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 static int ecp_nistz256_set_words(BIGNUM *a, BN_ULONG words[P256_LIMBS])
227 if (bn_wexpand(a, P256_LIMBS) == NULL) {
228 ECerr(EC_F_ECP_NISTZ256_SET_WORDS, ERR_R_MALLOC_FAILURE);
231 memcpy(a->d, words, sizeof(BN_ULONG) * P256_LIMBS);
237 #ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
238 void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a);
239 void ecp_nistz256_point_add(P256_POINT *r,
240 const P256_POINT *a, const P256_POINT *b);
241 void ecp_nistz256_point_add_affine(P256_POINT *r,
243 const P256_POINT_AFFINE *b);
245 /* Point double: r = 2*a */
246 static void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a)
248 BN_ULONG S[P256_LIMBS];
249 BN_ULONG M[P256_LIMBS];
250 BN_ULONG Zsqr[P256_LIMBS];
251 BN_ULONG tmp0[P256_LIMBS];
253 const BN_ULONG *in_x = a->X;
254 const BN_ULONG *in_y = a->Y;
255 const BN_ULONG *in_z = a->Z;
257 BN_ULONG *res_x = r->X;
258 BN_ULONG *res_y = r->Y;
259 BN_ULONG *res_z = r->Z;
261 ecp_nistz256_mul_by_2(S, in_y);
263 ecp_nistz256_sqr_mont(Zsqr, in_z);
265 ecp_nistz256_sqr_mont(S, S);
267 ecp_nistz256_mul_mont(res_z, in_z, in_y);
268 ecp_nistz256_mul_by_2(res_z, res_z);
270 ecp_nistz256_add(M, in_x, Zsqr);
271 ecp_nistz256_sub(Zsqr, in_x, Zsqr);
273 ecp_nistz256_sqr_mont(res_y, S);
274 ecp_nistz256_div_by_2(res_y, res_y);
276 ecp_nistz256_mul_mont(M, M, Zsqr);
277 ecp_nistz256_mul_by_3(M, M);
279 ecp_nistz256_mul_mont(S, S, in_x);
280 ecp_nistz256_mul_by_2(tmp0, S);
282 ecp_nistz256_sqr_mont(res_x, M);
284 ecp_nistz256_sub(res_x, res_x, tmp0);
285 ecp_nistz256_sub(S, S, res_x);
287 ecp_nistz256_mul_mont(S, S, M);
288 ecp_nistz256_sub(res_y, S, res_y);
291 /* Point addition: r = a+b */
292 static void ecp_nistz256_point_add(P256_POINT *r,
293 const P256_POINT *a, const P256_POINT *b)
295 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
296 BN_ULONG U1[P256_LIMBS], S1[P256_LIMBS];
297 BN_ULONG Z1sqr[P256_LIMBS];
298 BN_ULONG Z2sqr[P256_LIMBS];
299 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
300 BN_ULONG Hsqr[P256_LIMBS];
301 BN_ULONG Rsqr[P256_LIMBS];
302 BN_ULONG Hcub[P256_LIMBS];
304 BN_ULONG res_x[P256_LIMBS];
305 BN_ULONG res_y[P256_LIMBS];
306 BN_ULONG res_z[P256_LIMBS];
308 BN_ULONG in1infty, in2infty;
310 const BN_ULONG *in1_x = a->X;
311 const BN_ULONG *in1_y = a->Y;
312 const BN_ULONG *in1_z = a->Z;
314 const BN_ULONG *in2_x = b->X;
315 const BN_ULONG *in2_y = b->Y;
316 const BN_ULONG *in2_z = b->Z;
318 /* We encode infinity as (0,0), which is not on the curve,
320 in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
321 in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]);
323 in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
324 in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]);
326 in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
327 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
329 in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
330 in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
332 in1infty = is_zero(in1infty);
333 in2infty = is_zero(in2infty);
335 ecp_nistz256_sqr_mont(Z2sqr, in2_z); /* Z2^2 */
336 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
338 ecp_nistz256_mul_mont(S1, Z2sqr, in2_z); /* S1 = Z2^3 */
339 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
341 ecp_nistz256_mul_mont(S1, S1, in1_y); /* S1 = Y1*Z2^3 */
342 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
343 ecp_nistz256_sub(R, S2, S1); /* R = S2 - S1 */
345 ecp_nistz256_mul_mont(U1, in1_x, Z2sqr); /* U1 = X1*Z2^2 */
346 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
347 ecp_nistz256_sub(H, U2, U1); /* H = U2 - U1 */
350 * This should not happen during sign/ecdh, so no constant time violation
352 if (is_equal(U1, U2) && !in1infty && !in2infty) {
353 if (is_equal(S1, S2)) {
354 ecp_nistz256_point_double(r, a);
357 memset(r, 0, sizeof(*r));
362 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
363 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
364 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
365 ecp_nistz256_mul_mont(res_z, res_z, in2_z); /* Z3 = H*Z1*Z2 */
366 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
368 ecp_nistz256_mul_mont(U2, U1, Hsqr); /* U1*H^2 */
369 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
371 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
372 ecp_nistz256_sub(res_x, res_x, Hcub);
374 ecp_nistz256_sub(res_y, U2, res_x);
376 ecp_nistz256_mul_mont(S2, S1, Hcub);
377 ecp_nistz256_mul_mont(res_y, R, res_y);
378 ecp_nistz256_sub(res_y, res_y, S2);
380 copy_conditional(res_x, in2_x, in1infty);
381 copy_conditional(res_y, in2_y, in1infty);
382 copy_conditional(res_z, in2_z, in1infty);
384 copy_conditional(res_x, in1_x, in2infty);
385 copy_conditional(res_y, in1_y, in2infty);
386 copy_conditional(res_z, in1_z, in2infty);
388 memcpy(r->X, res_x, sizeof(res_x));
389 memcpy(r->Y, res_y, sizeof(res_y));
390 memcpy(r->Z, res_z, sizeof(res_z));
393 /* Point addition when b is known to be affine: r = a+b */
394 static void ecp_nistz256_point_add_affine(P256_POINT *r,
396 const P256_POINT_AFFINE *b)
398 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
399 BN_ULONG Z1sqr[P256_LIMBS];
400 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
401 BN_ULONG Hsqr[P256_LIMBS];
402 BN_ULONG Rsqr[P256_LIMBS];
403 BN_ULONG Hcub[P256_LIMBS];
405 BN_ULONG res_x[P256_LIMBS];
406 BN_ULONG res_y[P256_LIMBS];
407 BN_ULONG res_z[P256_LIMBS];
409 BN_ULONG in1infty, in2infty;
411 const BN_ULONG *in1_x = a->X;
412 const BN_ULONG *in1_y = a->Y;
413 const BN_ULONG *in1_z = a->Z;
415 const BN_ULONG *in2_x = b->X;
416 const BN_ULONG *in2_y = b->Y;
419 * In affine representation we encode infty as (0,0), which is not on the
422 in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
423 in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]);
425 in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
426 in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]);
428 in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
429 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
431 in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
432 in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
434 in1infty = is_zero(in1infty);
435 in2infty = is_zero(in2infty);
437 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
439 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
440 ecp_nistz256_sub(H, U2, in1_x); /* H = U2 - U1 */
442 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
444 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
446 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
447 ecp_nistz256_sub(R, S2, in1_y); /* R = S2 - S1 */
449 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
450 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
451 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
453 ecp_nistz256_mul_mont(U2, in1_x, Hsqr); /* U1*H^2 */
454 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
456 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
457 ecp_nistz256_sub(res_x, res_x, Hcub);
458 ecp_nistz256_sub(H, U2, res_x);
460 ecp_nistz256_mul_mont(S2, in1_y, Hcub);
461 ecp_nistz256_mul_mont(H, H, R);
462 ecp_nistz256_sub(res_y, H, S2);
464 copy_conditional(res_x, in2_x, in1infty);
465 copy_conditional(res_x, in1_x, in2infty);
467 copy_conditional(res_y, in2_y, in1infty);
468 copy_conditional(res_y, in1_y, in2infty);
470 copy_conditional(res_z, ONE, in1infty);
471 copy_conditional(res_z, in1_z, in2infty);
473 memcpy(r->X, res_x, sizeof(res_x));
474 memcpy(r->Y, res_y, sizeof(res_y));
475 memcpy(r->Z, res_z, sizeof(res_z));
479 /* r = in^-1 mod p */
480 static void ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS],
481 const BN_ULONG in[P256_LIMBS])
484 * The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff
485 * ffffffff ffffffff We use FLT and used poly-2 as exponent
487 BN_ULONG p2[P256_LIMBS];
488 BN_ULONG p4[P256_LIMBS];
489 BN_ULONG p8[P256_LIMBS];
490 BN_ULONG p16[P256_LIMBS];
491 BN_ULONG p32[P256_LIMBS];
492 BN_ULONG res[P256_LIMBS];
495 ecp_nistz256_sqr_mont(res, in);
496 ecp_nistz256_mul_mont(p2, res, in); /* 3*p */
498 ecp_nistz256_sqr_mont(res, p2);
499 ecp_nistz256_sqr_mont(res, res);
500 ecp_nistz256_mul_mont(p4, res, p2); /* f*p */
502 ecp_nistz256_sqr_mont(res, p4);
503 ecp_nistz256_sqr_mont(res, res);
504 ecp_nistz256_sqr_mont(res, res);
505 ecp_nistz256_sqr_mont(res, res);
506 ecp_nistz256_mul_mont(p8, res, p4); /* ff*p */
508 ecp_nistz256_sqr_mont(res, p8);
509 for (i = 0; i < 7; i++)
510 ecp_nistz256_sqr_mont(res, res);
511 ecp_nistz256_mul_mont(p16, res, p8); /* ffff*p */
513 ecp_nistz256_sqr_mont(res, p16);
514 for (i = 0; i < 15; i++)
515 ecp_nistz256_sqr_mont(res, res);
516 ecp_nistz256_mul_mont(p32, res, p16); /* ffffffff*p */
518 ecp_nistz256_sqr_mont(res, p32);
519 for (i = 0; i < 31; i++)
520 ecp_nistz256_sqr_mont(res, res);
521 ecp_nistz256_mul_mont(res, res, in);
523 for (i = 0; i < 32 * 4; i++)
524 ecp_nistz256_sqr_mont(res, res);
525 ecp_nistz256_mul_mont(res, res, p32);
527 for (i = 0; i < 32; i++)
528 ecp_nistz256_sqr_mont(res, res);
529 ecp_nistz256_mul_mont(res, res, p32);
531 for (i = 0; i < 16; i++)
532 ecp_nistz256_sqr_mont(res, res);
533 ecp_nistz256_mul_mont(res, res, p16);
535 for (i = 0; i < 8; i++)
536 ecp_nistz256_sqr_mont(res, res);
537 ecp_nistz256_mul_mont(res, res, p8);
539 ecp_nistz256_sqr_mont(res, res);
540 ecp_nistz256_sqr_mont(res, res);
541 ecp_nistz256_sqr_mont(res, res);
542 ecp_nistz256_sqr_mont(res, res);
543 ecp_nistz256_mul_mont(res, res, p4);
545 ecp_nistz256_sqr_mont(res, res);
546 ecp_nistz256_sqr_mont(res, res);
547 ecp_nistz256_mul_mont(res, res, p2);
549 ecp_nistz256_sqr_mont(res, res);
550 ecp_nistz256_sqr_mont(res, res);
551 ecp_nistz256_mul_mont(res, res, in);
553 memcpy(r, res, sizeof(res));
557 * ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
558 * returns one if it fits. Otherwise it returns zero.
560 static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],
563 if (in->top > P256_LIMBS)
566 memset(out, 0, sizeof(BN_ULONG) * P256_LIMBS);
567 memcpy(out, in->d, sizeof(BN_ULONG) * in->top);
571 /* r = sum(scalar[i]*point[i]) */
572 static int ecp_nistz256_windowed_mul(const EC_GROUP *group,
574 const BIGNUM **scalar,
575 const EC_POINT **point,
576 int num, BN_CTX *ctx)
581 unsigned char (*p_str)[33] = NULL;
582 const unsigned int window_size = 5;
583 const unsigned int mask = (1 << (window_size + 1)) - 1;
585 BN_ULONG tmp[P256_LIMBS];
586 ALIGN32 P256_POINT h;
587 const BIGNUM **scalars = NULL;
588 P256_POINT (*table)[16] = NULL;
589 void *table_storage = NULL;
592 OPENSSL_malloc(num * 16 * sizeof(P256_POINT) + 64)) == NULL
594 OPENSSL_malloc(num * 33 * sizeof(unsigned char))) == NULL
595 || (scalars = OPENSSL_malloc(num * sizeof(BIGNUM *))) == NULL) {
596 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_MALLOC_FAILURE);
599 table = (void *)ALIGNPTR(table_storage, 64);
602 for (i = 0; i < num; i++) {
603 P256_POINT *row = table[i];
605 /* This is an unusual input, we don't guarantee constant-timeness. */
606 if ((BN_num_bits(scalar[i]) > 256) || BN_is_negative(scalar[i])) {
609 if ((mod = BN_CTX_get(ctx)) == NULL)
611 if (!BN_nnmod(mod, scalar[i], &group->order, ctx)) {
612 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_BN_LIB);
617 scalars[i] = scalar[i];
619 for (j = 0; j < scalars[i]->top * BN_BYTES; j += BN_BYTES) {
620 BN_ULONG d = scalars[i]->d[j / BN_BYTES];
622 p_str[i][j + 0] = d & 0xff;
623 p_str[i][j + 1] = (d >> 8) & 0xff;
624 p_str[i][j + 2] = (d >> 16) & 0xff;
625 p_str[i][j + 3] = (d >>= 24) & 0xff;
628 p_str[i][j + 4] = d & 0xff;
629 p_str[i][j + 5] = (d >> 8) & 0xff;
630 p_str[i][j + 6] = (d >> 16) & 0xff;
631 p_str[i][j + 7] = (d >> 24) & 0xff;
637 /* table[0] is implicitly (0,0,0) (the point at infinity),
638 * therefore it is not stored. All other values are actually
639 * stored with an offset of -1 in table.
642 if (!ecp_nistz256_bignum_to_field_elem(row[1 - 1].X, &point[i]->X)
643 || !ecp_nistz256_bignum_to_field_elem(row[1 - 1].Y, &point[i]->Y)
644 || !ecp_nistz256_bignum_to_field_elem(row[1 - 1].Z, &point[i]->Z)) {
645 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, EC_R_COORDINATES_OUT_OF_RANGE);
649 ecp_nistz256_point_double(&row[ 2 - 1], &row[ 1 - 1]);
650 ecp_nistz256_point_add (&row[ 3 - 1], &row[ 2 - 1], &row[1 - 1]);
651 ecp_nistz256_point_double(&row[ 4 - 1], &row[ 2 - 1]);
652 ecp_nistz256_point_double(&row[ 6 - 1], &row[ 3 - 1]);
653 ecp_nistz256_point_double(&row[ 8 - 1], &row[ 4 - 1]);
654 ecp_nistz256_point_double(&row[12 - 1], &row[ 6 - 1]);
655 ecp_nistz256_point_add (&row[ 5 - 1], &row[ 4 - 1], &row[1 - 1]);
656 ecp_nistz256_point_add (&row[ 7 - 1], &row[ 6 - 1], &row[1 - 1]);
657 ecp_nistz256_point_add (&row[ 9 - 1], &row[ 8 - 1], &row[1 - 1]);
658 ecp_nistz256_point_add (&row[13 - 1], &row[12 - 1], &row[1 - 1]);
659 ecp_nistz256_point_double(&row[14 - 1], &row[ 7 - 1]);
660 ecp_nistz256_point_double(&row[10 - 1], &row[ 5 - 1]);
661 ecp_nistz256_point_add (&row[15 - 1], &row[14 - 1], &row[1 - 1]);
662 ecp_nistz256_point_add (&row[11 - 1], &row[10 - 1], &row[1 - 1]);
663 ecp_nistz256_point_add (&row[16 - 1], &row[15 - 1], &row[1 - 1]);
668 wvalue = p_str[0][(index - 1) / 8];
669 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
671 ecp_nistz256_select_w5(r, table[0], _booth_recode_w5(wvalue) >> 1);
674 for (i = (index == 255 ? 1 : 0); i < num; i++) {
675 unsigned int off = (index - 1) / 8;
677 wvalue = p_str[i][off] | p_str[i][off + 1] << 8;
678 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
680 wvalue = _booth_recode_w5(wvalue);
682 ecp_nistz256_select_w5(&h, table[i], wvalue >> 1);
684 ecp_nistz256_neg(tmp, h.Y);
685 copy_conditional(h.Y, tmp, (wvalue & 1));
687 ecp_nistz256_point_add(r, r, &h);
690 index -= window_size;
692 ecp_nistz256_point_double(r, r);
693 ecp_nistz256_point_double(r, r);
694 ecp_nistz256_point_double(r, r);
695 ecp_nistz256_point_double(r, r);
696 ecp_nistz256_point_double(r, r);
700 for (i = 0; i < num; i++) {
701 wvalue = p_str[i][0];
702 wvalue = (wvalue << 1) & mask;
704 wvalue = _booth_recode_w5(wvalue);
706 ecp_nistz256_select_w5(&h, table[i], wvalue >> 1);
708 ecp_nistz256_neg(tmp, h.Y);
709 copy_conditional(h.Y, tmp, wvalue & 1);
711 ecp_nistz256_point_add(r, r, &h);
717 OPENSSL_free(table_storage);
721 OPENSSL_free(scalars);
725 /* Coordinates of G, for which we have precomputed tables */
726 const static BN_ULONG def_xG[P256_LIMBS] = {
727 TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
728 TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
731 const static BN_ULONG def_yG[P256_LIMBS] = {
732 TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
733 TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
737 * ecp_nistz256_is_affine_G returns one if |generator| is the standard, P-256
740 static int ecp_nistz256_is_affine_G(const EC_POINT *generator)
742 return (generator->X.top == P256_LIMBS) &&
743 (generator->Y.top == P256_LIMBS) &&
744 (generator->Z.top == (P256_LIMBS - P256_LIMBS / 8)) &&
745 is_equal(generator->X.d, def_xG) &&
746 is_equal(generator->Y.d, def_yG) && is_one(generator->Z.d);
749 static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx)
752 * We precompute a table for a Booth encoded exponent (wNAF) based
753 * computation. Each table holds 64 values for safe access, with an
754 * implicit value of infinity at index zero. We use window of size 7, and
755 * therefore require ceil(256/7) = 37 tables.
758 EC_POINT *P = NULL, *T = NULL;
759 const EC_POINT *generator;
760 EC_PRE_COMP *pre_comp;
761 BN_CTX *new_ctx = NULL;
762 int i, j, k, ret = 0;
765 PRECOMP256_ROW *preComputedTable = NULL;
766 unsigned char *precomp_storage = NULL;
768 /* if there is an old EC_PRE_COMP object, throw it away */
769 EC_EX_DATA_free_data(&group->extra_data, ecp_nistz256_pre_comp_dup,
770 ecp_nistz256_pre_comp_free,
771 ecp_nistz256_pre_comp_clear_free);
773 generator = EC_GROUP_get0_generator(group);
774 if (generator == NULL) {
775 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNDEFINED_GENERATOR);
779 if (ecp_nistz256_is_affine_G(generator)) {
781 * No need to calculate tables for the standard generator because we
782 * have them statically.
787 if ((pre_comp = ecp_nistz256_pre_comp_new(group)) == NULL)
791 ctx = new_ctx = BN_CTX_new();
797 order = BN_CTX_get(ctx);
802 if (!EC_GROUP_get_order(group, order, ctx))
805 if (BN_is_zero(order)) {
806 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNKNOWN_ORDER);
812 if ((precomp_storage =
813 OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE) + 64)) == NULL) {
814 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, ERR_R_MALLOC_FAILURE);
817 preComputedTable = (void *)ALIGNPTR(precomp_storage, 64);
820 P = EC_POINT_new(group);
821 T = EC_POINT_new(group);
822 if (P == NULL || T == NULL)
826 * The zero entry is implicitly infinity, and we skip it, storing other
827 * values with -1 offset.
829 if (!EC_POINT_copy(T, generator))
832 for (k = 0; k < 64; k++) {
833 if (!EC_POINT_copy(P, T))
835 for (j = 0; j < 37; j++) {
837 * It would be faster to use
838 * ec_GFp_simple_points_make_affine and make multiple
839 * points affine at the same time.
841 ec_GFp_simple_make_affine(group, P, ctx);
842 ecp_nistz256_bignum_to_field_elem(preComputedTable[j]
844 ecp_nistz256_bignum_to_field_elem(preComputedTable[j]
846 for (i = 0; i < 7; i++)
847 ec_GFp_simple_dbl(group, P, P, ctx);
849 ec_GFp_simple_add(group, T, T, generator, ctx);
852 pre_comp->group = group;
854 pre_comp->precomp = preComputedTable;
855 pre_comp->precomp_storage = precomp_storage;
857 precomp_storage = NULL;
859 if (!EC_EX_DATA_set_data(&group->extra_data, pre_comp,
860 ecp_nistz256_pre_comp_dup,
861 ecp_nistz256_pre_comp_free,
862 ecp_nistz256_pre_comp_clear_free)) {
873 BN_CTX_free(new_ctx);
876 ecp_nistz256_pre_comp_free(pre_comp);
878 OPENSSL_free(precomp_storage);
887 * Note that by default ECP_NISTZ256_AVX2 is undefined. While it's great
888 * code processing 4 points in parallel, corresponding serial operation
889 * is several times slower, because it uses 29x29=58-bit multiplication
890 * as opposite to 64x64=128-bit in integer-only scalar case. As result
891 * it doesn't provide *significant* performance improvement. Note that
892 * just defining ECP_NISTZ256_AVX2 is not sufficient to make it work,
893 * you'd need to compile even asm/ecp_nistz256-avx.pl module.
895 #if defined(ECP_NISTZ256_AVX2)
896 # if !(defined(__x86_64) || defined(__x86_64__)) || \
897 defined(_M_AMD64) || defined(_MX64)) || \
898 !(defined(__GNUC__) || defined(_MSC_VER)) /* this is for ALIGN32 */
899 # undef ECP_NISTZ256_AVX2
901 /* Constant time access, loading four values, from four consecutive tables */
902 void ecp_nistz256_avx2_select_w7(P256_POINT_AFFINE * val,
903 const P256_POINT_AFFINE * in_t, int index);
904 void ecp_nistz256_avx2_multi_select_w7(void *result, const void *in, int index0,
905 int index1, int index2, int index3);
906 void ecp_nistz256_avx2_transpose_convert(void *RESULTx4, const void *in);
907 void ecp_nistz256_avx2_convert_transpose_back(void *result, const void *Ax4);
908 void ecp_nistz256_avx2_point_add_affine_x4(void *RESULTx4, const void *Ax4,
910 void ecp_nistz256_avx2_point_add_affines_x4(void *RESULTx4, const void *Ax4,
912 void ecp_nistz256_avx2_to_mont(void *RESULTx4, const void *Ax4);
913 void ecp_nistz256_avx2_from_mont(void *RESULTx4, const void *Ax4);
914 void ecp_nistz256_avx2_set1(void *RESULTx4);
915 int ecp_nistz_avx2_eligible(void);
917 static void booth_recode_w7(unsigned char *sign,
918 unsigned char *digit, unsigned char in)
922 s = ~((in >> 7) - 1);
923 d = (1 << 8) - in - 1;
924 d = (d & s) | (in & ~s);
925 d = (d >> 1) + (d & 1);
932 * ecp_nistz256_avx2_mul_g performs multiplication by G, using only the
933 * precomputed table. It does 4 affine point additions in parallel,
934 * significantly speeding up point multiplication for a fixed value.
936 static void ecp_nistz256_avx2_mul_g(P256_POINT *r,
937 unsigned char p_str[33],
938 const P256_POINT_AFFINE(*preComputedTable)[64])
940 const unsigned int window_size = 7;
941 const unsigned int mask = (1 << (window_size + 1)) - 1;
943 /* Using 4 windows at a time */
944 unsigned char sign0, digit0;
945 unsigned char sign1, digit1;
946 unsigned char sign2, digit2;
947 unsigned char sign3, digit3;
948 unsigned int index = 0;
949 BN_ULONG tmp[P256_LIMBS];
952 ALIGN32 BN_ULONG aX4[4 * 9 * 3] = { 0 };
953 ALIGN32 BN_ULONG bX4[4 * 9 * 2] = { 0 };
954 ALIGN32 P256_POINT_AFFINE point_arr[P256_LIMBS];
955 ALIGN32 P256_POINT res_point_arr[P256_LIMBS];
957 /* Initial four windows */
958 wvalue = *((u16 *) & p_str[0]);
959 wvalue = (wvalue << 1) & mask;
960 index += window_size;
961 booth_recode_w7(&sign0, &digit0, wvalue);
962 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
963 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
964 index += window_size;
965 booth_recode_w7(&sign1, &digit1, wvalue);
966 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
967 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
968 index += window_size;
969 booth_recode_w7(&sign2, &digit2, wvalue);
970 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
971 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
972 index += window_size;
973 booth_recode_w7(&sign3, &digit3, wvalue);
975 ecp_nistz256_avx2_multi_select_w7(point_arr, preComputedTable[0],
976 digit0, digit1, digit2, digit3);
978 ecp_nistz256_neg(tmp, point_arr[0].Y);
979 copy_conditional(point_arr[0].Y, tmp, sign0);
980 ecp_nistz256_neg(tmp, point_arr[1].Y);
981 copy_conditional(point_arr[1].Y, tmp, sign1);
982 ecp_nistz256_neg(tmp, point_arr[2].Y);
983 copy_conditional(point_arr[2].Y, tmp, sign2);
984 ecp_nistz256_neg(tmp, point_arr[3].Y);
985 copy_conditional(point_arr[3].Y, tmp, sign3);
987 ecp_nistz256_avx2_transpose_convert(aX4, point_arr);
988 ecp_nistz256_avx2_to_mont(aX4, aX4);
989 ecp_nistz256_avx2_to_mont(&aX4[4 * 9], &aX4[4 * 9]);
990 ecp_nistz256_avx2_set1(&aX4[4 * 9 * 2]);
992 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
993 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
994 index += window_size;
995 booth_recode_w7(&sign0, &digit0, wvalue);
996 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
997 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
998 index += window_size;
999 booth_recode_w7(&sign1, &digit1, wvalue);
1000 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1001 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1002 index += window_size;
1003 booth_recode_w7(&sign2, &digit2, wvalue);
1004 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1005 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1006 index += window_size;
1007 booth_recode_w7(&sign3, &digit3, wvalue);
1009 ecp_nistz256_avx2_multi_select_w7(point_arr, preComputedTable[4 * 1],
1010 digit0, digit1, digit2, digit3);
1012 ecp_nistz256_neg(tmp, point_arr[0].Y);
1013 copy_conditional(point_arr[0].Y, tmp, sign0);
1014 ecp_nistz256_neg(tmp, point_arr[1].Y);
1015 copy_conditional(point_arr[1].Y, tmp, sign1);
1016 ecp_nistz256_neg(tmp, point_arr[2].Y);
1017 copy_conditional(point_arr[2].Y, tmp, sign2);
1018 ecp_nistz256_neg(tmp, point_arr[3].Y);
1019 copy_conditional(point_arr[3].Y, tmp, sign3);
1021 ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
1022 ecp_nistz256_avx2_to_mont(bX4, bX4);
1023 ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
1024 /* Optimized when both inputs are affine */
1025 ecp_nistz256_avx2_point_add_affines_x4(aX4, aX4, bX4);
1027 for (i = 2; i < 9; i++) {
1028 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1029 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1030 index += window_size;
1031 booth_recode_w7(&sign0, &digit0, wvalue);
1032 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1033 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1034 index += window_size;
1035 booth_recode_w7(&sign1, &digit1, wvalue);
1036 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1037 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1038 index += window_size;
1039 booth_recode_w7(&sign2, &digit2, wvalue);
1040 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1041 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1042 index += window_size;
1043 booth_recode_w7(&sign3, &digit3, wvalue);
1045 ecp_nistz256_avx2_multi_select_w7(point_arr,
1046 preComputedTable[4 * i],
1047 digit0, digit1, digit2, digit3);
1049 ecp_nistz256_neg(tmp, point_arr[0].Y);
1050 copy_conditional(point_arr[0].Y, tmp, sign0);
1051 ecp_nistz256_neg(tmp, point_arr[1].Y);
1052 copy_conditional(point_arr[1].Y, tmp, sign1);
1053 ecp_nistz256_neg(tmp, point_arr[2].Y);
1054 copy_conditional(point_arr[2].Y, tmp, sign2);
1055 ecp_nistz256_neg(tmp, point_arr[3].Y);
1056 copy_conditional(point_arr[3].Y, tmp, sign3);
1058 ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
1059 ecp_nistz256_avx2_to_mont(bX4, bX4);
1060 ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
1062 ecp_nistz256_avx2_point_add_affine_x4(aX4, aX4, bX4);
1065 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 0], &aX4[4 * 9 * 0]);
1066 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 1], &aX4[4 * 9 * 1]);
1067 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 2], &aX4[4 * 9 * 2]);
1069 ecp_nistz256_avx2_convert_transpose_back(res_point_arr, aX4);
1070 /* Last window is performed serially */
1071 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1072 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1073 booth_recode_w7(&sign0, &digit0, wvalue);
1074 ecp_nistz256_avx2_select_w7((P256_POINT_AFFINE *) r,
1075 preComputedTable[36], digit0);
1076 ecp_nistz256_neg(tmp, r->Y);
1077 copy_conditional(r->Y, tmp, sign0);
1078 memcpy(r->Z, ONE, sizeof(ONE));
1079 /* Sum the four windows */
1080 ecp_nistz256_point_add(r, r, &res_point_arr[0]);
1081 ecp_nistz256_point_add(r, r, &res_point_arr[1]);
1082 ecp_nistz256_point_add(r, r, &res_point_arr[2]);
1083 ecp_nistz256_point_add(r, r, &res_point_arr[3]);
1088 static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group,
1089 const P256_POINT_AFFINE *in,
1093 BN_ULONG d_x[P256_LIMBS], d_y[P256_LIMBS];
1096 memcpy(d_x, in->X, sizeof(d_x));
1098 x.dmax = x.top = P256_LIMBS;
1100 x.flags = BN_FLG_STATIC_DATA;
1102 memcpy(d_y, in->Y, sizeof(d_y));
1104 y.dmax = y.top = P256_LIMBS;
1106 y.flags = BN_FLG_STATIC_DATA;
1108 ret = EC_POINT_set_affine_coordinates_GFp(group, out, &x, &y, ctx);
1113 /* r = scalar*G + sum(scalars[i]*points[i]) */
1114 static int ecp_nistz256_points_mul(const EC_GROUP *group,
1116 const BIGNUM *scalar,
1118 const EC_POINT *points[],
1119 const BIGNUM *scalars[], BN_CTX *ctx)
1121 int i = 0, ret = 0, no_precomp_for_generator = 0, p_is_infinity = 0;
1123 unsigned char p_str[33] = { 0 };
1124 const PRECOMP256_ROW *preComputedTable = NULL;
1125 const EC_PRE_COMP *pre_comp = NULL;
1126 const EC_POINT *generator = NULL;
1127 unsigned int index = 0;
1128 BN_CTX *new_ctx = NULL;
1129 const BIGNUM **new_scalars = NULL;
1130 const EC_POINT **new_points = NULL;
1131 const unsigned int window_size = 7;
1132 const unsigned int mask = (1 << (window_size + 1)) - 1;
1133 unsigned int wvalue;
1136 P256_POINT_AFFINE a;
1140 if (group->meth != r->meth) {
1141 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
1145 if ((scalar == NULL) && (num == 0))
1146 return EC_POINT_set_to_infinity(group, r);
1148 for (j = 0; j < num; j++) {
1149 if (group->meth != points[j]->meth) {
1150 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
1156 ctx = new_ctx = BN_CTX_new();
1164 generator = EC_GROUP_get0_generator(group);
1165 if (generator == NULL) {
1166 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_UNDEFINED_GENERATOR);
1170 /* look if we can use precomputed multiples of generator */
1172 EC_EX_DATA_get_data(group->extra_data, ecp_nistz256_pre_comp_dup,
1173 ecp_nistz256_pre_comp_free,
1174 ecp_nistz256_pre_comp_clear_free);
1178 * If there is a precomputed table for the generator, check that
1179 * it was generated with the same generator.
1181 EC_POINT *pre_comp_generator = EC_POINT_new(group);
1182 if (pre_comp_generator == NULL)
1185 if (!ecp_nistz256_set_from_affine
1186 (pre_comp_generator, group, pre_comp->precomp[0], ctx)) {
1187 EC_POINT_free(pre_comp_generator);
1191 if (0 == EC_POINT_cmp(group, generator, pre_comp_generator, ctx))
1192 preComputedTable = (const PRECOMP256_ROW *)pre_comp->precomp;
1194 EC_POINT_free(pre_comp_generator);
1197 if (preComputedTable == NULL && ecp_nistz256_is_affine_G(generator)) {
1199 * If there is no precomputed data, but the generator
1200 * is the default, a hardcoded table of precomputed
1201 * data is used. This is because applications, such as
1202 * Apache, do not use EC_KEY_precompute_mult.
1204 preComputedTable = (const PRECOMP256_ROW *)ecp_nistz256_precomputed;
1207 if (preComputedTable) {
1208 if ((BN_num_bits(scalar) > 256)
1209 || BN_is_negative(scalar)) {
1210 if ((tmp_scalar = BN_CTX_get(ctx)) == NULL)
1213 if (!BN_nnmod(tmp_scalar, scalar, &group->order, ctx)) {
1214 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_BN_LIB);
1217 scalar = tmp_scalar;
1220 for (i = 0; i < scalar->top * BN_BYTES; i += BN_BYTES) {
1221 BN_ULONG d = scalar->d[i / BN_BYTES];
1223 p_str[i + 0] = d & 0xff;
1224 p_str[i + 1] = (d >> 8) & 0xff;
1225 p_str[i + 2] = (d >> 16) & 0xff;
1226 p_str[i + 3] = (d >>= 24) & 0xff;
1227 if (BN_BYTES == 8) {
1229 p_str[i + 4] = d & 0xff;
1230 p_str[i + 5] = (d >> 8) & 0xff;
1231 p_str[i + 6] = (d >> 16) & 0xff;
1232 p_str[i + 7] = (d >> 24) & 0xff;
1239 #if defined(ECP_NISTZ256_AVX2)
1240 if (ecp_nistz_avx2_eligible()) {
1241 ecp_nistz256_avx2_mul_g(&p.p, p_str, preComputedTable);
1246 wvalue = (p_str[0] << 1) & mask;
1247 index += window_size;
1249 wvalue = _booth_recode_w7(wvalue);
1251 ecp_nistz256_select_w7(&p.a, preComputedTable[0], wvalue >> 1);
1253 ecp_nistz256_neg(p.p.Z, p.p.Y);
1254 copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
1256 memcpy(p.p.Z, ONE, sizeof(ONE));
1258 for (i = 1; i < 37; i++) {
1259 unsigned int off = (index - 1) / 8;
1260 wvalue = p_str[off] | p_str[off + 1] << 8;
1261 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1262 index += window_size;
1264 wvalue = _booth_recode_w7(wvalue);
1266 ecp_nistz256_select_w7(&t.a,
1267 preComputedTable[i], wvalue >> 1);
1269 ecp_nistz256_neg(t.p.Z, t.a.Y);
1270 copy_conditional(t.a.Y, t.p.Z, wvalue & 1);
1272 ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a);
1277 no_precomp_for_generator = 1;
1282 if (no_precomp_for_generator) {
1284 * Without a precomputed table for the generator, it has to be
1285 * handled like a normal point.
1287 new_scalars = OPENSSL_malloc((num + 1) * sizeof(BIGNUM *));
1289 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1293 new_points = OPENSSL_malloc((num + 1) * sizeof(EC_POINT *));
1295 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1299 memcpy(new_scalars, scalars, num * sizeof(BIGNUM *));
1300 new_scalars[num] = scalar;
1301 memcpy(new_points, points, num * sizeof(EC_POINT *));
1302 new_points[num] = generator;
1304 scalars = new_scalars;
1305 points = new_points;
1310 P256_POINT *out = &t.p;
1314 if (!ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx))
1318 ecp_nistz256_point_add(&p.p, &p.p, out);
1321 /* Not constant-time, but we're only operating on the public output. */
1322 if (!ecp_nistz256_set_words(&r->X, p.p.X) ||
1323 !ecp_nistz256_set_words(&r->Y, p.p.Y) ||
1324 !ecp_nistz256_set_words(&r->Z, p.p.Z)) {
1327 r->Z_is_one = is_one(p.p.Z) & 1;
1334 BN_CTX_free(new_ctx);
1336 OPENSSL_free(new_points);
1338 OPENSSL_free(new_scalars);
1342 static int ecp_nistz256_get_affine(const EC_GROUP *group,
1343 const EC_POINT *point,
1344 BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
1346 BN_ULONG z_inv2[P256_LIMBS];
1347 BN_ULONG z_inv3[P256_LIMBS];
1348 BN_ULONG x_aff[P256_LIMBS];
1349 BN_ULONG y_aff[P256_LIMBS];
1350 BN_ULONG point_x[P256_LIMBS], point_y[P256_LIMBS], point_z[P256_LIMBS];
1351 BN_ULONG x_ret[P256_LIMBS], y_ret[P256_LIMBS];
1353 if (EC_POINT_is_at_infinity(group, point)) {
1354 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_POINT_AT_INFINITY);
1358 if (!ecp_nistz256_bignum_to_field_elem(point_x, &point->X) ||
1359 !ecp_nistz256_bignum_to_field_elem(point_y, &point->Y) ||
1360 !ecp_nistz256_bignum_to_field_elem(point_z, &point->Z)) {
1361 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_COORDINATES_OUT_OF_RANGE);
1365 ecp_nistz256_mod_inverse(z_inv3, point_z);
1366 ecp_nistz256_sqr_mont(z_inv2, z_inv3);
1367 ecp_nistz256_mul_mont(x_aff, z_inv2, point_x);
1370 ecp_nistz256_from_mont(x_ret, x_aff);
1371 if (!ecp_nistz256_set_words(x, x_ret))
1376 ecp_nistz256_mul_mont(z_inv3, z_inv3, z_inv2);
1377 ecp_nistz256_mul_mont(y_aff, z_inv3, point_y);
1378 ecp_nistz256_from_mont(y_ret, y_aff);
1379 if (!ecp_nistz256_set_words(y, y_ret))
1386 static EC_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group)
1388 EC_PRE_COMP *ret = NULL;
1393 ret = (EC_PRE_COMP *)OPENSSL_malloc(sizeof(EC_PRE_COMP));
1396 ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1401 ret->w = 6; /* default */
1402 ret->precomp = NULL;
1403 ret->precomp_storage = NULL;
1404 ret->references = 1;
1408 static void *ecp_nistz256_pre_comp_dup(void *src_)
1410 EC_PRE_COMP *src = src_;
1412 /* no need to actually copy, these objects never change! */
1413 CRYPTO_add(&src->references, 1, CRYPTO_LOCK_EC_PRE_COMP);
1418 static void ecp_nistz256_pre_comp_free(void *pre_)
1421 EC_PRE_COMP *pre = pre_;
1426 i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1430 if (pre->precomp_storage)
1431 OPENSSL_free(pre->precomp_storage);
1436 static void ecp_nistz256_pre_comp_clear_free(void *pre_)
1439 EC_PRE_COMP *pre = pre_;
1444 i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1448 if (pre->precomp_storage) {
1449 OPENSSL_cleanse(pre->precomp,
1450 32 * sizeof(unsigned char) * (1 << pre->w) * 2 * 37);
1451 OPENSSL_free(pre->precomp_storage);
1453 OPENSSL_cleanse(pre, sizeof *pre);
1457 static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP *group)
1459 /* There is a hard-coded table for the default generator. */
1460 const EC_POINT *generator = EC_GROUP_get0_generator(group);
1461 if (generator != NULL && ecp_nistz256_is_affine_G(generator)) {
1462 /* There is a hard-coded table for the default generator. */
1466 return EC_EX_DATA_get_data(group->extra_data, ecp_nistz256_pre_comp_dup,
1467 ecp_nistz256_pre_comp_free,
1468 ecp_nistz256_pre_comp_clear_free) != NULL;
1471 const EC_METHOD *EC_GFp_nistz256_method(void)
1473 static const EC_METHOD ret = {
1474 EC_FLAGS_DEFAULT_OCT,
1475 NID_X9_62_prime_field,
1476 ec_GFp_mont_group_init,
1477 ec_GFp_mont_group_finish,
1478 ec_GFp_mont_group_clear_finish,
1479 ec_GFp_mont_group_copy,
1480 ec_GFp_mont_group_set_curve,
1481 ec_GFp_simple_group_get_curve,
1482 ec_GFp_simple_group_get_degree,
1483 ec_GFp_simple_group_check_discriminant,
1484 ec_GFp_simple_point_init,
1485 ec_GFp_simple_point_finish,
1486 ec_GFp_simple_point_clear_finish,
1487 ec_GFp_simple_point_copy,
1488 ec_GFp_simple_point_set_to_infinity,
1489 ec_GFp_simple_set_Jprojective_coordinates_GFp,
1490 ec_GFp_simple_get_Jprojective_coordinates_GFp,
1491 ec_GFp_simple_point_set_affine_coordinates,
1492 ecp_nistz256_get_affine,
1496 ec_GFp_simple_invert,
1497 ec_GFp_simple_is_at_infinity,
1498 ec_GFp_simple_is_on_curve,
1500 ec_GFp_simple_make_affine,
1501 ec_GFp_simple_points_make_affine,
1502 ecp_nistz256_points_mul, /* mul */
1503 ecp_nistz256_mult_precompute, /* precompute_mult */
1504 ecp_nistz256_window_have_precompute_mult, /* have_precompute_mult */
1505 ec_GFp_mont_field_mul,
1506 ec_GFp_mont_field_sqr,
1508 ec_GFp_mont_field_encode,
1509 ec_GFp_mont_field_decode,
1510 ec_GFp_mont_field_set_to_one