = ((t0) | (t1 << 26)) & UINT64_C(0xfffffffffff);
660 st->HH[1] = ((t1 >> 18) | (t2 << 8) | (t3 << 34)) & UINT64_C(0xfffffffffff);
661 st->HH[2] = ((t3 >> 10) | (t4 << 16)) & UINT64_C(0x3ffffffffff);
662
663 return consumed;
664 }
665
CRYPTO_poly1305_update(poly1305_state * state,const uint8_t * m,size_t bytes)666 void CRYPTO_poly1305_update(poly1305_state *state, const uint8_t *m,
667 size_t bytes) {
668 poly1305_state_internal *st = poly1305_aligned_state(state);
669 size_t want;
670
671 // Work around a C language bug. See https://crbug.com/1019588.
672 if (bytes == 0) {
673 return;
674 }
675
676 // need at least 32 initial bytes to start the accelerated branch
677 if (!st->started) {
678 if ((st->leftover == 0) && (bytes > 32)) {
679 poly1305_first_block(st, m);
680 m += 32;
681 bytes -= 32;
682 } else {
683 want = poly1305_min(32 - st->leftover, bytes);
684 OPENSSL_memcpy(st->buffer + st->leftover, m, want);
685 bytes -= want;
686 m += want;
687 st->leftover += want;
688 if ((st->leftover < 32) || (bytes == 0)) {
689 return;
690 }
691 poly1305_first_block(st, st->buffer);
692 st->leftover = 0;
693 }
694 st->started = 1;
695 }
696
697 // handle leftover
698 if (st->leftover) {
699 want = poly1305_min(64 - st->leftover, bytes);
700 OPENSSL_memcpy(st->buffer + st->leftover, m, want);
701 bytes -= want;
702 m += want;
703 st->leftover += want;
704 if (st->leftover < 64) {
705 return;
706 }
707 poly1305_blocks(st, st->buffer, 64);
708 st->leftover = 0;
709 }
710
711 // process 64 byte blocks
712 if (bytes >= 64) {
713 want = (bytes & ~63);
714 poly1305_blocks(st, m, want);
715 m += want;
716 bytes -= want;
717 }
718
719 if (bytes) {
720 OPENSSL_memcpy(st->buffer + st->leftover, m, bytes);
721 st->leftover += bytes;
722 }
723 }
724
CRYPTO_poly1305_finish(poly1305_state * state,uint8_t mac[16])725 void CRYPTO_poly1305_finish(poly1305_state *state, uint8_t mac[16]) {
726 poly1305_state_internal *st = poly1305_aligned_state(state);
727 size_t leftover = st->leftover;
728 uint8_t *m = st->buffer;
729 uint128_t d[3];
730 uint64_t h0, h1, h2;
731 uint64_t t0, t1;
732 uint64_t g0, g1, g2, c, nc;
733 uint64_t r0, r1, r2, s1, s2;
734 poly1305_power *p;
735
736 if (st->started) {
737 size_t consumed = poly1305_combine(st, m, leftover);
738 leftover -= consumed;
739 m += consumed;
740 }
741
742 // st->HH will either be 0 or have the combined result
743 h0 = st->HH[0];
744 h1 = st->HH[1];
745 h2 = st->HH[2];
746
747 p = &st->P[1];
748 r0 = ((uint64_t)p->R20.d[3] << 32) | (uint64_t)p->R20.d[1];
749 r1 = ((uint64_t)p->R21.d[3] << 32) | (uint64_t)p->R21.d[1];
750 r2 = ((uint64_t)p->R22.d[3] << 32) | (uint64_t)p->R22.d[1];
751 s1 = r1 * (5 << 2);
752 s2 = r2 * (5 << 2);
753
754 if (leftover < 16) {
755 goto poly1305_donna_atmost15bytes;
756 }
757
758 poly1305_donna_atleast16bytes:
759 t0 = CRYPTO_load_u64_le(m + 0);
760 t1 = CRYPTO_load_u64_le(m + 8);
761 h0 += t0 & 0xfffffffffff;
762 t0 = shr128_pair(t1, t0, 44);
763 h1 += t0 & 0xfffffffffff;
764 h2 += (t1 >> 24) | ((uint64_t)1 << 40);
765
766 poly1305_donna_mul:
767 d[0] = add128(add128(mul64x64_128(h0, r0), mul64x64_128(h1, s2)),
768 mul64x64_128(h2, s1));
769 d[1] = add128(add128(mul64x64_128(h0, r1), mul64x64_128(h1, r0)),
770 mul64x64_128(h2, s2));
771 d[2] = add128(add128(mul64x64_128(h0, r2), mul64x64_128(h1, r1)),
772 mul64x64_128(h2, r0));
773 h0 = lo128(d[0]) & 0xfffffffffff;
774 c = shr128(d[0], 44);
775 d[1] = add128_64(d[1], c);
776 h1 = lo128(d[1]) & 0xfffffffffff;
777 c = shr128(d[1], 44);
778 d[2] = add128_64(d[2], c);
779 h2 = lo128(d[2]) & 0x3ffffffffff;
780 c = shr128(d[2], 42);
781 h0 += c * 5;
782
783 m += 16;
784 leftover -= 16;
785 if (leftover >= 16) {
786 goto poly1305_donna_atleast16bytes;
787 }
788
789 // final bytes
790 poly1305_donna_atmost15bytes:
791 if (!leftover) {
792 goto poly1305_donna_finish;
793 }
794
795 m[leftover++] = 1;
796 OPENSSL_memset(m + leftover, 0, 16 - leftover);
797 leftover = 16;
798
799 t0 = CRYPTO_load_u64_le(m + 0);
800 t1 = CRYPTO_load_u64_le(m + 8);
801 h0 += t0 & 0xfffffffffff;
802 t0 = shr128_pair(t1, t0, 44);
803 h1 += t0 & 0xfffffffffff;
804 h2 += (t1 >> 24);
805
806 goto poly1305_donna_mul;
807
808 poly1305_donna_finish:
809 c = (h0 >> 44);
810 h0 &= 0xfffffffffff;
811 h1 += c;
812 c = (h1 >> 44);
813 h1 &= 0xfffffffffff;
814 h2 += c;
815 c = (h2 >> 42);
816 h2 &= 0x3ffffffffff;
817 h0 += c * 5;
818
819 g0 = h0 + 5;
820 c = (g0 >> 44);
821 g0 &= 0xfffffffffff;
822 g1 = h1 + c;
823 c = (g1 >> 44);
824 g1 &= 0xfffffffffff;
825 g2 = h2 + c - ((uint64_t)1 << 42);
826
827 c = (g2 >> 63) - 1;
828 nc = ~c;
829 h0 = (h0 & nc) | (g0 & c);
830 h1 = (h1 & nc) | (g1 & c);
831 h2 = (h2 & nc) | (g2 & c);
832
833 // pad
834 t0 = ((uint64_t)p->R23.d[3] << 32) | (uint64_t)p->R23.d[1];
835 t1 = ((uint64_t)p->R24.d[3] << 32) | (uint64_t)p->R24.d[1];
836 h0 += (t0 & 0xfffffffffff);
837 c = (h0 >> 44);
838 h0 &= 0xfffffffffff;
839 t0 = shr128_pair(t1, t0, 44);
840 h1 += (t0 & 0xfffffffffff) + c;
841 c = (h1 >> 44);
842 h1 &= 0xfffffffffff;
843 t1 = (t1 >> 24);
844 h2 += (t1) + c;
845
846 CRYPTO_store_u64_le(mac + 0, ((h0) | (h1 << 44)));
847 CRYPTO_store_u64_le(mac + 8, ((h1 >> 20) | (h2 << 24)));
848 }
849
850 #endif // BORINGSSL_HAS_UINT128 && OPENSSL_X86_64
851