]>
Commit | Line | Data |
---|---|---|
4d3fa06f AP |
1 | /****************************************************************************** |
2 | * * | |
3 | * Copyright 2014 Intel Corporation * | |
4 | * * | |
5 | * Licensed under the Apache License, Version 2.0 (the "License"); * | |
6 | * you may not use this file except in compliance with the License. * | |
7 | * You may obtain a copy of the License at * | |
8 | * * | |
9 | * http://www.apache.org/licenses/LICENSE-2.0 * | |
10 | * * | |
11 | * Unless required by applicable law or agreed to in writing, software * | |
12 | * distributed under the License is distributed on an "AS IS" BASIS, * | |
13 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * | |
14 | * See the License for the specific language governing permissions and * | |
15 | * limitations under the License. * | |
16 | * * | |
17 | ****************************************************************************** | |
18 | * * | |
19 | * Developers and authors: * | |
20 | * Shay Gueron (1, 2), and Vlad Krasnov (1) * | |
21 | * (1) Intel Corporation, Israel Development Center * | |
22 | * (2) University of Haifa * | |
23 | * Reference: * | |
24 | * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with * | |
25 | * 256 Bit Primes" * | |
26 | * * | |
27 | ******************************************************************************/ | |
28 | ||
29 | #include <string.h> | |
30 | ||
b39fc560 | 31 | #include "internal/cryptlib.h" |
5afc296a | 32 | #include "internal/bn_int.h" |
4d3fa06f AP |
33 | #include "ec_lcl.h" |
34 | ||
35 | #if BN_BITS2 != 64 | |
58d47cf0 | 36 | # define TOBN(hi,lo) lo,hi |
4d3fa06f | 37 | #else |
58d47cf0 | 38 | # define TOBN(hi,lo) ((BN_ULONG)hi<<32|lo) |
4d3fa06f AP |
39 | #endif |
40 | ||
41 | #if defined(__GNUC__) | |
58d47cf0 | 42 | # define ALIGN32 __attribute((aligned(32))) |
4d3fa06f | 43 | #elif defined(_MSC_VER) |
58d47cf0 | 44 | # define ALIGN32 __declspec(align(32)) |
4d3fa06f AP |
45 | #else |
46 | # define ALIGN32 | |
47 | #endif | |
48 | ||
58d47cf0 AP |
49 | #define ALIGNPTR(p,N) ((unsigned char *)p+N-(size_t)p%N) |
50 | #define P256_LIMBS (256/BN_BITS2) | |
4d3fa06f AP |
51 | |
52 | typedef unsigned short u16; | |
53 | ||
54 | typedef struct { | |
55 | BN_ULONG X[P256_LIMBS]; | |
56 | BN_ULONG Y[P256_LIMBS]; | |
57 | BN_ULONG Z[P256_LIMBS]; | |
58 | } P256_POINT; | |
59 | ||
60 | typedef struct { | |
61 | BN_ULONG X[P256_LIMBS]; | |
62 | BN_ULONG Y[P256_LIMBS]; | |
63 | } P256_POINT_AFFINE; | |
64 | ||
65 | typedef P256_POINT_AFFINE PRECOMP256_ROW[64]; | |
66 | ||
67 | /* structure for precomputed multiples of the generator */ | |
3aef36ff | 68 | struct nistz256_pre_comp_st { |
4d3fa06f AP |
69 | const EC_GROUP *group; /* Parent EC_GROUP object */ |
70 | size_t w; /* Window size */ | |
20728adc AP |
71 | /* |
72 | * Constant time access to the X and Y coordinates of the pre-computed, | |
4d3fa06f | 73 | * generator multiplies, in the Montgomery domain. Pre-calculated |
20728adc AP |
74 | * multiplies are stored in affine form. |
75 | */ | |
4d3fa06f AP |
76 | PRECOMP256_ROW *precomp; |
77 | void *precomp_storage; | |
78 | int references; | |
3aef36ff | 79 | }; |
4d3fa06f AP |
80 | |
81 | /* Functions implemented in assembly */ | |
82 | /* Modular mul by 2: res = 2*a mod P */ | |
83 | void ecp_nistz256_mul_by_2(BN_ULONG res[P256_LIMBS], | |
84 | const BN_ULONG a[P256_LIMBS]); | |
85 | /* Modular div by 2: res = a/2 mod P */ | |
86 | void ecp_nistz256_div_by_2(BN_ULONG res[P256_LIMBS], | |
87 | const BN_ULONG a[P256_LIMBS]); | |
88 | /* Modular mul by 3: res = 3*a mod P */ | |
89 | void ecp_nistz256_mul_by_3(BN_ULONG res[P256_LIMBS], | |
90 | const BN_ULONG a[P256_LIMBS]); | |
20728adc | 91 | /* Modular add: res = a+b mod P */ |
4d3fa06f AP |
92 | void ecp_nistz256_add(BN_ULONG res[P256_LIMBS], |
93 | const BN_ULONG a[P256_LIMBS], | |
94 | const BN_ULONG b[P256_LIMBS]); | |
20728adc | 95 | /* Modular sub: res = a-b mod P */ |
4d3fa06f AP |
96 | void ecp_nistz256_sub(BN_ULONG res[P256_LIMBS], |
97 | const BN_ULONG a[P256_LIMBS], | |
98 | const BN_ULONG b[P256_LIMBS]); | |
20728adc | 99 | /* Modular neg: res = -a mod P */ |
4d3fa06f AP |
100 | void ecp_nistz256_neg(BN_ULONG res[P256_LIMBS], const BN_ULONG a[P256_LIMBS]); |
101 | /* Montgomery mul: res = a*b*2^-256 mod P */ | |
102 | void ecp_nistz256_mul_mont(BN_ULONG res[P256_LIMBS], | |
103 | const BN_ULONG a[P256_LIMBS], | |
104 | const BN_ULONG b[P256_LIMBS]); | |
105 | /* Montgomery sqr: res = a*a*2^-256 mod P */ | |
106 | void ecp_nistz256_sqr_mont(BN_ULONG res[P256_LIMBS], | |
107 | const BN_ULONG a[P256_LIMBS]); | |
108 | /* Convert a number from Montgomery domain, by multiplying with 1 */ | |
109 | void ecp_nistz256_from_mont(BN_ULONG res[P256_LIMBS], | |
110 | const BN_ULONG in[P256_LIMBS]); | |
111 | /* Convert a number to Montgomery domain, by multiplying with 2^512 mod P*/ | |
112 | void ecp_nistz256_to_mont(BN_ULONG res[P256_LIMBS], | |
113 | const BN_ULONG in[P256_LIMBS]); | |
114 | /* Functions that perform constant time access to the precomputed tables */ | |
58d47cf0 | 115 | void ecp_nistz256_scatter_w5(P256_POINT *val, |
49b05c7d | 116 | const P256_POINT *in_t, int idx); |
20728adc | 117 | void ecp_nistz256_gather_w5(P256_POINT *val, |
49b05c7d | 118 | const P256_POINT *in_t, int idx); |
58d47cf0 | 119 | void ecp_nistz256_scatter_w7(P256_POINT_AFFINE *val, |
49b05c7d | 120 | const P256_POINT_AFFINE *in_t, int idx); |
58d47cf0 | 121 | void ecp_nistz256_gather_w7(P256_POINT_AFFINE *val, |
49b05c7d | 122 | const P256_POINT_AFFINE *in_t, int idx); |
4d3fa06f AP |
123 | |
124 | /* One converted into the Montgomery domain */ | |
125 | static const BN_ULONG ONE[P256_LIMBS] = { | |
126 | TOBN(0x00000000, 0x00000001), TOBN(0xffffffff, 0x00000000), | |
127 | TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe) | |
128 | }; | |
129 | ||
3aef36ff | 130 | static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group); |
4d3fa06f AP |
131 | |
132 | /* Precomputed tables for the default generator */ | |
3ff08e1d | 133 | extern const PRECOMP256_ROW ecp_nistz256_precomputed[37]; |
4d3fa06f AP |
134 | |
135 | /* Recode window to a signed digit, see ecp_nistputil.c for details */ | |
136 | static unsigned int _booth_recode_w5(unsigned int in) | |
137 | { | |
138 | unsigned int s, d; | |
139 | ||
140 | s = ~((in >> 5) - 1); | |
141 | d = (1 << 6) - in - 1; | |
142 | d = (d & s) | (in & ~s); | |
143 | d = (d >> 1) + (d & 1); | |
144 | ||
145 | return (d << 1) + (s & 1); | |
146 | } | |
147 | ||
148 | static unsigned int _booth_recode_w7(unsigned int in) | |
149 | { | |
150 | unsigned int s, d; | |
151 | ||
152 | s = ~((in >> 7) - 1); | |
153 | d = (1 << 8) - in - 1; | |
154 | d = (d & s) | (in & ~s); | |
155 | d = (d >> 1) + (d & 1); | |
156 | ||
157 | return (d << 1) + (s & 1); | |
158 | } | |
159 | ||
160 | static void copy_conditional(BN_ULONG dst[P256_LIMBS], | |
161 | const BN_ULONG src[P256_LIMBS], BN_ULONG move) | |
162 | { | |
5afc296a | 163 | BN_ULONG mask1 = 0-move; |
4d3fa06f AP |
164 | BN_ULONG mask2 = ~mask1; |
165 | ||
166 | dst[0] = (src[0] & mask1) ^ (dst[0] & mask2); | |
167 | dst[1] = (src[1] & mask1) ^ (dst[1] & mask2); | |
168 | dst[2] = (src[2] & mask1) ^ (dst[2] & mask2); | |
169 | dst[3] = (src[3] & mask1) ^ (dst[3] & mask2); | |
170 | if (P256_LIMBS == 8) { | |
171 | dst[4] = (src[4] & mask1) ^ (dst[4] & mask2); | |
172 | dst[5] = (src[5] & mask1) ^ (dst[5] & mask2); | |
173 | dst[6] = (src[6] & mask1) ^ (dst[6] & mask2); | |
174 | dst[7] = (src[7] & mask1) ^ (dst[7] & mask2); | |
175 | } | |
176 | } | |
177 | ||
178 | static BN_ULONG is_zero(BN_ULONG in) | |
179 | { | |
180 | in |= (0 - in); | |
181 | in = ~in; | |
4d3fa06f AP |
182 | in >>= BN_BITS2 - 1; |
183 | return in; | |
184 | } | |
185 | ||
186 | static BN_ULONG is_equal(const BN_ULONG a[P256_LIMBS], | |
187 | const BN_ULONG b[P256_LIMBS]) | |
188 | { | |
189 | BN_ULONG res; | |
190 | ||
191 | res = a[0] ^ b[0]; | |
192 | res |= a[1] ^ b[1]; | |
193 | res |= a[2] ^ b[2]; | |
194 | res |= a[3] ^ b[3]; | |
195 | if (P256_LIMBS == 8) { | |
196 | res |= a[4] ^ b[4]; | |
197 | res |= a[5] ^ b[5]; | |
198 | res |= a[6] ^ b[6]; | |
199 | res |= a[7] ^ b[7]; | |
200 | } | |
201 | ||
202 | return is_zero(res); | |
203 | } | |
204 | ||
205 | static BN_ULONG is_one(const BN_ULONG a[P256_LIMBS]) | |
206 | { | |
207 | BN_ULONG res; | |
208 | ||
209 | res = a[0] ^ ONE[0]; | |
210 | res |= a[1] ^ ONE[1]; | |
211 | res |= a[2] ^ ONE[2]; | |
212 | res |= a[3] ^ ONE[3]; | |
213 | if (P256_LIMBS == 8) { | |
214 | res |= a[4] ^ ONE[4]; | |
215 | res |= a[5] ^ ONE[5]; | |
216 | res |= a[6] ^ ONE[6]; | |
217 | } | |
218 | ||
219 | return is_zero(res); | |
220 | } | |
221 | ||
222 | #ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION | |
58d47cf0 AP |
223 | void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a); |
224 | void ecp_nistz256_point_add(P256_POINT *r, | |
225 | const P256_POINT *a, const P256_POINT *b); | |
226 | void ecp_nistz256_point_add_affine(P256_POINT *r, | |
227 | const P256_POINT *a, | |
228 | const P256_POINT_AFFINE *b); | |
4d3fa06f AP |
229 | #else |
230 | /* Point double: r = 2*a */ | |
58d47cf0 | 231 | static void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a) |
4d3fa06f AP |
232 | { |
233 | BN_ULONG S[P256_LIMBS]; | |
234 | BN_ULONG M[P256_LIMBS]; | |
235 | BN_ULONG Zsqr[P256_LIMBS]; | |
236 | BN_ULONG tmp0[P256_LIMBS]; | |
237 | ||
238 | const BN_ULONG *in_x = a->X; | |
239 | const BN_ULONG *in_y = a->Y; | |
240 | const BN_ULONG *in_z = a->Z; | |
241 | ||
242 | BN_ULONG *res_x = r->X; | |
243 | BN_ULONG *res_y = r->Y; | |
244 | BN_ULONG *res_z = r->Z; | |
245 | ||
246 | ecp_nistz256_mul_by_2(S, in_y); | |
247 | ||
248 | ecp_nistz256_sqr_mont(Zsqr, in_z); | |
249 | ||
250 | ecp_nistz256_sqr_mont(S, S); | |
251 | ||
252 | ecp_nistz256_mul_mont(res_z, in_z, in_y); | |
253 | ecp_nistz256_mul_by_2(res_z, res_z); | |
254 | ||
255 | ecp_nistz256_add(M, in_x, Zsqr); | |
256 | ecp_nistz256_sub(Zsqr, in_x, Zsqr); | |
257 | ||
258 | ecp_nistz256_sqr_mont(res_y, S); | |
259 | ecp_nistz256_div_by_2(res_y, res_y); | |
260 | ||
261 | ecp_nistz256_mul_mont(M, M, Zsqr); | |
262 | ecp_nistz256_mul_by_3(M, M); | |
263 | ||
264 | ecp_nistz256_mul_mont(S, S, in_x); | |
265 | ecp_nistz256_mul_by_2(tmp0, S); | |
266 | ||
267 | ecp_nistz256_sqr_mont(res_x, M); | |
268 | ||
269 | ecp_nistz256_sub(res_x, res_x, tmp0); | |
270 | ecp_nistz256_sub(S, S, res_x); | |
271 | ||
272 | ecp_nistz256_mul_mont(S, S, M); | |
273 | ecp_nistz256_sub(res_y, S, res_y); | |
274 | } | |
275 | ||
276 | /* Point addition: r = a+b */ | |
20728adc AP |
277 | static void ecp_nistz256_point_add(P256_POINT *r, |
278 | const P256_POINT *a, const P256_POINT *b) | |
4d3fa06f AP |
279 | { |
280 | BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS]; | |
281 | BN_ULONG U1[P256_LIMBS], S1[P256_LIMBS]; | |
282 | BN_ULONG Z1sqr[P256_LIMBS]; | |
283 | BN_ULONG Z2sqr[P256_LIMBS]; | |
284 | BN_ULONG H[P256_LIMBS], R[P256_LIMBS]; | |
285 | BN_ULONG Hsqr[P256_LIMBS]; | |
286 | BN_ULONG Rsqr[P256_LIMBS]; | |
287 | BN_ULONG Hcub[P256_LIMBS]; | |
288 | ||
289 | BN_ULONG res_x[P256_LIMBS]; | |
290 | BN_ULONG res_y[P256_LIMBS]; | |
291 | BN_ULONG res_z[P256_LIMBS]; | |
292 | ||
293 | BN_ULONG in1infty, in2infty; | |
294 | ||
295 | const BN_ULONG *in1_x = a->X; | |
296 | const BN_ULONG *in1_y = a->Y; | |
297 | const BN_ULONG *in1_z = a->Z; | |
298 | ||
299 | const BN_ULONG *in2_x = b->X; | |
300 | const BN_ULONG *in2_y = b->Y; | |
301 | const BN_ULONG *in2_z = b->Z; | |
302 | ||
303 | /* We encode infinity as (0,0), which is not on the curve, | |
304 | * so it is OK. */ | |
58d47cf0 AP |
305 | in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] | |
306 | in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]); | |
4d3fa06f | 307 | if (P256_LIMBS == 8) |
58d47cf0 AP |
308 | in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] | |
309 | in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]); | |
4d3fa06f | 310 | |
58d47cf0 AP |
311 | in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] | |
312 | in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]); | |
4d3fa06f | 313 | if (P256_LIMBS == 8) |
58d47cf0 AP |
314 | in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] | |
315 | in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]); | |
4d3fa06f AP |
316 | |
317 | in1infty = is_zero(in1infty); | |
318 | in2infty = is_zero(in2infty); | |
319 | ||
320 | ecp_nistz256_sqr_mont(Z2sqr, in2_z); /* Z2^2 */ | |
321 | ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */ | |
322 | ||
323 | ecp_nistz256_mul_mont(S1, Z2sqr, in2_z); /* S1 = Z2^3 */ | |
324 | ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */ | |
325 | ||
326 | ecp_nistz256_mul_mont(S1, S1, in1_y); /* S1 = Y1*Z2^3 */ | |
327 | ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */ | |
328 | ecp_nistz256_sub(R, S2, S1); /* R = S2 - S1 */ | |
329 | ||
330 | ecp_nistz256_mul_mont(U1, in1_x, Z2sqr); /* U1 = X1*Z2^2 */ | |
331 | ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */ | |
332 | ecp_nistz256_sub(H, U2, U1); /* H = U2 - U1 */ | |
333 | ||
20728adc AP |
334 | /* |
335 | * This should not happen during sign/ecdh, so no constant time violation | |
336 | */ | |
4d3fa06f AP |
337 | if (is_equal(U1, U2) && !in1infty && !in2infty) { |
338 | if (is_equal(S1, S2)) { | |
339 | ecp_nistz256_point_double(r, a); | |
340 | return; | |
341 | } else { | |
342 | memset(r, 0, sizeof(*r)); | |
343 | return; | |
344 | } | |
345 | } | |
346 | ||
347 | ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */ | |
348 | ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */ | |
349 | ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */ | |
350 | ecp_nistz256_mul_mont(res_z, res_z, in2_z); /* Z3 = H*Z1*Z2 */ | |
351 | ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */ | |
352 | ||
353 | ecp_nistz256_mul_mont(U2, U1, Hsqr); /* U1*H^2 */ | |
354 | ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */ | |
355 | ||
356 | ecp_nistz256_sub(res_x, Rsqr, Hsqr); | |
357 | ecp_nistz256_sub(res_x, res_x, Hcub); | |
358 | ||
359 | ecp_nistz256_sub(res_y, U2, res_x); | |
360 | ||
361 | ecp_nistz256_mul_mont(S2, S1, Hcub); | |
362 | ecp_nistz256_mul_mont(res_y, R, res_y); | |
363 | ecp_nistz256_sub(res_y, res_y, S2); | |
364 | ||
365 | copy_conditional(res_x, in2_x, in1infty); | |
366 | copy_conditional(res_y, in2_y, in1infty); | |
367 | copy_conditional(res_z, in2_z, in1infty); | |
368 | ||
369 | copy_conditional(res_x, in1_x, in2infty); | |
370 | copy_conditional(res_y, in1_y, in2infty); | |
371 | copy_conditional(res_z, in1_z, in2infty); | |
372 | ||
373 | memcpy(r->X, res_x, sizeof(res_x)); | |
374 | memcpy(r->Y, res_y, sizeof(res_y)); | |
375 | memcpy(r->Z, res_z, sizeof(res_z)); | |
376 | } | |
377 | ||
378 | /* Point addition when b is known to be affine: r = a+b */ | |
58d47cf0 AP |
379 | static void ecp_nistz256_point_add_affine(P256_POINT *r, |
380 | const P256_POINT *a, | |
381 | const P256_POINT_AFFINE *b) | |
4d3fa06f AP |
382 | { |
383 | BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS]; | |
384 | BN_ULONG Z1sqr[P256_LIMBS]; | |
385 | BN_ULONG H[P256_LIMBS], R[P256_LIMBS]; | |
386 | BN_ULONG Hsqr[P256_LIMBS]; | |
387 | BN_ULONG Rsqr[P256_LIMBS]; | |
388 | BN_ULONG Hcub[P256_LIMBS]; | |
389 | ||
390 | BN_ULONG res_x[P256_LIMBS]; | |
391 | BN_ULONG res_y[P256_LIMBS]; | |
392 | BN_ULONG res_z[P256_LIMBS]; | |
393 | ||
394 | BN_ULONG in1infty, in2infty; | |
395 | ||
396 | const BN_ULONG *in1_x = a->X; | |
397 | const BN_ULONG *in1_y = a->Y; | |
398 | const BN_ULONG *in1_z = a->Z; | |
399 | ||
400 | const BN_ULONG *in2_x = b->X; | |
401 | const BN_ULONG *in2_y = b->Y; | |
402 | ||
20728adc AP |
403 | /* |
404 | * In affine representation we encode infty as (0,0), which is not on the | |
405 | * curve, so it is OK | |
406 | */ | |
58d47cf0 AP |
407 | in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] | |
408 | in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]); | |
4d3fa06f | 409 | if (P256_LIMBS == 8) |
58d47cf0 AP |
410 | in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] | |
411 | in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]); | |
4d3fa06f | 412 | |
58d47cf0 AP |
413 | in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] | |
414 | in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]); | |
4d3fa06f | 415 | if (P256_LIMBS == 8) |
58d47cf0 AP |
416 | in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] | |
417 | in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]); | |
4d3fa06f AP |
418 | |
419 | in1infty = is_zero(in1infty); | |
420 | in2infty = is_zero(in2infty); | |
421 | ||
422 | ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */ | |
423 | ||
424 | ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */ | |
425 | ecp_nistz256_sub(H, U2, in1_x); /* H = U2 - U1 */ | |
426 | ||
427 | ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */ | |
428 | ||
429 | ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */ | |
430 | ||
431 | ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */ | |
432 | ecp_nistz256_sub(R, S2, in1_y); /* R = S2 - S1 */ | |
433 | ||
434 | ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */ | |
435 | ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */ | |
436 | ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */ | |
437 | ||
438 | ecp_nistz256_mul_mont(U2, in1_x, Hsqr); /* U1*H^2 */ | |
439 | ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */ | |
440 | ||
441 | ecp_nistz256_sub(res_x, Rsqr, Hsqr); | |
442 | ecp_nistz256_sub(res_x, res_x, Hcub); | |
443 | ecp_nistz256_sub(H, U2, res_x); | |
444 | ||
445 | ecp_nistz256_mul_mont(S2, in1_y, Hcub); | |
446 | ecp_nistz256_mul_mont(H, H, R); | |
447 | ecp_nistz256_sub(res_y, H, S2); | |
448 | ||
449 | copy_conditional(res_x, in2_x, in1infty); | |
450 | copy_conditional(res_x, in1_x, in2infty); | |
451 | ||
452 | copy_conditional(res_y, in2_y, in1infty); | |
453 | copy_conditional(res_y, in1_y, in2infty); | |
454 | ||
455 | copy_conditional(res_z, ONE, in1infty); | |
456 | copy_conditional(res_z, in1_z, in2infty); | |
457 | ||
458 | memcpy(r->X, res_x, sizeof(res_x)); | |
459 | memcpy(r->Y, res_y, sizeof(res_y)); | |
460 | memcpy(r->Z, res_z, sizeof(res_z)); | |
461 | } | |
462 | #endif | |
463 | ||
464 | /* r = in^-1 mod p */ | |
465 | static void ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS], | |
466 | const BN_ULONG in[P256_LIMBS]) | |
467 | { | |
20728adc AP |
468 | /* |
469 | * The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff | |
470 | * ffffffff ffffffff We use FLT and used poly-2 as exponent | |
471 | */ | |
4d3fa06f AP |
472 | BN_ULONG p2[P256_LIMBS]; |
473 | BN_ULONG p4[P256_LIMBS]; | |
474 | BN_ULONG p8[P256_LIMBS]; | |
475 | BN_ULONG p16[P256_LIMBS]; | |
476 | BN_ULONG p32[P256_LIMBS]; | |
477 | BN_ULONG res[P256_LIMBS]; | |
478 | int i; | |
479 | ||
480 | ecp_nistz256_sqr_mont(res, in); | |
481 | ecp_nistz256_mul_mont(p2, res, in); /* 3*p */ | |
482 | ||
483 | ecp_nistz256_sqr_mont(res, p2); | |
484 | ecp_nistz256_sqr_mont(res, res); | |
485 | ecp_nistz256_mul_mont(p4, res, p2); /* f*p */ | |
486 | ||
487 | ecp_nistz256_sqr_mont(res, p4); | |
488 | ecp_nistz256_sqr_mont(res, res); | |
489 | ecp_nistz256_sqr_mont(res, res); | |
490 | ecp_nistz256_sqr_mont(res, res); | |
491 | ecp_nistz256_mul_mont(p8, res, p4); /* ff*p */ | |
492 | ||
493 | ecp_nistz256_sqr_mont(res, p8); | |
494 | for (i = 0; i < 7; i++) | |
495 | ecp_nistz256_sqr_mont(res, res); | |
496 | ecp_nistz256_mul_mont(p16, res, p8); /* ffff*p */ | |
497 | ||
498 | ecp_nistz256_sqr_mont(res, p16); | |
499 | for (i = 0; i < 15; i++) | |
500 | ecp_nistz256_sqr_mont(res, res); | |
501 | ecp_nistz256_mul_mont(p32, res, p16); /* ffffffff*p */ | |
502 | ||
503 | ecp_nistz256_sqr_mont(res, p32); | |
504 | for (i = 0; i < 31; i++) | |
505 | ecp_nistz256_sqr_mont(res, res); | |
506 | ecp_nistz256_mul_mont(res, res, in); | |
507 | ||
508 | for (i = 0; i < 32 * 4; i++) | |
509 | ecp_nistz256_sqr_mont(res, res); | |
510 | ecp_nistz256_mul_mont(res, res, p32); | |
511 | ||
512 | for (i = 0; i < 32; i++) | |
513 | ecp_nistz256_sqr_mont(res, res); | |
514 | ecp_nistz256_mul_mont(res, res, p32); | |
515 | ||
516 | for (i = 0; i < 16; i++) | |
517 | ecp_nistz256_sqr_mont(res, res); | |
518 | ecp_nistz256_mul_mont(res, res, p16); | |
519 | ||
520 | for (i = 0; i < 8; i++) | |
521 | ecp_nistz256_sqr_mont(res, res); | |
522 | ecp_nistz256_mul_mont(res, res, p8); | |
523 | ||
524 | ecp_nistz256_sqr_mont(res, res); | |
525 | ecp_nistz256_sqr_mont(res, res); | |
526 | ecp_nistz256_sqr_mont(res, res); | |
527 | ecp_nistz256_sqr_mont(res, res); | |
528 | ecp_nistz256_mul_mont(res, res, p4); | |
529 | ||
530 | ecp_nistz256_sqr_mont(res, res); | |
531 | ecp_nistz256_sqr_mont(res, res); | |
532 | ecp_nistz256_mul_mont(res, res, p2); | |
533 | ||
534 | ecp_nistz256_sqr_mont(res, res); | |
535 | ecp_nistz256_sqr_mont(res, res); | |
536 | ecp_nistz256_mul_mont(res, res, in); | |
537 | ||
538 | memcpy(r, res, sizeof(res)); | |
539 | } | |
540 | ||
20728adc AP |
541 | /* |
542 | * ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and | |
543 | * returns one if it fits. Otherwise it returns zero. | |
544 | */ | |
5956b110 EK |
545 | __owur static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS], |
546 | const BIGNUM *in) | |
4d3fa06f | 547 | { |
5784a521 | 548 | return bn_copy_words(out, in, P256_LIMBS); |
4d3fa06f AP |
549 | } |
550 | ||
551 | /* r = sum(scalar[i]*point[i]) */ | |
5956b110 EK |
552 | __owur static int ecp_nistz256_windowed_mul(const EC_GROUP *group, |
553 | P256_POINT *r, | |
554 | const BIGNUM **scalar, | |
555 | const EC_POINT **point, | |
556 | size_t num, BN_CTX *ctx) | |
4d3fa06f | 557 | { |
5afc296a | 558 | size_t i; |
a4d5269e | 559 | int j, ret = 0; |
49b05c7d | 560 | unsigned int idx; |
4d3fa06f AP |
561 | unsigned char (*p_str)[33] = NULL; |
562 | const unsigned int window_size = 5; | |
563 | const unsigned int mask = (1 << (window_size + 1)) - 1; | |
564 | unsigned int wvalue; | |
20728adc | 565 | P256_POINT *temp; /* place for 5 temporary points */ |
4d3fa06f | 566 | const BIGNUM **scalars = NULL; |
20728adc | 567 | P256_POINT (*table)[16] = NULL; |
4d3fa06f AP |
568 | void *table_storage = NULL; |
569 | ||
5afc296a AP |
570 | if ((num * 16 + 6) > OPENSSL_MALLOC_MAX_NELEMS(P256_POINT) |
571 | || (table_storage = | |
572 | OPENSSL_malloc((num * 16 + 5) * sizeof(P256_POINT) + 64)) == NULL | |
4d3fa06f AP |
573 | || (p_str = |
574 | OPENSSL_malloc(num * 33 * sizeof(unsigned char))) == NULL | |
575 | || (scalars = OPENSSL_malloc(num * sizeof(BIGNUM *))) == NULL) { | |
be07ae9b | 576 | ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_MALLOC_FAILURE); |
4d3fa06f | 577 | goto err; |
4d3fa06f AP |
578 | } |
579 | ||
3ff08e1d | 580 | table = (void *)ALIGNPTR(table_storage, 64); |
20728adc | 581 | temp = (P256_POINT *)(table + num); |
3ff08e1d | 582 | |
4d3fa06f AP |
583 | for (i = 0; i < num; i++) { |
584 | P256_POINT *row = table[i]; | |
585 | ||
c028254b | 586 | /* This is an unusual input, we don't guarantee constant-timeness. */ |
4d3fa06f AP |
587 | if ((BN_num_bits(scalar[i]) > 256) || BN_is_negative(scalar[i])) { |
588 | BIGNUM *mod; | |
589 | ||
590 | if ((mod = BN_CTX_get(ctx)) == NULL) | |
591 | goto err; | |
5784a521 | 592 | if (!BN_nnmod(mod, scalar[i], group->order, ctx)) { |
be07ae9b | 593 | ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_BN_LIB); |
4d3fa06f AP |
594 | goto err; |
595 | } | |
596 | scalars[i] = mod; | |
597 | } else | |
598 | scalars[i] = scalar[i]; | |
599 | ||
5784a521 MC |
600 | for (j = 0; j < bn_get_top(scalars[i]) * BN_BYTES; j += BN_BYTES) { |
601 | BN_ULONG d = bn_get_words(scalars[i])[j / BN_BYTES]; | |
4d3fa06f | 602 | |
5afc296a AP |
603 | p_str[i][j + 0] = (unsigned char)d; |
604 | p_str[i][j + 1] = (unsigned char)(d >> 8); | |
605 | p_str[i][j + 2] = (unsigned char)(d >> 16); | |
606 | p_str[i][j + 3] = (unsigned char)(d >>= 24); | |
4d3fa06f AP |
607 | if (BN_BYTES == 8) { |
608 | d >>= 8; | |
5afc296a AP |
609 | p_str[i][j + 4] = (unsigned char)d; |
610 | p_str[i][j + 5] = (unsigned char)(d >> 8); | |
611 | p_str[i][j + 6] = (unsigned char)(d >> 16); | |
612 | p_str[i][j + 7] = (unsigned char)(d >> 24); | |
4d3fa06f AP |
613 | } |
614 | } | |
615 | for (; j < 33; j++) | |
616 | p_str[i][j] = 0; | |
617 | ||
5784a521 MC |
618 | if (!ecp_nistz256_bignum_to_field_elem(temp[0].X, point[i]->X) |
619 | || !ecp_nistz256_bignum_to_field_elem(temp[0].Y, point[i]->Y) | |
620 | || !ecp_nistz256_bignum_to_field_elem(temp[0].Z, point[i]->Z)) { | |
58d47cf0 AP |
621 | ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, |
622 | EC_R_COORDINATES_OUT_OF_RANGE); | |
4d3fa06f AP |
623 | goto err; |
624 | } | |
625 | ||
20728adc AP |
626 | /* |
627 | * row[0] is implicitly (0,0,0) (the point at infinity), therefore it | |
628 | * is not stored. All other values are actually stored with an offset | |
629 | * of -1 in table. | |
3ff08e1d AP |
630 | */ |
631 | ||
632 | ecp_nistz256_scatter_w5 (row, &temp[0], 1); | |
633 | ecp_nistz256_point_double(&temp[1], &temp[0]); /*1+1=2 */ | |
634 | ecp_nistz256_scatter_w5 (row, &temp[1], 2); | |
635 | ecp_nistz256_point_add (&temp[2], &temp[1], &temp[0]); /*2+1=3 */ | |
636 | ecp_nistz256_scatter_w5 (row, &temp[2], 3); | |
637 | ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*2=4 */ | |
638 | ecp_nistz256_scatter_w5 (row, &temp[1], 4); | |
639 | ecp_nistz256_point_double(&temp[2], &temp[2]); /*2*3=6 */ | |
640 | ecp_nistz256_scatter_w5 (row, &temp[2], 6); | |
641 | ecp_nistz256_point_add (&temp[3], &temp[1], &temp[0]); /*4+1=5 */ | |
642 | ecp_nistz256_scatter_w5 (row, &temp[3], 5); | |
643 | ecp_nistz256_point_add (&temp[4], &temp[2], &temp[0]); /*6+1=7 */ | |
644 | ecp_nistz256_scatter_w5 (row, &temp[4], 7); | |
645 | ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*4=8 */ | |
646 | ecp_nistz256_scatter_w5 (row, &temp[1], 8); | |
647 | ecp_nistz256_point_double(&temp[2], &temp[2]); /*2*6=12 */ | |
648 | ecp_nistz256_scatter_w5 (row, &temp[2], 12); | |
649 | ecp_nistz256_point_double(&temp[3], &temp[3]); /*2*5=10 */ | |
650 | ecp_nistz256_scatter_w5 (row, &temp[3], 10); | |
651 | ecp_nistz256_point_double(&temp[4], &temp[4]); /*2*7=14 */ | |
652 | ecp_nistz256_scatter_w5 (row, &temp[4], 14); | |
653 | ecp_nistz256_point_add (&temp[2], &temp[2], &temp[0]); /*12+1=13*/ | |
654 | ecp_nistz256_scatter_w5 (row, &temp[2], 13); | |
655 | ecp_nistz256_point_add (&temp[3], &temp[3], &temp[0]); /*10+1=11*/ | |
656 | ecp_nistz256_scatter_w5 (row, &temp[3], 11); | |
657 | ecp_nistz256_point_add (&temp[4], &temp[4], &temp[0]); /*14+1=15*/ | |
658 | ecp_nistz256_scatter_w5 (row, &temp[4], 15); | |
659 | ecp_nistz256_point_add (&temp[2], &temp[1], &temp[0]); /*8+1=9 */ | |
660 | ecp_nistz256_scatter_w5 (row, &temp[2], 9); | |
661 | ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*8=16 */ | |
662 | ecp_nistz256_scatter_w5 (row, &temp[1], 16); | |
4d3fa06f AP |
663 | } |
664 | ||
49b05c7d | 665 | idx = 255; |
4d3fa06f | 666 | |
49b05c7d RS |
667 | wvalue = p_str[0][(idx - 1) / 8]; |
668 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
4d3fa06f | 669 | |
3ff08e1d AP |
670 | /* |
671 | * We gather to temp[0], because we know it's position relative | |
672 | * to table | |
673 | */ | |
674 | ecp_nistz256_gather_w5(&temp[0], table[0], _booth_recode_w5(wvalue) >> 1); | |
675 | memcpy(r, &temp[0], sizeof(temp[0])); | |
4d3fa06f | 676 | |
49b05c7d RS |
677 | while (idx >= 5) { |
678 | for (i = (idx == 255 ? 1 : 0); i < num; i++) { | |
679 | unsigned int off = (idx - 1) / 8; | |
4d3fa06f AP |
680 | |
681 | wvalue = p_str[i][off] | p_str[i][off + 1] << 8; | |
49b05c7d | 682 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; |
4d3fa06f AP |
683 | |
684 | wvalue = _booth_recode_w5(wvalue); | |
685 | ||
3ff08e1d | 686 | ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1); |
4d3fa06f | 687 | |
3ff08e1d AP |
688 | ecp_nistz256_neg(temp[1].Y, temp[0].Y); |
689 | copy_conditional(temp[0].Y, temp[1].Y, (wvalue & 1)); | |
4d3fa06f | 690 | |
3ff08e1d | 691 | ecp_nistz256_point_add(r, r, &temp[0]); |
4d3fa06f AP |
692 | } |
693 | ||
49b05c7d | 694 | idx -= window_size; |
4d3fa06f AP |
695 | |
696 | ecp_nistz256_point_double(r, r); | |
697 | ecp_nistz256_point_double(r, r); | |
698 | ecp_nistz256_point_double(r, r); | |
699 | ecp_nistz256_point_double(r, r); | |
700 | ecp_nistz256_point_double(r, r); | |
701 | } | |
702 | ||
703 | /* Final window */ | |
704 | for (i = 0; i < num; i++) { | |
705 | wvalue = p_str[i][0]; | |
706 | wvalue = (wvalue << 1) & mask; | |
707 | ||
708 | wvalue = _booth_recode_w5(wvalue); | |
709 | ||
3ff08e1d | 710 | ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1); |
4d3fa06f | 711 | |
3ff08e1d AP |
712 | ecp_nistz256_neg(temp[1].Y, temp[0].Y); |
713 | copy_conditional(temp[0].Y, temp[1].Y, wvalue & 1); | |
4d3fa06f | 714 | |
3ff08e1d | 715 | ecp_nistz256_point_add(r, r, &temp[0]); |
4d3fa06f AP |
716 | } |
717 | ||
a4d5269e | 718 | ret = 1; |
58d47cf0 | 719 | err: |
b548a1f1 RS |
720 | OPENSSL_free(table_storage); |
721 | OPENSSL_free(p_str); | |
722 | OPENSSL_free(scalars); | |
a4d5269e | 723 | return ret; |
4d3fa06f AP |
724 | } |
725 | ||
726 | /* Coordinates of G, for which we have precomputed tables */ | |
727 | const static BN_ULONG def_xG[P256_LIMBS] = { | |
728 | TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601), | |
729 | TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6) | |
730 | }; | |
731 | ||
732 | const static BN_ULONG def_yG[P256_LIMBS] = { | |
733 | TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c), | |
734 | TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85) | |
735 | }; | |
736 | ||
20728adc AP |
737 | /* |
738 | * ecp_nistz256_is_affine_G returns one if |generator| is the standard, P-256 | |
739 | * generator. | |
740 | */ | |
58d47cf0 | 741 | static int ecp_nistz256_is_affine_G(const EC_POINT *generator) |
4d3fa06f | 742 | { |
5784a521 MC |
743 | return (bn_get_top(generator->X) == P256_LIMBS) && |
744 | (bn_get_top(generator->Y) == P256_LIMBS) && | |
745 | (bn_get_top(generator->Z) == (P256_LIMBS - P256_LIMBS / 8)) && | |
746 | is_equal(bn_get_words(generator->X), def_xG) && | |
747 | is_equal(bn_get_words(generator->Y), def_yG) && | |
748 | is_one(bn_get_words(generator->Z)); | |
4d3fa06f AP |
749 | } |
750 | ||
5956b110 | 751 | __owur static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx) |
4d3fa06f | 752 | { |
20728adc AP |
753 | /* |
754 | * We precompute a table for a Booth encoded exponent (wNAF) based | |
4d3fa06f | 755 | * computation. Each table holds 64 values for safe access, with an |
20728adc AP |
756 | * implicit value of infinity at index zero. We use window of size 7, and |
757 | * therefore require ceil(256/7) = 37 tables. | |
758 | */ | |
be2e334f | 759 | const BIGNUM *order; |
4d3fa06f AP |
760 | EC_POINT *P = NULL, *T = NULL; |
761 | const EC_POINT *generator; | |
3aef36ff | 762 | NISTZ256_PRE_COMP *pre_comp; |
53dd4ddf | 763 | BN_CTX *new_ctx = NULL; |
4d3fa06f AP |
764 | int i, j, k, ret = 0; |
765 | size_t w; | |
766 | ||
767 | PRECOMP256_ROW *preComputedTable = NULL; | |
768 | unsigned char *precomp_storage = NULL; | |
769 | ||
3aef36ff | 770 | /* if there is an old NISTZ256_PRE_COMP object, throw it away */ |
2c52ac9b | 771 | EC_pre_comp_free(group); |
4d3fa06f AP |
772 | generator = EC_GROUP_get0_generator(group); |
773 | if (generator == NULL) { | |
be07ae9b | 774 | ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNDEFINED_GENERATOR); |
4d3fa06f AP |
775 | return 0; |
776 | } | |
777 | ||
778 | if (ecp_nistz256_is_affine_G(generator)) { | |
20728adc AP |
779 | /* |
780 | * No need to calculate tables for the standard generator because we | |
781 | * have them statically. | |
782 | */ | |
4d3fa06f AP |
783 | return 1; |
784 | } | |
785 | ||
be07ae9b | 786 | if ((pre_comp = ecp_nistz256_pre_comp_new(group)) == NULL) |
4d3fa06f AP |
787 | return 0; |
788 | ||
789 | if (ctx == NULL) { | |
53dd4ddf | 790 | ctx = new_ctx = BN_CTX_new(); |
4d3fa06f AP |
791 | if (ctx == NULL) |
792 | goto err; | |
793 | } | |
794 | ||
795 | BN_CTX_start(ctx); | |
4d3fa06f | 796 | |
be2e334f | 797 | order = EC_GROUP_get0_order(group); |
4d3fa06f AP |
798 | if (order == NULL) |
799 | goto err; | |
800 | ||
4d3fa06f | 801 | if (BN_is_zero(order)) { |
be07ae9b | 802 | ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNKNOWN_ORDER); |
4d3fa06f AP |
803 | goto err; |
804 | } | |
805 | ||
806 | w = 7; | |
807 | ||
808 | if ((precomp_storage = | |
809 | OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE) + 64)) == NULL) { | |
be07ae9b | 810 | ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, ERR_R_MALLOC_FAILURE); |
4d3fa06f | 811 | goto err; |
4d3fa06f AP |
812 | } |
813 | ||
3ff08e1d AP |
814 | preComputedTable = (void *)ALIGNPTR(precomp_storage, 64); |
815 | ||
4d3fa06f AP |
816 | P = EC_POINT_new(group); |
817 | T = EC_POINT_new(group); | |
53dd4ddf EK |
818 | if (P == NULL || T == NULL) |
819 | goto err; | |
4d3fa06f | 820 | |
20728adc AP |
821 | /* |
822 | * The zero entry is implicitly infinity, and we skip it, storing other | |
823 | * values with -1 offset. | |
824 | */ | |
53dd4ddf EK |
825 | if (!EC_POINT_copy(T, generator)) |
826 | goto err; | |
4d3fa06f AP |
827 | |
828 | for (k = 0; k < 64; k++) { | |
53dd4ddf EK |
829 | if (!EC_POINT_copy(P, T)) |
830 | goto err; | |
4d3fa06f | 831 | for (j = 0; j < 37; j++) { |
3ff08e1d | 832 | P256_POINT_AFFINE temp; |
20728adc | 833 | /* |
6038354c | 834 | * It would be faster to use EC_POINTs_make_affine and |
20728adc AP |
835 | * make multiple points affine at the same time. |
836 | */ | |
6038354c EK |
837 | if (!EC_POINT_make_affine(group, P, ctx)) |
838 | goto err; | |
839 | if (!ecp_nistz256_bignum_to_field_elem(temp.X, P->X) || | |
840 | !ecp_nistz256_bignum_to_field_elem(temp.Y, P->Y)) { | |
841 | ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, | |
842 | EC_R_COORDINATES_OUT_OF_RANGE); | |
843 | goto err; | |
844 | } | |
3ff08e1d | 845 | ecp_nistz256_scatter_w7(preComputedTable[j], &temp, k); |
6038354c EK |
846 | for (i = 0; i < 7; i++) { |
847 | if (!EC_POINT_dbl(group, P, P, ctx)) | |
848 | goto err; | |
849 | } | |
4d3fa06f | 850 | } |
6038354c EK |
851 | if (!EC_POINT_add(group, T, T, generator, ctx)) |
852 | goto err; | |
4d3fa06f AP |
853 | } |
854 | ||
855 | pre_comp->group = group; | |
856 | pre_comp->w = w; | |
857 | pre_comp->precomp = preComputedTable; | |
858 | pre_comp->precomp_storage = precomp_storage; | |
4d3fa06f | 859 | precomp_storage = NULL; |
3aef36ff | 860 | SETPRECOMP(group, nistz256, pre_comp); |
4d3fa06f | 861 | pre_comp = NULL; |
4d3fa06f AP |
862 | ret = 1; |
863 | ||
58d47cf0 | 864 | err: |
4d3fa06f AP |
865 | if (ctx != NULL) |
866 | BN_CTX_end(ctx); | |
53dd4ddf EK |
867 | BN_CTX_free(new_ctx); |
868 | ||
3aef36ff | 869 | EC_nistz256_pre_comp_free(pre_comp); |
b548a1f1 | 870 | OPENSSL_free(precomp_storage); |
8fdc3734 RS |
871 | EC_POINT_free(P); |
872 | EC_POINT_free(T); | |
4d3fa06f AP |
873 | return ret; |
874 | } | |
875 | ||
876 | /* | |
877 | * Note that by default ECP_NISTZ256_AVX2 is undefined. While it's great | |
878 | * code processing 4 points in parallel, corresponding serial operation | |
879 | * is several times slower, because it uses 29x29=58-bit multiplication | |
880 | * as opposite to 64x64=128-bit in integer-only scalar case. As result | |
881 | * it doesn't provide *significant* performance improvement. Note that | |
882 | * just defining ECP_NISTZ256_AVX2 is not sufficient to make it work, | |
883 | * you'd need to compile even asm/ecp_nistz256-avx.pl module. | |
884 | */ | |
885 | #if defined(ECP_NISTZ256_AVX2) | |
3ff08e1d | 886 | # if !(defined(__x86_64) || defined(__x86_64__) || \ |
4d3fa06f AP |
887 | defined(_M_AMD64) || defined(_MX64)) || \ |
888 | !(defined(__GNUC__) || defined(_MSC_VER)) /* this is for ALIGN32 */ | |
889 | # undef ECP_NISTZ256_AVX2 | |
890 | # else | |
891 | /* Constant time access, loading four values, from four consecutive tables */ | |
58d47cf0 AP |
892 | void ecp_nistz256_avx2_multi_gather_w7(void *result, const void *in, |
893 | int index0, int index1, int index2, | |
894 | int index3); | |
4d3fa06f AP |
895 | void ecp_nistz256_avx2_transpose_convert(void *RESULTx4, const void *in); |
896 | void ecp_nistz256_avx2_convert_transpose_back(void *result, const void *Ax4); | |
897 | void ecp_nistz256_avx2_point_add_affine_x4(void *RESULTx4, const void *Ax4, | |
898 | const void *Bx4); | |
899 | void ecp_nistz256_avx2_point_add_affines_x4(void *RESULTx4, const void *Ax4, | |
900 | const void *Bx4); | |
901 | void ecp_nistz256_avx2_to_mont(void *RESULTx4, const void *Ax4); | |
902 | void ecp_nistz256_avx2_from_mont(void *RESULTx4, const void *Ax4); | |
903 | void ecp_nistz256_avx2_set1(void *RESULTx4); | |
904 | int ecp_nistz_avx2_eligible(void); | |
905 | ||
906 | static void booth_recode_w7(unsigned char *sign, | |
907 | unsigned char *digit, unsigned char in) | |
908 | { | |
909 | unsigned char s, d; | |
910 | ||
911 | s = ~((in >> 7) - 1); | |
912 | d = (1 << 8) - in - 1; | |
913 | d = (d & s) | (in & ~s); | |
914 | d = (d >> 1) + (d & 1); | |
915 | ||
916 | *sign = s & 1; | |
917 | *digit = d; | |
918 | } | |
919 | ||
20728adc AP |
920 | /* |
921 | * ecp_nistz256_avx2_mul_g performs multiplication by G, using only the | |
4d3fa06f | 922 | * precomputed table. It does 4 affine point additions in parallel, |
20728adc AP |
923 | * significantly speeding up point multiplication for a fixed value. |
924 | */ | |
58d47cf0 | 925 | static void ecp_nistz256_avx2_mul_g(P256_POINT *r, |
4d3fa06f | 926 | unsigned char p_str[33], |
58d47cf0 | 927 | const P256_POINT_AFFINE(*preComputedTable)[64]) |
4d3fa06f AP |
928 | { |
929 | const unsigned int window_size = 7; | |
930 | const unsigned int mask = (1 << (window_size + 1)) - 1; | |
931 | unsigned int wvalue; | |
932 | /* Using 4 windows at a time */ | |
933 | unsigned char sign0, digit0; | |
934 | unsigned char sign1, digit1; | |
935 | unsigned char sign2, digit2; | |
936 | unsigned char sign3, digit3; | |
49b05c7d | 937 | unsigned int idx = 0; |
4d3fa06f AP |
938 | BN_ULONG tmp[P256_LIMBS]; |
939 | int i; | |
940 | ||
941 | ALIGN32 BN_ULONG aX4[4 * 9 * 3] = { 0 }; | |
942 | ALIGN32 BN_ULONG bX4[4 * 9 * 2] = { 0 }; | |
3ff08e1d AP |
943 | ALIGN32 P256_POINT_AFFINE point_arr[4]; |
944 | ALIGN32 P256_POINT res_point_arr[4]; | |
4d3fa06f AP |
945 | |
946 | /* Initial four windows */ | |
947 | wvalue = *((u16 *) & p_str[0]); | |
948 | wvalue = (wvalue << 1) & mask; | |
49b05c7d | 949 | idx += window_size; |
4d3fa06f | 950 | booth_recode_w7(&sign0, &digit0, wvalue); |
49b05c7d RS |
951 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
952 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
953 | idx += window_size; | |
4d3fa06f | 954 | booth_recode_w7(&sign1, &digit1, wvalue); |
49b05c7d RS |
955 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
956 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
957 | idx += window_size; | |
4d3fa06f | 958 | booth_recode_w7(&sign2, &digit2, wvalue); |
49b05c7d RS |
959 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
960 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
961 | idx += window_size; | |
4d3fa06f AP |
962 | booth_recode_w7(&sign3, &digit3, wvalue); |
963 | ||
3ff08e1d | 964 | ecp_nistz256_avx2_multi_gather_w7(point_arr, preComputedTable[0], |
4d3fa06f AP |
965 | digit0, digit1, digit2, digit3); |
966 | ||
967 | ecp_nistz256_neg(tmp, point_arr[0].Y); | |
968 | copy_conditional(point_arr[0].Y, tmp, sign0); | |
969 | ecp_nistz256_neg(tmp, point_arr[1].Y); | |
970 | copy_conditional(point_arr[1].Y, tmp, sign1); | |
971 | ecp_nistz256_neg(tmp, point_arr[2].Y); | |
972 | copy_conditional(point_arr[2].Y, tmp, sign2); | |
973 | ecp_nistz256_neg(tmp, point_arr[3].Y); | |
974 | copy_conditional(point_arr[3].Y, tmp, sign3); | |
975 | ||
976 | ecp_nistz256_avx2_transpose_convert(aX4, point_arr); | |
977 | ecp_nistz256_avx2_to_mont(aX4, aX4); | |
978 | ecp_nistz256_avx2_to_mont(&aX4[4 * 9], &aX4[4 * 9]); | |
979 | ecp_nistz256_avx2_set1(&aX4[4 * 9 * 2]); | |
980 | ||
49b05c7d RS |
981 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
982 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
983 | idx += window_size; | |
4d3fa06f | 984 | booth_recode_w7(&sign0, &digit0, wvalue); |
49b05c7d RS |
985 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
986 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
987 | idx += window_size; | |
4d3fa06f | 988 | booth_recode_w7(&sign1, &digit1, wvalue); |
49b05c7d RS |
989 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
990 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
991 | idx += window_size; | |
4d3fa06f | 992 | booth_recode_w7(&sign2, &digit2, wvalue); |
49b05c7d RS |
993 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
994 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
995 | idx += window_size; | |
4d3fa06f AP |
996 | booth_recode_w7(&sign3, &digit3, wvalue); |
997 | ||
3ff08e1d | 998 | ecp_nistz256_avx2_multi_gather_w7(point_arr, preComputedTable[4 * 1], |
4d3fa06f AP |
999 | digit0, digit1, digit2, digit3); |
1000 | ||
1001 | ecp_nistz256_neg(tmp, point_arr[0].Y); | |
1002 | copy_conditional(point_arr[0].Y, tmp, sign0); | |
1003 | ecp_nistz256_neg(tmp, point_arr[1].Y); | |
1004 | copy_conditional(point_arr[1].Y, tmp, sign1); | |
1005 | ecp_nistz256_neg(tmp, point_arr[2].Y); | |
1006 | copy_conditional(point_arr[2].Y, tmp, sign2); | |
1007 | ecp_nistz256_neg(tmp, point_arr[3].Y); | |
1008 | copy_conditional(point_arr[3].Y, tmp, sign3); | |
1009 | ||
1010 | ecp_nistz256_avx2_transpose_convert(bX4, point_arr); | |
1011 | ecp_nistz256_avx2_to_mont(bX4, bX4); | |
1012 | ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]); | |
1013 | /* Optimized when both inputs are affine */ | |
1014 | ecp_nistz256_avx2_point_add_affines_x4(aX4, aX4, bX4); | |
1015 | ||
1016 | for (i = 2; i < 9; i++) { | |
49b05c7d RS |
1017 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
1018 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
1019 | idx += window_size; | |
4d3fa06f | 1020 | booth_recode_w7(&sign0, &digit0, wvalue); |
49b05c7d RS |
1021 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
1022 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
1023 | idx += window_size; | |
4d3fa06f | 1024 | booth_recode_w7(&sign1, &digit1, wvalue); |
49b05c7d RS |
1025 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
1026 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
1027 | idx += window_size; | |
4d3fa06f | 1028 | booth_recode_w7(&sign2, &digit2, wvalue); |
49b05c7d RS |
1029 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
1030 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
1031 | idx += window_size; | |
4d3fa06f AP |
1032 | booth_recode_w7(&sign3, &digit3, wvalue); |
1033 | ||
3ff08e1d | 1034 | ecp_nistz256_avx2_multi_gather_w7(point_arr, |
4d3fa06f AP |
1035 | preComputedTable[4 * i], |
1036 | digit0, digit1, digit2, digit3); | |
1037 | ||
1038 | ecp_nistz256_neg(tmp, point_arr[0].Y); | |
1039 | copy_conditional(point_arr[0].Y, tmp, sign0); | |
1040 | ecp_nistz256_neg(tmp, point_arr[1].Y); | |
1041 | copy_conditional(point_arr[1].Y, tmp, sign1); | |
1042 | ecp_nistz256_neg(tmp, point_arr[2].Y); | |
1043 | copy_conditional(point_arr[2].Y, tmp, sign2); | |
1044 | ecp_nistz256_neg(tmp, point_arr[3].Y); | |
1045 | copy_conditional(point_arr[3].Y, tmp, sign3); | |
1046 | ||
1047 | ecp_nistz256_avx2_transpose_convert(bX4, point_arr); | |
1048 | ecp_nistz256_avx2_to_mont(bX4, bX4); | |
1049 | ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]); | |
1050 | ||
1051 | ecp_nistz256_avx2_point_add_affine_x4(aX4, aX4, bX4); | |
1052 | } | |
1053 | ||
1054 | ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 0], &aX4[4 * 9 * 0]); | |
1055 | ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 1], &aX4[4 * 9 * 1]); | |
1056 | ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 2], &aX4[4 * 9 * 2]); | |
1057 | ||
1058 | ecp_nistz256_avx2_convert_transpose_back(res_point_arr, aX4); | |
1059 | /* Last window is performed serially */ | |
49b05c7d RS |
1060 | wvalue = *((u16 *) & p_str[(idx - 1) / 8]); |
1061 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; | |
4d3fa06f | 1062 | booth_recode_w7(&sign0, &digit0, wvalue); |
58d47cf0 AP |
1063 | ecp_nistz256_gather_w7((P256_POINT_AFFINE *)r, |
1064 | preComputedTable[36], digit0); | |
4d3fa06f AP |
1065 | ecp_nistz256_neg(tmp, r->Y); |
1066 | copy_conditional(r->Y, tmp, sign0); | |
1067 | memcpy(r->Z, ONE, sizeof(ONE)); | |
1068 | /* Sum the four windows */ | |
1069 | ecp_nistz256_point_add(r, r, &res_point_arr[0]); | |
1070 | ecp_nistz256_point_add(r, r, &res_point_arr[1]); | |
1071 | ecp_nistz256_point_add(r, r, &res_point_arr[2]); | |
1072 | ecp_nistz256_point_add(r, r, &res_point_arr[3]); | |
1073 | } | |
1074 | # endif | |
1075 | #endif | |
1076 | ||
5956b110 EK |
1077 | __owur static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group, |
1078 | const P256_POINT_AFFINE *in, | |
1079 | BN_CTX *ctx) | |
4d3fa06f | 1080 | { |
5784a521 | 1081 | BIGNUM *x, *y; |
4d3fa06f AP |
1082 | BN_ULONG d_x[P256_LIMBS], d_y[P256_LIMBS]; |
1083 | int ret = 0; | |
1084 | ||
5784a521 | 1085 | x = BN_new(); |
90945fa3 | 1086 | if (x == NULL) |
5784a521 MC |
1087 | return 0; |
1088 | y = BN_new(); | |
90945fa3 | 1089 | if (y == NULL) { |
5784a521 MC |
1090 | BN_free(x); |
1091 | return 0; | |
1092 | } | |
4d3fa06f | 1093 | memcpy(d_x, in->X, sizeof(d_x)); |
5784a521 | 1094 | bn_set_static_words(x, d_x, P256_LIMBS); |
4d3fa06f AP |
1095 | |
1096 | memcpy(d_y, in->Y, sizeof(d_y)); | |
5784a521 MC |
1097 | bn_set_static_words(y, d_y, P256_LIMBS); |
1098 | ||
1099 | ret = EC_POINT_set_affine_coordinates_GFp(group, out, x, y, ctx); | |
4d3fa06f | 1100 | |
23a1d5e9 RS |
1101 | BN_free(x); |
1102 | BN_free(y); | |
4d3fa06f AP |
1103 | |
1104 | return ret; | |
1105 | } | |
1106 | ||
1107 | /* r = scalar*G + sum(scalars[i]*points[i]) */ | |
5956b110 EK |
1108 | __owur static int ecp_nistz256_points_mul(const EC_GROUP *group, |
1109 | EC_POINT *r, | |
1110 | const BIGNUM *scalar, | |
1111 | size_t num, | |
1112 | const EC_POINT *points[], | |
1113 | const BIGNUM *scalars[], BN_CTX *ctx) | |
4d3fa06f AP |
1114 | { |
1115 | int i = 0, ret = 0, no_precomp_for_generator = 0, p_is_infinity = 0; | |
16e5b45f | 1116 | size_t j; |
4d3fa06f AP |
1117 | unsigned char p_str[33] = { 0 }; |
1118 | const PRECOMP256_ROW *preComputedTable = NULL; | |
3aef36ff | 1119 | const NISTZ256_PRE_COMP *pre_comp = NULL; |
4d3fa06f | 1120 | const EC_POINT *generator = NULL; |
e22d2199 | 1121 | BN_CTX *new_ctx = NULL; |
a4d5269e EK |
1122 | const BIGNUM **new_scalars = NULL; |
1123 | const EC_POINT **new_points = NULL; | |
49b05c7d | 1124 | unsigned int idx = 0; |
4d3fa06f AP |
1125 | const unsigned int window_size = 7; |
1126 | const unsigned int mask = (1 << (window_size + 1)) - 1; | |
1127 | unsigned int wvalue; | |
1128 | ALIGN32 union { | |
1129 | P256_POINT p; | |
1130 | P256_POINT_AFFINE a; | |
1131 | } t, p; | |
1132 | BIGNUM *tmp_scalar; | |
1133 | ||
58d47cf0 | 1134 | if ((num + 1) == 0 || (num + 1) > OPENSSL_MALLOC_MAX_NELEMS(void *)) { |
3ff08e1d AP |
1135 | ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE); |
1136 | return 0; | |
1137 | } | |
1138 | ||
4d3fa06f | 1139 | if (group->meth != r->meth) { |
be07ae9b | 1140 | ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS); |
4d3fa06f AP |
1141 | return 0; |
1142 | } | |
e22d2199 | 1143 | |
4d3fa06f AP |
1144 | if ((scalar == NULL) && (num == 0)) |
1145 | return EC_POINT_set_to_infinity(group, r); | |
1146 | ||
16e5b45f DSH |
1147 | for (j = 0; j < num; j++) { |
1148 | if (group->meth != points[j]->meth) { | |
be07ae9b | 1149 | ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS); |
4d3fa06f AP |
1150 | return 0; |
1151 | } | |
1152 | } | |
1153 | ||
e22d2199 EK |
1154 | if (ctx == NULL) { |
1155 | ctx = new_ctx = BN_CTX_new(); | |
1156 | if (ctx == NULL) | |
1157 | goto err; | |
1158 | } | |
1159 | ||
1160 | BN_CTX_start(ctx); | |
4d3fa06f AP |
1161 | |
1162 | if (scalar) { | |
1163 | generator = EC_GROUP_get0_generator(group); | |
1164 | if (generator == NULL) { | |
be07ae9b | 1165 | ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_UNDEFINED_GENERATOR); |
4d3fa06f AP |
1166 | goto err; |
1167 | } | |
1168 | ||
1169 | /* look if we can use precomputed multiples of generator */ | |
3aef36ff | 1170 | pre_comp = group->pre_comp.nistz256; |
4d3fa06f AP |
1171 | |
1172 | if (pre_comp) { | |
20728adc AP |
1173 | /* |
1174 | * If there is a precomputed table for the generator, check that | |
1175 | * it was generated with the same generator. | |
1176 | */ | |
4d3fa06f AP |
1177 | EC_POINT *pre_comp_generator = EC_POINT_new(group); |
1178 | if (pre_comp_generator == NULL) | |
1179 | goto err; | |
1180 | ||
3ff08e1d AP |
1181 | if (!ecp_nistz256_set_from_affine(pre_comp_generator, |
1182 | group, pre_comp->precomp[0], | |
e22d2199 EK |
1183 | ctx)) { |
1184 | EC_POINT_free(pre_comp_generator); | |
4d3fa06f | 1185 | goto err; |
e22d2199 | 1186 | } |
4d3fa06f AP |
1187 | |
1188 | if (0 == EC_POINT_cmp(group, generator, pre_comp_generator, ctx)) | |
1189 | preComputedTable = (const PRECOMP256_ROW *)pre_comp->precomp; | |
1190 | ||
1191 | EC_POINT_free(pre_comp_generator); | |
1192 | } | |
1193 | ||
1194 | if (preComputedTable == NULL && ecp_nistz256_is_affine_G(generator)) { | |
20728adc AP |
1195 | /* |
1196 | * If there is no precomputed data, but the generator is the | |
1197 | * default, a hardcoded table of precomputed data is used. This | |
1198 | * is because applications, such as Apache, do not use | |
1199 | * EC_KEY_precompute_mult. | |
1200 | */ | |
3ff08e1d | 1201 | preComputedTable = ecp_nistz256_precomputed; |
4d3fa06f AP |
1202 | } |
1203 | ||
1204 | if (preComputedTable) { | |
1205 | if ((BN_num_bits(scalar) > 256) | |
1206 | || BN_is_negative(scalar)) { | |
1207 | if ((tmp_scalar = BN_CTX_get(ctx)) == NULL) | |
1208 | goto err; | |
1209 | ||
5784a521 | 1210 | if (!BN_nnmod(tmp_scalar, scalar, group->order, ctx)) { |
be07ae9b | 1211 | ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_BN_LIB); |
4d3fa06f AP |
1212 | goto err; |
1213 | } | |
1214 | scalar = tmp_scalar; | |
1215 | } | |
1216 | ||
5784a521 MC |
1217 | for (i = 0; i < bn_get_top(scalar) * BN_BYTES; i += BN_BYTES) { |
1218 | BN_ULONG d = bn_get_words(scalar)[i / BN_BYTES]; | |
4d3fa06f | 1219 | |
5afc296a AP |
1220 | p_str[i + 0] = (unsigned char)d; |
1221 | p_str[i + 1] = (unsigned char)(d >> 8); | |
1222 | p_str[i + 2] = (unsigned char)(d >> 16); | |
1223 | p_str[i + 3] = (unsigned char)(d >>= 24); | |
4d3fa06f AP |
1224 | if (BN_BYTES == 8) { |
1225 | d >>= 8; | |
5afc296a AP |
1226 | p_str[i + 4] = (unsigned char)d; |
1227 | p_str[i + 5] = (unsigned char)(d >> 8); | |
1228 | p_str[i + 6] = (unsigned char)(d >> 16); | |
1229 | p_str[i + 7] = (unsigned char)(d >> 24); | |
4d3fa06f AP |
1230 | } |
1231 | } | |
1232 | ||
1233 | for (; i < 33; i++) | |
1234 | p_str[i] = 0; | |
1235 | ||
1236 | #if defined(ECP_NISTZ256_AVX2) | |
1237 | if (ecp_nistz_avx2_eligible()) { | |
1238 | ecp_nistz256_avx2_mul_g(&p.p, p_str, preComputedTable); | |
1239 | } else | |
1240 | #endif | |
1241 | { | |
1242 | /* First window */ | |
1243 | wvalue = (p_str[0] << 1) & mask; | |
49b05c7d | 1244 | idx += window_size; |
4d3fa06f AP |
1245 | |
1246 | wvalue = _booth_recode_w7(wvalue); | |
1247 | ||
58d47cf0 AP |
1248 | ecp_nistz256_gather_w7(&p.a, preComputedTable[0], |
1249 | wvalue >> 1); | |
4d3fa06f AP |
1250 | |
1251 | ecp_nistz256_neg(p.p.Z, p.p.Y); | |
1252 | copy_conditional(p.p.Y, p.p.Z, wvalue & 1); | |
1253 | ||
1254 | memcpy(p.p.Z, ONE, sizeof(ONE)); | |
1255 | ||
1256 | for (i = 1; i < 37; i++) { | |
49b05c7d | 1257 | unsigned int off = (idx - 1) / 8; |
4d3fa06f | 1258 | wvalue = p_str[off] | p_str[off + 1] << 8; |
49b05c7d RS |
1259 | wvalue = (wvalue >> ((idx - 1) % 8)) & mask; |
1260 | idx += window_size; | |
4d3fa06f AP |
1261 | |
1262 | wvalue = _booth_recode_w7(wvalue); | |
1263 | ||
3ff08e1d | 1264 | ecp_nistz256_gather_w7(&t.a, |
4d3fa06f AP |
1265 | preComputedTable[i], wvalue >> 1); |
1266 | ||
1267 | ecp_nistz256_neg(t.p.Z, t.a.Y); | |
1268 | copy_conditional(t.a.Y, t.p.Z, wvalue & 1); | |
1269 | ||
1270 | ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a); | |
1271 | } | |
1272 | } | |
1273 | } else { | |
1274 | p_is_infinity = 1; | |
1275 | no_precomp_for_generator = 1; | |
1276 | } | |
1277 | } else | |
1278 | p_is_infinity = 1; | |
1279 | ||
1280 | if (no_precomp_for_generator) { | |
20728adc AP |
1281 | /* |
1282 | * Without a precomputed table for the generator, it has to be | |
1283 | * handled like a normal point. | |
1284 | */ | |
4d3fa06f | 1285 | new_scalars = OPENSSL_malloc((num + 1) * sizeof(BIGNUM *)); |
90945fa3 | 1286 | if (new_scalars == NULL) { |
be07ae9b | 1287 | ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE); |
e22d2199 | 1288 | goto err; |
4d3fa06f AP |
1289 | } |
1290 | ||
1291 | new_points = OPENSSL_malloc((num + 1) * sizeof(EC_POINT *)); | |
90945fa3 | 1292 | if (new_points == NULL) { |
be07ae9b | 1293 | ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE); |
e22d2199 | 1294 | goto err; |
4d3fa06f AP |
1295 | } |
1296 | ||
1297 | memcpy(new_scalars, scalars, num * sizeof(BIGNUM *)); | |
1298 | new_scalars[num] = scalar; | |
1299 | memcpy(new_points, points, num * sizeof(EC_POINT *)); | |
1300 | new_points[num] = generator; | |
1301 | ||
1302 | scalars = new_scalars; | |
1303 | points = new_points; | |
1304 | num++; | |
1305 | } | |
1306 | ||
1307 | if (num) { | |
1308 | P256_POINT *out = &t.p; | |
1309 | if (p_is_infinity) | |
1310 | out = &p.p; | |
1311 | ||
a4d5269e EK |
1312 | if (!ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx)) |
1313 | goto err; | |
4d3fa06f AP |
1314 | |
1315 | if (!p_is_infinity) | |
1316 | ecp_nistz256_point_add(&p.p, &p.p, out); | |
1317 | } | |
1318 | ||
c028254b | 1319 | /* Not constant-time, but we're only operating on the public output. */ |
e22d2199 EK |
1320 | if (!bn_set_words(r->X, p.p.X, P256_LIMBS) || |
1321 | !bn_set_words(r->Y, p.p.Y, P256_LIMBS) || | |
1322 | !bn_set_words(r->Z, p.p.Z, P256_LIMBS)) { | |
1323 | goto err; | |
1324 | } | |
4446044a | 1325 | r->Z_is_one = is_one(p.p.Z) & 1; |
4d3fa06f AP |
1326 | |
1327 | ret = 1; | |
1328 | ||
e22d2199 EK |
1329 | err: |
1330 | if (ctx) | |
1331 | BN_CTX_end(ctx); | |
1332 | BN_CTX_free(new_ctx); | |
b548a1f1 RS |
1333 | OPENSSL_free(new_points); |
1334 | OPENSSL_free(new_scalars); | |
4d3fa06f AP |
1335 | return ret; |
1336 | } | |
1337 | ||
5956b110 EK |
1338 | __owur static int ecp_nistz256_get_affine(const EC_GROUP *group, |
1339 | const EC_POINT *point, | |
1340 | BIGNUM *x, BIGNUM *y, BN_CTX *ctx) | |
4d3fa06f AP |
1341 | { |
1342 | BN_ULONG z_inv2[P256_LIMBS]; | |
1343 | BN_ULONG z_inv3[P256_LIMBS]; | |
1344 | BN_ULONG x_aff[P256_LIMBS]; | |
1345 | BN_ULONG y_aff[P256_LIMBS]; | |
1346 | BN_ULONG point_x[P256_LIMBS], point_y[P256_LIMBS], point_z[P256_LIMBS]; | |
e22d2199 | 1347 | BN_ULONG x_ret[P256_LIMBS], y_ret[P256_LIMBS]; |
4d3fa06f AP |
1348 | |
1349 | if (EC_POINT_is_at_infinity(group, point)) { | |
be07ae9b | 1350 | ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_POINT_AT_INFINITY); |
4d3fa06f AP |
1351 | return 0; |
1352 | } | |
1353 | ||
5784a521 MC |
1354 | if (!ecp_nistz256_bignum_to_field_elem(point_x, point->X) || |
1355 | !ecp_nistz256_bignum_to_field_elem(point_y, point->Y) || | |
1356 | !ecp_nistz256_bignum_to_field_elem(point_z, point->Z)) { | |
be07ae9b | 1357 | ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_COORDINATES_OUT_OF_RANGE); |
4d3fa06f AP |
1358 | return 0; |
1359 | } | |
1360 | ||
1361 | ecp_nistz256_mod_inverse(z_inv3, point_z); | |
1362 | ecp_nistz256_sqr_mont(z_inv2, z_inv3); | |
1363 | ecp_nistz256_mul_mont(x_aff, z_inv2, point_x); | |
1364 | ||
1365 | if (x != NULL) { | |
e22d2199 EK |
1366 | ecp_nistz256_from_mont(x_ret, x_aff); |
1367 | if (!bn_set_words(x, x_ret, P256_LIMBS)) | |
1368 | return 0; | |
4d3fa06f AP |
1369 | } |
1370 | ||
1371 | if (y != NULL) { | |
1372 | ecp_nistz256_mul_mont(z_inv3, z_inv3, z_inv2); | |
1373 | ecp_nistz256_mul_mont(y_aff, z_inv3, point_y); | |
e22d2199 EK |
1374 | ecp_nistz256_from_mont(y_ret, y_aff); |
1375 | if (!bn_set_words(y, y_ret, P256_LIMBS)) | |
1376 | return 0; | |
4d3fa06f AP |
1377 | } |
1378 | ||
1379 | return 1; | |
1380 | } | |
1381 | ||
3aef36ff | 1382 | static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group) |
4d3fa06f | 1383 | { |
3aef36ff | 1384 | NISTZ256_PRE_COMP *ret = NULL; |
4d3fa06f AP |
1385 | |
1386 | if (!group) | |
1387 | return NULL; | |
1388 | ||
3aef36ff | 1389 | ret = OPENSSL_zalloc(sizeof(*ret)); |
4d3fa06f | 1390 | |
90945fa3 | 1391 | if (ret == NULL) { |
be07ae9b | 1392 | ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE); |
4d3fa06f AP |
1393 | return ret; |
1394 | } | |
1395 | ||
1396 | ret->group = group; | |
1397 | ret->w = 6; /* default */ | |
4d3fa06f AP |
1398 | ret->references = 1; |
1399 | return ret; | |
1400 | } | |
1401 | ||
3aef36ff | 1402 | NISTZ256_PRE_COMP *EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP *p) |
4d3fa06f | 1403 | { |
3aef36ff RS |
1404 | if (p != NULL) |
1405 | CRYPTO_add(&p->references, 1, CRYPTO_LOCK_EC_PRE_COMP); | |
1406 | return p; | |
4d3fa06f AP |
1407 | } |
1408 | ||
3aef36ff | 1409 | void EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP *pre) |
4d3fa06f | 1410 | { |
3aef36ff RS |
1411 | if (pre == NULL |
1412 | || CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP) > 0) | |
4d3fa06f | 1413 | return; |
b548a1f1 | 1414 | OPENSSL_free(pre->precomp_storage); |
4d3fa06f AP |
1415 | OPENSSL_free(pre); |
1416 | } | |
1417 | ||
4d3fa06f | 1418 | |
58d47cf0 | 1419 | static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP *group) |
4d3fa06f AP |
1420 | { |
1421 | /* There is a hard-coded table for the default generator. */ | |
1422 | const EC_POINT *generator = EC_GROUP_get0_generator(group); | |
3aef36ff | 1423 | |
4d3fa06f AP |
1424 | if (generator != NULL && ecp_nistz256_is_affine_G(generator)) { |
1425 | /* There is a hard-coded table for the default generator. */ | |
1426 | return 1; | |
1427 | } | |
1428 | ||
3aef36ff | 1429 | return HAVEPRECOMP(group, nistz256); |
4d3fa06f AP |
1430 | } |
1431 | ||
1432 | const EC_METHOD *EC_GFp_nistz256_method(void) | |
1433 | { | |
1434 | static const EC_METHOD ret = { | |
1435 | EC_FLAGS_DEFAULT_OCT, | |
1436 | NID_X9_62_prime_field, | |
1437 | ec_GFp_mont_group_init, | |
1438 | ec_GFp_mont_group_finish, | |
1439 | ec_GFp_mont_group_clear_finish, | |
1440 | ec_GFp_mont_group_copy, | |
1441 | ec_GFp_mont_group_set_curve, | |
1442 | ec_GFp_simple_group_get_curve, | |
1443 | ec_GFp_simple_group_get_degree, | |
9ff9bccc | 1444 | ec_group_simple_order_bits, |
4d3fa06f AP |
1445 | ec_GFp_simple_group_check_discriminant, |
1446 | ec_GFp_simple_point_init, | |
1447 | ec_GFp_simple_point_finish, | |
1448 | ec_GFp_simple_point_clear_finish, | |
1449 | ec_GFp_simple_point_copy, | |
1450 | ec_GFp_simple_point_set_to_infinity, | |
1451 | ec_GFp_simple_set_Jprojective_coordinates_GFp, | |
1452 | ec_GFp_simple_get_Jprojective_coordinates_GFp, | |
1453 | ec_GFp_simple_point_set_affine_coordinates, | |
1454 | ecp_nistz256_get_affine, | |
1455 | 0, 0, 0, | |
1456 | ec_GFp_simple_add, | |
1457 | ec_GFp_simple_dbl, | |
1458 | ec_GFp_simple_invert, | |
1459 | ec_GFp_simple_is_at_infinity, | |
1460 | ec_GFp_simple_is_on_curve, | |
1461 | ec_GFp_simple_cmp, | |
1462 | ec_GFp_simple_make_affine, | |
1463 | ec_GFp_simple_points_make_affine, | |
1464 | ecp_nistz256_points_mul, /* mul */ | |
1465 | ecp_nistz256_mult_precompute, /* precompute_mult */ | |
1466 | ecp_nistz256_window_have_precompute_mult, /* have_precompute_mult */ | |
1467 | ec_GFp_mont_field_mul, | |
1468 | ec_GFp_mont_field_sqr, | |
1469 | 0, /* field_div */ | |
1470 | ec_GFp_mont_field_encode, | |
1471 | ec_GFp_mont_field_decode, | |
9ff9bccc DSH |
1472 | ec_GFp_mont_field_set_to_one, |
1473 | ec_key_simple_priv2oct, | |
1474 | ec_key_simple_oct2priv, | |
1475 | 0, /* set private */ | |
1476 | ec_key_simple_generate_key, | |
1477 | ec_key_simple_check_key, | |
1478 | ec_key_simple_generate_public_key, | |
1479 | 0, /* keycopy */ | |
1480 | 0, /* keyfinish */ | |
1481 | ecdh_simple_compute_key | |
4d3fa06f AP |
1482 | }; |
1483 | ||
1484 | return &ret; | |
1485 | } |