1 /******************************************************************************
3 * Copyright 2014 Intel Corporation *
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 *
9 * http://www.apache.org/licenses/LICENSE-2.0 *
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. *
17 ******************************************************************************
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 *
24 * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with *
27 ******************************************************************************/
31 #include <openssl/bn.h>
32 #include <openssl/err.h>
33 #include <openssl/ec.h>
39 # define TOBN(hi,lo) lo,hi
41 # define TOBN(hi,lo) ((BN_ULONG)hi<<32|lo)
45 # define ALIGN32 __attribute((aligned(32)))
46 #elif defined(_MSC_VER)
47 # define ALIGN32 __declspec(align(32))
52 #define ALIGNPTR(p,N) ((unsigned char *)p+N-(size_t)p%N)
53 #define P256_LIMBS (256/BN_BITS2)
55 typedef unsigned short u16;
58 BN_ULONG X[P256_LIMBS];
59 BN_ULONG Y[P256_LIMBS];
60 BN_ULONG Z[P256_LIMBS];
64 BN_ULONG X[P256_LIMBS];
65 BN_ULONG Y[P256_LIMBS];
68 typedef P256_POINT_AFFINE PRECOMP256_ROW[64];
70 /* structure for precomputed multiples of the generator */
71 typedef struct ec_pre_comp_st {
72 const EC_GROUP *group; /* Parent EC_GROUP object */
73 size_t w; /* Window size */
75 * Constant time access to the X and Y coordinates of the pre-computed,
76 * generator multiplies, in the Montgomery domain. Pre-calculated
77 * multiplies are stored in affine form.
79 PRECOMP256_ROW *precomp;
80 void *precomp_storage;
84 /* Functions implemented in assembly */
85 /* Modular mul by 2: res = 2*a mod P */
86 void ecp_nistz256_mul_by_2(BN_ULONG res[P256_LIMBS],
87 const BN_ULONG a[P256_LIMBS]);
88 /* Modular div by 2: res = a/2 mod P */
89 void ecp_nistz256_div_by_2(BN_ULONG res[P256_LIMBS],
90 const BN_ULONG a[P256_LIMBS]);
91 /* Modular mul by 3: res = 3*a mod P */
92 void ecp_nistz256_mul_by_3(BN_ULONG res[P256_LIMBS],
93 const BN_ULONG a[P256_LIMBS]);
94 /* Modular add: res = a+b mod P */
95 void ecp_nistz256_add(BN_ULONG res[P256_LIMBS],
96 const BN_ULONG a[P256_LIMBS],
97 const BN_ULONG b[P256_LIMBS]);
98 /* Modular sub: res = a-b mod P */
99 void ecp_nistz256_sub(BN_ULONG res[P256_LIMBS],
100 const BN_ULONG a[P256_LIMBS],
101 const BN_ULONG b[P256_LIMBS]);
102 /* Modular neg: res = -a mod P */
103 void ecp_nistz256_neg(BN_ULONG res[P256_LIMBS], const BN_ULONG a[P256_LIMBS]);
104 /* Montgomery mul: res = a*b*2^-256 mod P */
105 void ecp_nistz256_mul_mont(BN_ULONG res[P256_LIMBS],
106 const BN_ULONG a[P256_LIMBS],
107 const BN_ULONG b[P256_LIMBS]);
108 /* Montgomery sqr: res = a*a*2^-256 mod P */
109 void ecp_nistz256_sqr_mont(BN_ULONG res[P256_LIMBS],
110 const BN_ULONG a[P256_LIMBS]);
111 /* Convert a number from Montgomery domain, by multiplying with 1 */
112 void ecp_nistz256_from_mont(BN_ULONG res[P256_LIMBS],
113 const BN_ULONG in[P256_LIMBS]);
114 /* Convert a number to Montgomery domain, by multiplying with 2^512 mod P*/
115 void ecp_nistz256_to_mont(BN_ULONG res[P256_LIMBS],
116 const BN_ULONG in[P256_LIMBS]);
117 /* Functions that perform constant time access to the precomputed tables */
118 void ecp_nistz256_select_w5(P256_POINT * val,
119 const P256_POINT * in_t, int index);
120 void ecp_nistz256_select_w7(P256_POINT_AFFINE * val,
121 const P256_POINT_AFFINE * in_t, int index);
123 /* One converted into the Montgomery domain */
124 static const BN_ULONG ONE[P256_LIMBS] = {
125 TOBN(0x00000000, 0x00000001), TOBN(0xffffffff, 0x00000000),
126 TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe)
129 static void *ecp_nistz256_pre_comp_dup(void *);
130 static void ecp_nistz256_pre_comp_free(void *);
131 static void ecp_nistz256_pre_comp_clear_free(void *);
132 static EC_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group);
134 /* Precomputed tables for the default generator */
135 #include "ecp_nistz256_table.c"
137 /* Recode window to a signed digit, see ecp_nistputil.c for details */
138 static unsigned int _booth_recode_w5(unsigned int in)
142 s = ~((in >> 5) - 1);
143 d = (1 << 6) - in - 1;
144 d = (d & s) | (in & ~s);
145 d = (d >> 1) + (d & 1);
147 return (d << 1) + (s & 1);
150 static unsigned int _booth_recode_w7(unsigned int in)
154 s = ~((in >> 7) - 1);
155 d = (1 << 8) - in - 1;
156 d = (d & s) | (in & ~s);
157 d = (d >> 1) + (d & 1);
159 return (d << 1) + (s & 1);
162 static void copy_conditional(BN_ULONG dst[P256_LIMBS],
163 const BN_ULONG src[P256_LIMBS], BN_ULONG move)
165 BN_ULONG mask1 = -move;
166 BN_ULONG mask2 = ~mask1;
168 dst[0] = (src[0] & mask1) ^ (dst[0] & mask2);
169 dst[1] = (src[1] & mask1) ^ (dst[1] & mask2);
170 dst[2] = (src[2] & mask1) ^ (dst[2] & mask2);
171 dst[3] = (src[3] & mask1) ^ (dst[3] & mask2);
172 if (P256_LIMBS == 8) {
173 dst[4] = (src[4] & mask1) ^ (dst[4] & mask2);
174 dst[5] = (src[5] & mask1) ^ (dst[5] & mask2);
175 dst[6] = (src[6] & mask1) ^ (dst[6] & mask2);
176 dst[7] = (src[7] & mask1) ^ (dst[7] & mask2);
180 static BN_ULONG is_zero(BN_ULONG in)
189 static BN_ULONG is_equal(const BN_ULONG a[P256_LIMBS],
190 const BN_ULONG b[P256_LIMBS])
198 if (P256_LIMBS == 8) {
208 static BN_ULONG is_one(const BN_ULONG a[P256_LIMBS])
213 res |= a[1] ^ ONE[1];
214 res |= a[2] ^ ONE[2];
215 res |= a[3] ^ ONE[3];
216 if (P256_LIMBS == 8) {
217 res |= a[4] ^ ONE[4];
218 res |= a[5] ^ ONE[5];
219 res |= a[6] ^ ONE[6];
225 #ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
226 void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a);
227 void ecp_nistz256_point_add(P256_POINT *r,
228 const P256_POINT *a, const P256_POINT *b);
229 void ecp_nistz256_point_add_affine(P256_POINT *r,
231 const P256_POINT_AFFINE *b);
233 /* Point double: r = 2*a */
234 static void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a)
236 BN_ULONG S[P256_LIMBS];
237 BN_ULONG M[P256_LIMBS];
238 BN_ULONG Zsqr[P256_LIMBS];
239 BN_ULONG tmp0[P256_LIMBS];
241 const BN_ULONG *in_x = a->X;
242 const BN_ULONG *in_y = a->Y;
243 const BN_ULONG *in_z = a->Z;
245 BN_ULONG *res_x = r->X;
246 BN_ULONG *res_y = r->Y;
247 BN_ULONG *res_z = r->Z;
249 ecp_nistz256_mul_by_2(S, in_y);
251 ecp_nistz256_sqr_mont(Zsqr, in_z);
253 ecp_nistz256_sqr_mont(S, S);
255 ecp_nistz256_mul_mont(res_z, in_z, in_y);
256 ecp_nistz256_mul_by_2(res_z, res_z);
258 ecp_nistz256_add(M, in_x, Zsqr);
259 ecp_nistz256_sub(Zsqr, in_x, Zsqr);
261 ecp_nistz256_sqr_mont(res_y, S);
262 ecp_nistz256_div_by_2(res_y, res_y);
264 ecp_nistz256_mul_mont(M, M, Zsqr);
265 ecp_nistz256_mul_by_3(M, M);
267 ecp_nistz256_mul_mont(S, S, in_x);
268 ecp_nistz256_mul_by_2(tmp0, S);
270 ecp_nistz256_sqr_mont(res_x, M);
272 ecp_nistz256_sub(res_x, res_x, tmp0);
273 ecp_nistz256_sub(S, S, res_x);
275 ecp_nistz256_mul_mont(S, S, M);
276 ecp_nistz256_sub(res_y, S, res_y);
279 /* Point addition: r = a+b */
280 static void ecp_nistz256_point_add(P256_POINT *r,
281 const P256_POINT *a, const P256_POINT *b)
283 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
284 BN_ULONG U1[P256_LIMBS], S1[P256_LIMBS];
285 BN_ULONG Z1sqr[P256_LIMBS];
286 BN_ULONG Z2sqr[P256_LIMBS];
287 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
288 BN_ULONG Hsqr[P256_LIMBS];
289 BN_ULONG Rsqr[P256_LIMBS];
290 BN_ULONG Hcub[P256_LIMBS];
292 BN_ULONG res_x[P256_LIMBS];
293 BN_ULONG res_y[P256_LIMBS];
294 BN_ULONG res_z[P256_LIMBS];
296 BN_ULONG in1infty, in2infty;
298 const BN_ULONG *in1_x = a->X;
299 const BN_ULONG *in1_y = a->Y;
300 const BN_ULONG *in1_z = a->Z;
302 const BN_ULONG *in2_x = b->X;
303 const BN_ULONG *in2_y = b->Y;
304 const BN_ULONG *in2_z = b->Z;
306 /* We encode infinity as (0,0), which is not on the curve,
308 in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
309 in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]);
311 in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
312 in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]);
314 in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
315 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
317 in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
318 in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
320 in1infty = is_zero(in1infty);
321 in2infty = is_zero(in2infty);
323 ecp_nistz256_sqr_mont(Z2sqr, in2_z); /* Z2^2 */
324 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
326 ecp_nistz256_mul_mont(S1, Z2sqr, in2_z); /* S1 = Z2^3 */
327 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
329 ecp_nistz256_mul_mont(S1, S1, in1_y); /* S1 = Y1*Z2^3 */
330 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
331 ecp_nistz256_sub(R, S2, S1); /* R = S2 - S1 */
333 ecp_nistz256_mul_mont(U1, in1_x, Z2sqr); /* U1 = X1*Z2^2 */
334 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
335 ecp_nistz256_sub(H, U2, U1); /* H = U2 - U1 */
338 * This should not happen during sign/ecdh, so no constant time violation
340 if (is_equal(U1, U2) && !in1infty && !in2infty) {
341 if (is_equal(S1, S2)) {
342 ecp_nistz256_point_double(r, a);
345 memset(r, 0, sizeof(*r));
350 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
351 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
352 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
353 ecp_nistz256_mul_mont(res_z, res_z, in2_z); /* Z3 = H*Z1*Z2 */
354 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
356 ecp_nistz256_mul_mont(U2, U1, Hsqr); /* U1*H^2 */
357 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
359 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
360 ecp_nistz256_sub(res_x, res_x, Hcub);
362 ecp_nistz256_sub(res_y, U2, res_x);
364 ecp_nistz256_mul_mont(S2, S1, Hcub);
365 ecp_nistz256_mul_mont(res_y, R, res_y);
366 ecp_nistz256_sub(res_y, res_y, S2);
368 copy_conditional(res_x, in2_x, in1infty);
369 copy_conditional(res_y, in2_y, in1infty);
370 copy_conditional(res_z, in2_z, in1infty);
372 copy_conditional(res_x, in1_x, in2infty);
373 copy_conditional(res_y, in1_y, in2infty);
374 copy_conditional(res_z, in1_z, in2infty);
376 memcpy(r->X, res_x, sizeof(res_x));
377 memcpy(r->Y, res_y, sizeof(res_y));
378 memcpy(r->Z, res_z, sizeof(res_z));
381 /* Point addition when b is known to be affine: r = a+b */
382 static void ecp_nistz256_point_add_affine(P256_POINT *r,
384 const P256_POINT_AFFINE *b)
386 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
387 BN_ULONG Z1sqr[P256_LIMBS];
388 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
389 BN_ULONG Hsqr[P256_LIMBS];
390 BN_ULONG Rsqr[P256_LIMBS];
391 BN_ULONG Hcub[P256_LIMBS];
393 BN_ULONG res_x[P256_LIMBS];
394 BN_ULONG res_y[P256_LIMBS];
395 BN_ULONG res_z[P256_LIMBS];
397 BN_ULONG in1infty, in2infty;
399 const BN_ULONG *in1_x = a->X;
400 const BN_ULONG *in1_y = a->Y;
401 const BN_ULONG *in1_z = a->Z;
403 const BN_ULONG *in2_x = b->X;
404 const BN_ULONG *in2_y = b->Y;
407 * In affine representation we encode infty as (0,0), which is not on the
410 in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
411 in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]);
413 in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
414 in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]);
416 in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
417 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
419 in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
420 in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
422 in1infty = is_zero(in1infty);
423 in2infty = is_zero(in2infty);
425 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
427 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
428 ecp_nistz256_sub(H, U2, in1_x); /* H = U2 - U1 */
430 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
432 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
434 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
435 ecp_nistz256_sub(R, S2, in1_y); /* R = S2 - S1 */
437 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
438 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
439 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
441 ecp_nistz256_mul_mont(U2, in1_x, Hsqr); /* U1*H^2 */
442 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
444 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
445 ecp_nistz256_sub(res_x, res_x, Hcub);
446 ecp_nistz256_sub(H, U2, res_x);
448 ecp_nistz256_mul_mont(S2, in1_y, Hcub);
449 ecp_nistz256_mul_mont(H, H, R);
450 ecp_nistz256_sub(res_y, H, S2);
452 copy_conditional(res_x, in2_x, in1infty);
453 copy_conditional(res_x, in1_x, in2infty);
455 copy_conditional(res_y, in2_y, in1infty);
456 copy_conditional(res_y, in1_y, in2infty);
458 copy_conditional(res_z, ONE, in1infty);
459 copy_conditional(res_z, in1_z, in2infty);
461 memcpy(r->X, res_x, sizeof(res_x));
462 memcpy(r->Y, res_y, sizeof(res_y));
463 memcpy(r->Z, res_z, sizeof(res_z));
467 /* r = in^-1 mod p */
468 static void ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS],
469 const BN_ULONG in[P256_LIMBS])
472 * The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff
473 * ffffffff ffffffff We use FLT and used poly-2 as exponent
475 BN_ULONG p2[P256_LIMBS];
476 BN_ULONG p4[P256_LIMBS];
477 BN_ULONG p8[P256_LIMBS];
478 BN_ULONG p16[P256_LIMBS];
479 BN_ULONG p32[P256_LIMBS];
480 BN_ULONG res[P256_LIMBS];
483 ecp_nistz256_sqr_mont(res, in);
484 ecp_nistz256_mul_mont(p2, res, in); /* 3*p */
486 ecp_nistz256_sqr_mont(res, p2);
487 ecp_nistz256_sqr_mont(res, res);
488 ecp_nistz256_mul_mont(p4, res, p2); /* f*p */
490 ecp_nistz256_sqr_mont(res, p4);
491 ecp_nistz256_sqr_mont(res, res);
492 ecp_nistz256_sqr_mont(res, res);
493 ecp_nistz256_sqr_mont(res, res);
494 ecp_nistz256_mul_mont(p8, res, p4); /* ff*p */
496 ecp_nistz256_sqr_mont(res, p8);
497 for (i = 0; i < 7; i++)
498 ecp_nistz256_sqr_mont(res, res);
499 ecp_nistz256_mul_mont(p16, res, p8); /* ffff*p */
501 ecp_nistz256_sqr_mont(res, p16);
502 for (i = 0; i < 15; i++)
503 ecp_nistz256_sqr_mont(res, res);
504 ecp_nistz256_mul_mont(p32, res, p16); /* ffffffff*p */
506 ecp_nistz256_sqr_mont(res, p32);
507 for (i = 0; i < 31; i++)
508 ecp_nistz256_sqr_mont(res, res);
509 ecp_nistz256_mul_mont(res, res, in);
511 for (i = 0; i < 32 * 4; i++)
512 ecp_nistz256_sqr_mont(res, res);
513 ecp_nistz256_mul_mont(res, res, p32);
515 for (i = 0; i < 32; i++)
516 ecp_nistz256_sqr_mont(res, res);
517 ecp_nistz256_mul_mont(res, res, p32);
519 for (i = 0; i < 16; i++)
520 ecp_nistz256_sqr_mont(res, res);
521 ecp_nistz256_mul_mont(res, res, p16);
523 for (i = 0; i < 8; i++)
524 ecp_nistz256_sqr_mont(res, res);
525 ecp_nistz256_mul_mont(res, res, p8);
527 ecp_nistz256_sqr_mont(res, res);
528 ecp_nistz256_sqr_mont(res, res);
529 ecp_nistz256_sqr_mont(res, res);
530 ecp_nistz256_sqr_mont(res, res);
531 ecp_nistz256_mul_mont(res, res, p4);
533 ecp_nistz256_sqr_mont(res, res);
534 ecp_nistz256_sqr_mont(res, res);
535 ecp_nistz256_mul_mont(res, res, p2);
537 ecp_nistz256_sqr_mont(res, res);
538 ecp_nistz256_sqr_mont(res, res);
539 ecp_nistz256_mul_mont(res, res, in);
541 memcpy(r, res, sizeof(res));
545 * ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
546 * returns one if it fits. Otherwise it returns zero.
548 static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],
551 if (in->top > P256_LIMBS)
554 memset(out, 0, sizeof(BN_ULONG) * P256_LIMBS);
555 memcpy(out, in->d, sizeof(BN_ULONG) * in->top);
559 /* r = sum(scalar[i]*point[i]) */
560 static void ecp_nistz256_windowed_mul(const EC_GROUP *group,
562 const BIGNUM **scalar,
563 const EC_POINT **point,
564 int num, BN_CTX *ctx)
568 unsigned char (*p_str)[33] = NULL;
569 const unsigned int window_size = 5;
570 const unsigned int mask = (1 << (window_size + 1)) - 1;
572 BN_ULONG tmp[P256_LIMBS];
573 ALIGN32 P256_POINT h;
574 const BIGNUM **scalars = NULL;
575 P256_POINT (*table)[16] = NULL;
576 void *table_storage = NULL;
579 OPENSSL_malloc(num * 16 * sizeof(P256_POINT) + 64)) == NULL
581 OPENSSL_malloc(num * 33 * sizeof(unsigned char))) == NULL
582 || (scalars = OPENSSL_malloc(num * sizeof(BIGNUM *))) == NULL) {
583 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_MALLOC_FAILURE);
586 table = (void *)ALIGNPTR(table_storage, 64);
589 for (i = 0; i < num; i++) {
590 P256_POINT *row = table[i];
592 if ((BN_num_bits(scalar[i]) > 256) || BN_is_negative(scalar[i])) {
595 if ((mod = BN_CTX_get(ctx)) == NULL)
597 if (!BN_nnmod(mod, scalar[i], &group->order, ctx)) {
598 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_BN_LIB);
603 scalars[i] = scalar[i];
605 for (j = 0; j < scalars[i]->top * BN_BYTES; j += BN_BYTES) {
606 BN_ULONG d = scalars[i]->d[j / BN_BYTES];
608 p_str[i][j + 0] = d & 0xff;
609 p_str[i][j + 1] = (d >> 8) & 0xff;
610 p_str[i][j + 2] = (d >> 16) & 0xff;
611 p_str[i][j + 3] = (d >>= 24) & 0xff;
614 p_str[i][j + 4] = d & 0xff;
615 p_str[i][j + 5] = (d >> 8) & 0xff;
616 p_str[i][j + 6] = (d >> 16) & 0xff;
617 p_str[i][j + 7] = (d >> 24) & 0xff;
623 /* table[0] is implicitly (0,0,0) (the point at infinity),
624 * therefore it is not stored. All other values are actually
625 * stored with an offset of -1 in table.
628 if (!ecp_nistz256_bignum_to_field_elem(row[1 - 1].X, &point[i]->X)
629 || !ecp_nistz256_bignum_to_field_elem(row[1 - 1].Y, &point[i]->Y)
630 || !ecp_nistz256_bignum_to_field_elem(row[1 - 1].Z, &point[i]->Z)) {
631 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, EC_R_COORDINATES_OUT_OF_RANGE);
635 ecp_nistz256_point_double(&row[ 2 - 1], &row[ 1 - 1]);
636 ecp_nistz256_point_add (&row[ 3 - 1], &row[ 2 - 1], &row[1 - 1]);
637 ecp_nistz256_point_double(&row[ 4 - 1], &row[ 2 - 1]);
638 ecp_nistz256_point_double(&row[ 6 - 1], &row[ 3 - 1]);
639 ecp_nistz256_point_double(&row[ 8 - 1], &row[ 4 - 1]);
640 ecp_nistz256_point_double(&row[12 - 1], &row[ 6 - 1]);
641 ecp_nistz256_point_add (&row[ 5 - 1], &row[ 4 - 1], &row[1 - 1]);
642 ecp_nistz256_point_add (&row[ 7 - 1], &row[ 6 - 1], &row[1 - 1]);
643 ecp_nistz256_point_add (&row[ 9 - 1], &row[ 8 - 1], &row[1 - 1]);
644 ecp_nistz256_point_add (&row[13 - 1], &row[12 - 1], &row[1 - 1]);
645 ecp_nistz256_point_double(&row[14 - 1], &row[ 7 - 1]);
646 ecp_nistz256_point_double(&row[10 - 1], &row[ 5 - 1]);
647 ecp_nistz256_point_add (&row[15 - 1], &row[14 - 1], &row[1 - 1]);
648 ecp_nistz256_point_add (&row[11 - 1], &row[10 - 1], &row[1 - 1]);
649 ecp_nistz256_point_add (&row[16 - 1], &row[15 - 1], &row[1 - 1]);
654 wvalue = p_str[0][(index - 1) / 8];
655 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
657 ecp_nistz256_select_w5(r, table[0], _booth_recode_w5(wvalue) >> 1);
660 for (i = (index == 255 ? 1 : 0); i < num; i++) {
661 unsigned int off = (index - 1) / 8;
663 wvalue = p_str[i][off] | p_str[i][off + 1] << 8;
664 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
666 wvalue = _booth_recode_w5(wvalue);
668 ecp_nistz256_select_w5(&h, table[i], wvalue >> 1);
670 ecp_nistz256_neg(tmp, h.Y);
671 copy_conditional(h.Y, tmp, (wvalue & 1));
673 ecp_nistz256_point_add(r, r, &h);
676 index -= window_size;
678 ecp_nistz256_point_double(r, r);
679 ecp_nistz256_point_double(r, r);
680 ecp_nistz256_point_double(r, r);
681 ecp_nistz256_point_double(r, r);
682 ecp_nistz256_point_double(r, r);
686 for (i = 0; i < num; i++) {
687 wvalue = p_str[i][0];
688 wvalue = (wvalue << 1) & mask;
690 wvalue = _booth_recode_w5(wvalue);
692 ecp_nistz256_select_w5(&h, table[i], wvalue >> 1);
694 ecp_nistz256_neg(tmp, h.Y);
695 copy_conditional(h.Y, tmp, wvalue & 1);
697 ecp_nistz256_point_add(r, r, &h);
702 OPENSSL_free(table_storage);
706 OPENSSL_free(scalars);
709 /* Coordinates of G, for which we have precomputed tables */
710 const static BN_ULONG def_xG[P256_LIMBS] = {
711 TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
712 TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
715 const static BN_ULONG def_yG[P256_LIMBS] = {
716 TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
717 TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
721 * ecp_nistz256_is_affine_G returns one if |generator| is the standard, P-256
724 static int ecp_nistz256_is_affine_G(const EC_POINT *generator)
726 return (generator->X.top == P256_LIMBS) &&
727 (generator->Y.top == P256_LIMBS) &&
728 (generator->Z.top == (P256_LIMBS - P256_LIMBS / 8)) &&
729 is_equal(generator->X.d, def_xG) &&
730 is_equal(generator->Y.d, def_yG) && is_one(generator->Z.d);
733 static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx)
736 * We precompute a table for a Booth encoded exponent (wNAF) based
737 * computation. Each table holds 64 values for safe access, with an
738 * implicit value of infinity at index zero. We use window of size 7, and
739 * therefore require ceil(256/7) = 37 tables.
742 EC_POINT *P = NULL, *T = NULL;
743 const EC_POINT *generator;
744 EC_PRE_COMP *pre_comp;
745 int i, j, k, ret = 0;
748 PRECOMP256_ROW *preComputedTable = NULL;
749 unsigned char *precomp_storage = NULL;
751 /* if there is an old EC_PRE_COMP object, throw it away */
752 EC_EX_DATA_free_data(&group->extra_data, ecp_nistz256_pre_comp_dup,
753 ecp_nistz256_pre_comp_free,
754 ecp_nistz256_pre_comp_clear_free);
756 generator = EC_GROUP_get0_generator(group);
757 if (generator == NULL) {
758 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNDEFINED_GENERATOR);
762 if (ecp_nistz256_is_affine_G(generator)) {
764 * No need to calculate tables for the standard generator because we
765 * have them statically.
770 if ((pre_comp = ecp_nistz256_pre_comp_new(group)) == NULL)
780 order = BN_CTX_get(ctx);
785 if (!EC_GROUP_get_order(group, order, ctx))
788 if (BN_is_zero(order)) {
789 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNKNOWN_ORDER);
795 if ((precomp_storage =
796 OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE) + 64)) == NULL) {
797 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, ERR_R_MALLOC_FAILURE);
800 preComputedTable = (void *)ALIGNPTR(precomp_storage, 64);
803 P = EC_POINT_new(group);
804 T = EC_POINT_new(group);
807 * The zero entry is implicitly infinity, and we skip it, storing other
808 * values with -1 offset.
810 EC_POINT_copy(T, generator);
812 for (k = 0; k < 64; k++) {
814 for (j = 0; j < 37; j++) {
816 * It would be faster to use
817 * ec_GFp_simple_points_make_affine and make multiple
818 * points affine at the same time.
820 ec_GFp_simple_make_affine(group, P, ctx);
821 ecp_nistz256_bignum_to_field_elem(preComputedTable[j]
823 ecp_nistz256_bignum_to_field_elem(preComputedTable[j]
825 for (i = 0; i < 7; i++)
826 ec_GFp_simple_dbl(group, P, P, ctx);
828 ec_GFp_simple_add(group, T, T, generator, ctx);
831 pre_comp->group = group;
833 pre_comp->precomp = preComputedTable;
834 pre_comp->precomp_storage = precomp_storage;
836 precomp_storage = NULL;
838 if (!EC_EX_DATA_set_data(&group->extra_data, pre_comp,
839 ecp_nistz256_pre_comp_dup,
840 ecp_nistz256_pre_comp_free,
841 ecp_nistz256_pre_comp_clear_free)) {
853 ecp_nistz256_pre_comp_free(pre_comp);
855 OPENSSL_free(precomp_storage);
864 * Note that by default ECP_NISTZ256_AVX2 is undefined. While it's great
865 * code processing 4 points in parallel, corresponding serial operation
866 * is several times slower, because it uses 29x29=58-bit multiplication
867 * as opposite to 64x64=128-bit in integer-only scalar case. As result
868 * it doesn't provide *significant* performance improvement. Note that
869 * just defining ECP_NISTZ256_AVX2 is not sufficient to make it work,
870 * you'd need to compile even asm/ecp_nistz256-avx.pl module.
872 #if defined(ECP_NISTZ256_AVX2)
873 # if !(defined(__x86_64) || defined(__x86_64__)) || \
874 defined(_M_AMD64) || defined(_MX64)) || \
875 !(defined(__GNUC__) || defined(_MSC_VER)) /* this is for ALIGN32 */
876 # undef ECP_NISTZ256_AVX2
878 /* Constant time access, loading four values, from four consecutive tables */
879 void ecp_nistz256_avx2_select_w7(P256_POINT_AFFINE * val,
880 const P256_POINT_AFFINE * in_t, int index);
881 void ecp_nistz256_avx2_multi_select_w7(void *result, const void *in, int index0,
882 int index1, int index2, int index3);
883 void ecp_nistz256_avx2_transpose_convert(void *RESULTx4, const void *in);
884 void ecp_nistz256_avx2_convert_transpose_back(void *result, const void *Ax4);
885 void ecp_nistz256_avx2_point_add_affine_x4(void *RESULTx4, const void *Ax4,
887 void ecp_nistz256_avx2_point_add_affines_x4(void *RESULTx4, const void *Ax4,
889 void ecp_nistz256_avx2_to_mont(void *RESULTx4, const void *Ax4);
890 void ecp_nistz256_avx2_from_mont(void *RESULTx4, const void *Ax4);
891 void ecp_nistz256_avx2_set1(void *RESULTx4);
892 int ecp_nistz_avx2_eligible(void);
894 static void booth_recode_w7(unsigned char *sign,
895 unsigned char *digit, unsigned char in)
899 s = ~((in >> 7) - 1);
900 d = (1 << 8) - in - 1;
901 d = (d & s) | (in & ~s);
902 d = (d >> 1) + (d & 1);
909 * ecp_nistz256_avx2_mul_g performs multiplication by G, using only the
910 * precomputed table. It does 4 affine point additions in parallel,
911 * significantly speeding up point multiplication for a fixed value.
913 static void ecp_nistz256_avx2_mul_g(P256_POINT *r,
914 unsigned char p_str[33],
915 const P256_POINT_AFFINE(*preComputedTable)[64])
917 const unsigned int window_size = 7;
918 const unsigned int mask = (1 << (window_size + 1)) - 1;
920 /* Using 4 windows at a time */
921 unsigned char sign0, digit0;
922 unsigned char sign1, digit1;
923 unsigned char sign2, digit2;
924 unsigned char sign3, digit3;
925 unsigned int index = 0;
926 BN_ULONG tmp[P256_LIMBS];
929 ALIGN32 BN_ULONG aX4[4 * 9 * 3] = { 0 };
930 ALIGN32 BN_ULONG bX4[4 * 9 * 2] = { 0 };
931 ALIGN32 P256_POINT_AFFINE point_arr[P256_LIMBS];
932 ALIGN32 P256_POINT res_point_arr[P256_LIMBS];
934 /* Initial four windows */
935 wvalue = *((u16 *) & p_str[0]);
936 wvalue = (wvalue << 1) & mask;
937 index += window_size;
938 booth_recode_w7(&sign0, &digit0, wvalue);
939 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
940 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
941 index += window_size;
942 booth_recode_w7(&sign1, &digit1, wvalue);
943 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
944 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
945 index += window_size;
946 booth_recode_w7(&sign2, &digit2, wvalue);
947 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
948 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
949 index += window_size;
950 booth_recode_w7(&sign3, &digit3, wvalue);
952 ecp_nistz256_avx2_multi_select_w7(point_arr, preComputedTable[0],
953 digit0, digit1, digit2, digit3);
955 ecp_nistz256_neg(tmp, point_arr[0].Y);
956 copy_conditional(point_arr[0].Y, tmp, sign0);
957 ecp_nistz256_neg(tmp, point_arr[1].Y);
958 copy_conditional(point_arr[1].Y, tmp, sign1);
959 ecp_nistz256_neg(tmp, point_arr[2].Y);
960 copy_conditional(point_arr[2].Y, tmp, sign2);
961 ecp_nistz256_neg(tmp, point_arr[3].Y);
962 copy_conditional(point_arr[3].Y, tmp, sign3);
964 ecp_nistz256_avx2_transpose_convert(aX4, point_arr);
965 ecp_nistz256_avx2_to_mont(aX4, aX4);
966 ecp_nistz256_avx2_to_mont(&aX4[4 * 9], &aX4[4 * 9]);
967 ecp_nistz256_avx2_set1(&aX4[4 * 9 * 2]);
969 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
970 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
971 index += window_size;
972 booth_recode_w7(&sign0, &digit0, wvalue);
973 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
974 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
975 index += window_size;
976 booth_recode_w7(&sign1, &digit1, wvalue);
977 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
978 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
979 index += window_size;
980 booth_recode_w7(&sign2, &digit2, wvalue);
981 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
982 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
983 index += window_size;
984 booth_recode_w7(&sign3, &digit3, wvalue);
986 ecp_nistz256_avx2_multi_select_w7(point_arr, preComputedTable[4 * 1],
987 digit0, digit1, digit2, digit3);
989 ecp_nistz256_neg(tmp, point_arr[0].Y);
990 copy_conditional(point_arr[0].Y, tmp, sign0);
991 ecp_nistz256_neg(tmp, point_arr[1].Y);
992 copy_conditional(point_arr[1].Y, tmp, sign1);
993 ecp_nistz256_neg(tmp, point_arr[2].Y);
994 copy_conditional(point_arr[2].Y, tmp, sign2);
995 ecp_nistz256_neg(tmp, point_arr[3].Y);
996 copy_conditional(point_arr[3].Y, tmp, sign3);
998 ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
999 ecp_nistz256_avx2_to_mont(bX4, bX4);
1000 ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
1001 /* Optimized when both inputs are affine */
1002 ecp_nistz256_avx2_point_add_affines_x4(aX4, aX4, bX4);
1004 for (i = 2; i < 9; i++) {
1005 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1006 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1007 index += window_size;
1008 booth_recode_w7(&sign0, &digit0, wvalue);
1009 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1010 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1011 index += window_size;
1012 booth_recode_w7(&sign1, &digit1, wvalue);
1013 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1014 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1015 index += window_size;
1016 booth_recode_w7(&sign2, &digit2, wvalue);
1017 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1018 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1019 index += window_size;
1020 booth_recode_w7(&sign3, &digit3, wvalue);
1022 ecp_nistz256_avx2_multi_select_w7(point_arr,
1023 preComputedTable[4 * i],
1024 digit0, digit1, digit2, digit3);
1026 ecp_nistz256_neg(tmp, point_arr[0].Y);
1027 copy_conditional(point_arr[0].Y, tmp, sign0);
1028 ecp_nistz256_neg(tmp, point_arr[1].Y);
1029 copy_conditional(point_arr[1].Y, tmp, sign1);
1030 ecp_nistz256_neg(tmp, point_arr[2].Y);
1031 copy_conditional(point_arr[2].Y, tmp, sign2);
1032 ecp_nistz256_neg(tmp, point_arr[3].Y);
1033 copy_conditional(point_arr[3].Y, tmp, sign3);
1035 ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
1036 ecp_nistz256_avx2_to_mont(bX4, bX4);
1037 ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
1039 ecp_nistz256_avx2_point_add_affine_x4(aX4, aX4, bX4);
1042 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 0], &aX4[4 * 9 * 0]);
1043 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 1], &aX4[4 * 9 * 1]);
1044 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 2], &aX4[4 * 9 * 2]);
1046 ecp_nistz256_avx2_convert_transpose_back(res_point_arr, aX4);
1047 /* Last window is performed serially */
1048 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1049 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1050 booth_recode_w7(&sign0, &digit0, wvalue);
1051 ecp_nistz256_avx2_select_w7((P256_POINT_AFFINE *) r,
1052 preComputedTable[36], digit0);
1053 ecp_nistz256_neg(tmp, r->Y);
1054 copy_conditional(r->Y, tmp, sign0);
1055 memcpy(r->Z, ONE, sizeof(ONE));
1056 /* Sum the four windows */
1057 ecp_nistz256_point_add(r, r, &res_point_arr[0]);
1058 ecp_nistz256_point_add(r, r, &res_point_arr[1]);
1059 ecp_nistz256_point_add(r, r, &res_point_arr[2]);
1060 ecp_nistz256_point_add(r, r, &res_point_arr[3]);
1065 static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group,
1066 const P256_POINT_AFFINE *in,
1070 BN_ULONG d_x[P256_LIMBS], d_y[P256_LIMBS];
1073 memcpy(d_x, in->X, sizeof(d_x));
1075 x.dmax = x.top = P256_LIMBS;
1077 x.flags = BN_FLG_STATIC_DATA;
1079 memcpy(d_y, in->Y, sizeof(d_y));
1081 y.dmax = y.top = P256_LIMBS;
1083 y.flags = BN_FLG_STATIC_DATA;
1085 ret = EC_POINT_set_affine_coordinates_GFp(group, out, &x, &y, ctx);
1090 /* r = scalar*G + sum(scalars[i]*points[i]) */
1091 static int ecp_nistz256_points_mul(const EC_GROUP *group,
1093 const BIGNUM *scalar,
1095 const EC_POINT *points[],
1096 const BIGNUM *scalars[], BN_CTX *ctx)
1098 int i = 0, ret = 0, no_precomp_for_generator = 0, p_is_infinity = 0;
1100 unsigned char p_str[33] = { 0 };
1101 const PRECOMP256_ROW *preComputedTable = NULL;
1102 const EC_PRE_COMP *pre_comp = NULL;
1103 const EC_POINT *generator = NULL;
1104 unsigned int index = 0;
1105 const unsigned int window_size = 7;
1106 const unsigned int mask = (1 << (window_size + 1)) - 1;
1107 unsigned int wvalue;
1110 P256_POINT_AFFINE a;
1114 if (group->meth != r->meth) {
1115 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
1118 if ((scalar == NULL) && (num == 0))
1119 return EC_POINT_set_to_infinity(group, r);
1121 for (j = 0; j < num; j++) {
1122 if (group->meth != points[j]->meth) {
1123 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
1128 /* Need 256 bits for space for all coordinates. */
1129 bn_wexpand(&r->X, P256_LIMBS);
1130 bn_wexpand(&r->Y, P256_LIMBS);
1131 bn_wexpand(&r->Z, P256_LIMBS);
1132 r->X.top = P256_LIMBS;
1133 r->Y.top = P256_LIMBS;
1134 r->Z.top = P256_LIMBS;
1137 generator = EC_GROUP_get0_generator(group);
1138 if (generator == NULL) {
1139 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_UNDEFINED_GENERATOR);
1143 /* look if we can use precomputed multiples of generator */
1145 EC_EX_DATA_get_data(group->extra_data, ecp_nistz256_pre_comp_dup,
1146 ecp_nistz256_pre_comp_free,
1147 ecp_nistz256_pre_comp_clear_free);
1151 * If there is a precomputed table for the generator, check that
1152 * it was generated with the same generator.
1154 EC_POINT *pre_comp_generator = EC_POINT_new(group);
1155 if (pre_comp_generator == NULL)
1158 if (!ecp_nistz256_set_from_affine
1159 (pre_comp_generator, group, pre_comp->precomp[0], ctx))
1162 if (0 == EC_POINT_cmp(group, generator, pre_comp_generator, ctx))
1163 preComputedTable = (const PRECOMP256_ROW *)pre_comp->precomp;
1165 EC_POINT_free(pre_comp_generator);
1168 if (preComputedTable == NULL && ecp_nistz256_is_affine_G(generator)) {
1170 * If there is no precomputed data, but the generator
1171 * is the default, a hardcoded table of precomputed
1172 * data is used. This is because applications, such as
1173 * Apache, do not use EC_KEY_precompute_mult.
1175 preComputedTable = (const PRECOMP256_ROW *)ecp_nistz256_precomputed;
1178 if (preComputedTable) {
1179 if ((BN_num_bits(scalar) > 256)
1180 || BN_is_negative(scalar)) {
1181 if ((tmp_scalar = BN_CTX_get(ctx)) == NULL)
1184 if (!BN_nnmod(tmp_scalar, scalar, &group->order, ctx)) {
1185 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_BN_LIB);
1188 scalar = tmp_scalar;
1191 for (i = 0; i < scalar->top * BN_BYTES; i += BN_BYTES) {
1192 BN_ULONG d = scalar->d[i / BN_BYTES];
1194 p_str[i + 0] = d & 0xff;
1195 p_str[i + 1] = (d >> 8) & 0xff;
1196 p_str[i + 2] = (d >> 16) & 0xff;
1197 p_str[i + 3] = (d >>= 24) & 0xff;
1198 if (BN_BYTES == 8) {
1200 p_str[i + 4] = d & 0xff;
1201 p_str[i + 5] = (d >> 8) & 0xff;
1202 p_str[i + 6] = (d >> 16) & 0xff;
1203 p_str[i + 7] = (d >> 24) & 0xff;
1210 #if defined(ECP_NISTZ256_AVX2)
1211 if (ecp_nistz_avx2_eligible()) {
1212 ecp_nistz256_avx2_mul_g(&p.p, p_str, preComputedTable);
1217 wvalue = (p_str[0] << 1) & mask;
1218 index += window_size;
1220 wvalue = _booth_recode_w7(wvalue);
1222 ecp_nistz256_select_w7(&p.a, preComputedTable[0], wvalue >> 1);
1224 ecp_nistz256_neg(p.p.Z, p.p.Y);
1225 copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
1227 memcpy(p.p.Z, ONE, sizeof(ONE));
1229 for (i = 1; i < 37; i++) {
1230 unsigned int off = (index - 1) / 8;
1231 wvalue = p_str[off] | p_str[off + 1] << 8;
1232 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1233 index += window_size;
1235 wvalue = _booth_recode_w7(wvalue);
1237 ecp_nistz256_select_w7(&t.a,
1238 preComputedTable[i], wvalue >> 1);
1240 ecp_nistz256_neg(t.p.Z, t.a.Y);
1241 copy_conditional(t.a.Y, t.p.Z, wvalue & 1);
1243 ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a);
1248 no_precomp_for_generator = 1;
1253 if (no_precomp_for_generator) {
1255 * Without a precomputed table for the generator, it has to be
1256 * handled like a normal point.
1258 const BIGNUM **new_scalars;
1259 const EC_POINT **new_points;
1261 new_scalars = OPENSSL_malloc((num + 1) * sizeof(BIGNUM *));
1263 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1267 new_points = OPENSSL_malloc((num + 1) * sizeof(EC_POINT *));
1269 OPENSSL_free(new_scalars);
1270 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1274 memcpy(new_scalars, scalars, num * sizeof(BIGNUM *));
1275 new_scalars[num] = scalar;
1276 memcpy(new_points, points, num * sizeof(EC_POINT *));
1277 new_points[num] = generator;
1279 scalars = new_scalars;
1280 points = new_points;
1285 P256_POINT *out = &t.p;
1289 ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx);
1292 ecp_nistz256_point_add(&p.p, &p.p, out);
1295 if (no_precomp_for_generator) {
1296 OPENSSL_free(points);
1297 OPENSSL_free(scalars);
1300 memcpy(r->X.d, p.p.X, sizeof(p.p.X));
1301 memcpy(r->Y.d, p.p.Y, sizeof(p.p.Y));
1302 memcpy(r->Z.d, p.p.Z, sizeof(p.p.Z));
1303 bn_correct_top(&r->X);
1304 bn_correct_top(&r->Y);
1305 bn_correct_top(&r->Z);
1313 static int ecp_nistz256_get_affine(const EC_GROUP *group,
1314 const EC_POINT *point,
1315 BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
1317 BN_ULONG z_inv2[P256_LIMBS];
1318 BN_ULONG z_inv3[P256_LIMBS];
1319 BN_ULONG x_aff[P256_LIMBS];
1320 BN_ULONG y_aff[P256_LIMBS];
1321 BN_ULONG point_x[P256_LIMBS], point_y[P256_LIMBS], point_z[P256_LIMBS];
1323 if (EC_POINT_is_at_infinity(group, point)) {
1324 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_POINT_AT_INFINITY);
1328 if (!ecp_nistz256_bignum_to_field_elem(point_x, &point->X) ||
1329 !ecp_nistz256_bignum_to_field_elem(point_y, &point->Y) ||
1330 !ecp_nistz256_bignum_to_field_elem(point_z, &point->Z)) {
1331 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_COORDINATES_OUT_OF_RANGE);
1335 ecp_nistz256_mod_inverse(z_inv3, point_z);
1336 ecp_nistz256_sqr_mont(z_inv2, z_inv3);
1337 ecp_nistz256_mul_mont(x_aff, z_inv2, point_x);
1340 bn_wexpand(x, P256_LIMBS);
1341 x->top = P256_LIMBS;
1342 ecp_nistz256_from_mont(x->d, x_aff);
1347 ecp_nistz256_mul_mont(z_inv3, z_inv3, z_inv2);
1348 ecp_nistz256_mul_mont(y_aff, z_inv3, point_y);
1349 bn_wexpand(y, P256_LIMBS);
1350 y->top = P256_LIMBS;
1351 ecp_nistz256_from_mont(y->d, y_aff);
1358 static EC_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group)
1360 EC_PRE_COMP *ret = NULL;
1365 ret = (EC_PRE_COMP *)OPENSSL_malloc(sizeof(EC_PRE_COMP));
1368 ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1373 ret->w = 6; /* default */
1374 ret->precomp = NULL;
1375 ret->precomp_storage = NULL;
1376 ret->references = 1;
1380 static void *ecp_nistz256_pre_comp_dup(void *src_)
1382 EC_PRE_COMP *src = src_;
1384 /* no need to actually copy, these objects never change! */
1385 CRYPTO_add(&src->references, 1, CRYPTO_LOCK_EC_PRE_COMP);
1390 static void ecp_nistz256_pre_comp_free(void *pre_)
1393 EC_PRE_COMP *pre = pre_;
1398 i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1402 if (pre->precomp_storage)
1403 OPENSSL_free(pre->precomp_storage);
1408 static void ecp_nistz256_pre_comp_clear_free(void *pre_)
1411 EC_PRE_COMP *pre = pre_;
1416 i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1420 if (pre->precomp_storage) {
1421 OPENSSL_cleanse(pre->precomp,
1422 32 * sizeof(unsigned char) * (1 << pre->w) * 2 * 37);
1423 OPENSSL_free(pre->precomp_storage);
1425 OPENSSL_cleanse(pre, sizeof *pre);
1429 static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP *group)
1431 /* There is a hard-coded table for the default generator. */
1432 const EC_POINT *generator = EC_GROUP_get0_generator(group);
1433 if (generator != NULL && ecp_nistz256_is_affine_G(generator)) {
1434 /* There is a hard-coded table for the default generator. */
1438 return EC_EX_DATA_get_data(group->extra_data, ecp_nistz256_pre_comp_dup,
1439 ecp_nistz256_pre_comp_free,
1440 ecp_nistz256_pre_comp_clear_free) != NULL;
1443 const EC_METHOD *EC_GFp_nistz256_method(void)
1445 static const EC_METHOD ret = {
1446 EC_FLAGS_DEFAULT_OCT,
1447 NID_X9_62_prime_field,
1448 ec_GFp_mont_group_init,
1449 ec_GFp_mont_group_finish,
1450 ec_GFp_mont_group_clear_finish,
1451 ec_GFp_mont_group_copy,
1452 ec_GFp_mont_group_set_curve,
1453 ec_GFp_simple_group_get_curve,
1454 ec_GFp_simple_group_get_degree,
1455 ec_GFp_simple_group_check_discriminant,
1456 ec_GFp_simple_point_init,
1457 ec_GFp_simple_point_finish,
1458 ec_GFp_simple_point_clear_finish,
1459 ec_GFp_simple_point_copy,
1460 ec_GFp_simple_point_set_to_infinity,
1461 ec_GFp_simple_set_Jprojective_coordinates_GFp,
1462 ec_GFp_simple_get_Jprojective_coordinates_GFp,
1463 ec_GFp_simple_point_set_affine_coordinates,
1464 ecp_nistz256_get_affine,
1468 ec_GFp_simple_invert,
1469 ec_GFp_simple_is_at_infinity,
1470 ec_GFp_simple_is_on_curve,
1472 ec_GFp_simple_make_affine,
1473 ec_GFp_simple_points_make_affine,
1474 ecp_nistz256_points_mul, /* mul */
1475 ecp_nistz256_mult_precompute, /* precompute_mult */
1476 ecp_nistz256_window_have_precompute_mult, /* have_precompute_mult */
1477 ec_GFp_mont_field_mul,
1478 ec_GFp_mont_field_sqr,
1480 ec_GFp_mont_field_encode,
1481 ec_GFp_mont_field_decode,
1482 ec_GFp_mont_field_set_to_one