2 * Copyright 2014-2016 The OpenSSL Project Authors. All Rights Reserved.
4 * Licensed under the OpenSSL license (the "License"). You may not use
5 * this file except in compliance with the License. You can obtain a copy
6 * in the file LICENSE in the source distribution or at
7 * https://www.openssl.org/source/license.html
10 /******************************************************************************
12 * Copyright 2014 Intel Corporation *
14 * Licensed under the Apache License, Version 2.0 (the "License"); *
15 * you may not use this file except in compliance with the License. *
16 * You may obtain a copy of the License at *
18 * http://www.apache.org/licenses/LICENSE-2.0 *
20 * Unless required by applicable law or agreed to in writing, software *
21 * distributed under the License is distributed on an "AS IS" BASIS, *
22 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. *
23 * See the License for the specific language governing permissions and *
24 * limitations under the License. *
26 ******************************************************************************
28 * Developers and authors: *
29 * Shay Gueron (1, 2), and Vlad Krasnov (1) *
30 * (1) Intel Corporation, Israel Development Center *
31 * (2) University of Haifa *
33 * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with *
36 ******************************************************************************/
40 #include "internal/cryptlib.h"
41 #include "internal/bn_int.h"
45 # define TOBN(hi,lo) lo,hi
47 # define TOBN(hi,lo) ((BN_ULONG)hi<<32|lo)
51 # define ALIGN32 __attribute((aligned(32)))
52 #elif defined(_MSC_VER)
53 # define ALIGN32 __declspec(align(32))
58 #define ALIGNPTR(p,N) ((unsigned char *)p+N-(size_t)p%N)
59 #define P256_LIMBS (256/BN_BITS2)
61 typedef unsigned short u16;
64 BN_ULONG X[P256_LIMBS];
65 BN_ULONG Y[P256_LIMBS];
66 BN_ULONG Z[P256_LIMBS];
70 BN_ULONG X[P256_LIMBS];
71 BN_ULONG Y[P256_LIMBS];
74 typedef P256_POINT_AFFINE PRECOMP256_ROW[64];
76 /* structure for precomputed multiples of the generator */
77 struct nistz256_pre_comp_st {
78 const EC_GROUP *group; /* Parent EC_GROUP object */
79 size_t w; /* Window size */
81 * Constant time access to the X and Y coordinates of the pre-computed,
82 * generator multiplies, in the Montgomery domain. Pre-calculated
83 * multiplies are stored in affine form.
85 PRECOMP256_ROW *precomp;
86 void *precomp_storage;
91 /* Functions implemented in assembly */
92 /* Modular mul by 2: res = 2*a mod P */
93 void ecp_nistz256_mul_by_2(BN_ULONG res[P256_LIMBS],
94 const BN_ULONG a[P256_LIMBS]);
95 /* Modular div by 2: res = a/2 mod P */
96 void ecp_nistz256_div_by_2(BN_ULONG res[P256_LIMBS],
97 const BN_ULONG a[P256_LIMBS]);
98 /* Modular mul by 3: res = 3*a mod P */
99 void ecp_nistz256_mul_by_3(BN_ULONG res[P256_LIMBS],
100 const BN_ULONG a[P256_LIMBS]);
101 /* Modular add: res = a+b mod P */
102 void ecp_nistz256_add(BN_ULONG res[P256_LIMBS],
103 const BN_ULONG a[P256_LIMBS],
104 const BN_ULONG b[P256_LIMBS]);
105 /* Modular sub: res = a-b mod P */
106 void ecp_nistz256_sub(BN_ULONG res[P256_LIMBS],
107 const BN_ULONG a[P256_LIMBS],
108 const BN_ULONG b[P256_LIMBS]);
109 /* Modular neg: res = -a mod P */
110 void ecp_nistz256_neg(BN_ULONG res[P256_LIMBS], const BN_ULONG a[P256_LIMBS]);
111 /* Montgomery mul: res = a*b*2^-256 mod P */
112 void ecp_nistz256_mul_mont(BN_ULONG res[P256_LIMBS],
113 const BN_ULONG a[P256_LIMBS],
114 const BN_ULONG b[P256_LIMBS]);
115 /* Montgomery sqr: res = a*a*2^-256 mod P */
116 void ecp_nistz256_sqr_mont(BN_ULONG res[P256_LIMBS],
117 const BN_ULONG a[P256_LIMBS]);
118 /* Convert a number from Montgomery domain, by multiplying with 1 */
119 void ecp_nistz256_from_mont(BN_ULONG res[P256_LIMBS],
120 const BN_ULONG in[P256_LIMBS]);
121 /* Convert a number to Montgomery domain, by multiplying with 2^512 mod P*/
122 void ecp_nistz256_to_mont(BN_ULONG res[P256_LIMBS],
123 const BN_ULONG in[P256_LIMBS]);
124 /* Functions that perform constant time access to the precomputed tables */
125 void ecp_nistz256_scatter_w5(P256_POINT *val,
126 const P256_POINT *in_t, int idx);
127 void ecp_nistz256_gather_w5(P256_POINT *val,
128 const P256_POINT *in_t, int idx);
129 void ecp_nistz256_scatter_w7(P256_POINT_AFFINE *val,
130 const P256_POINT_AFFINE *in_t, int idx);
131 void ecp_nistz256_gather_w7(P256_POINT_AFFINE *val,
132 const P256_POINT_AFFINE *in_t, int idx);
134 /* One converted into the Montgomery domain */
135 static const BN_ULONG ONE[P256_LIMBS] = {
136 TOBN(0x00000000, 0x00000001), TOBN(0xffffffff, 0x00000000),
137 TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe)
140 static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group);
142 /* Precomputed tables for the default generator */
143 extern const PRECOMP256_ROW ecp_nistz256_precomputed[37];
145 /* Recode window to a signed digit, see ecp_nistputil.c for details */
146 static unsigned int _booth_recode_w5(unsigned int in)
150 s = ~((in >> 5) - 1);
151 d = (1 << 6) - in - 1;
152 d = (d & s) | (in & ~s);
153 d = (d >> 1) + (d & 1);
155 return (d << 1) + (s & 1);
158 static unsigned int _booth_recode_w7(unsigned int in)
162 s = ~((in >> 7) - 1);
163 d = (1 << 8) - in - 1;
164 d = (d & s) | (in & ~s);
165 d = (d >> 1) + (d & 1);
167 return (d << 1) + (s & 1);
170 static void copy_conditional(BN_ULONG dst[P256_LIMBS],
171 const BN_ULONG src[P256_LIMBS], BN_ULONG move)
173 BN_ULONG mask1 = 0-move;
174 BN_ULONG mask2 = ~mask1;
176 dst[0] = (src[0] & mask1) ^ (dst[0] & mask2);
177 dst[1] = (src[1] & mask1) ^ (dst[1] & mask2);
178 dst[2] = (src[2] & mask1) ^ (dst[2] & mask2);
179 dst[3] = (src[3] & mask1) ^ (dst[3] & mask2);
180 if (P256_LIMBS == 8) {
181 dst[4] = (src[4] & mask1) ^ (dst[4] & mask2);
182 dst[5] = (src[5] & mask1) ^ (dst[5] & mask2);
183 dst[6] = (src[6] & mask1) ^ (dst[6] & mask2);
184 dst[7] = (src[7] & mask1) ^ (dst[7] & mask2);
188 static BN_ULONG is_zero(BN_ULONG in)
196 static BN_ULONG is_equal(const BN_ULONG a[P256_LIMBS],
197 const BN_ULONG b[P256_LIMBS])
205 if (P256_LIMBS == 8) {
215 static BN_ULONG is_one(const BIGNUM *z)
218 BN_ULONG *a = bn_get_words(z);
220 if (bn_get_top(z) == (P256_LIMBS - P256_LIMBS / 8)) {
222 res |= a[1] ^ ONE[1];
223 res |= a[2] ^ ONE[2];
224 res |= a[3] ^ ONE[3];
225 if (P256_LIMBS == 8) {
226 res |= a[4] ^ ONE[4];
227 res |= a[5] ^ ONE[5];
228 res |= a[6] ^ ONE[6];
230 * no check for a[7] (being zero) on 32-bit platforms,
231 * because value of "one" takes only 7 limbs.
240 #ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
241 void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a);
242 void ecp_nistz256_point_add(P256_POINT *r,
243 const P256_POINT *a, const P256_POINT *b);
244 void ecp_nistz256_point_add_affine(P256_POINT *r,
246 const P256_POINT_AFFINE *b);
248 /* Point double: r = 2*a */
249 static void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a)
251 BN_ULONG S[P256_LIMBS];
252 BN_ULONG M[P256_LIMBS];
253 BN_ULONG Zsqr[P256_LIMBS];
254 BN_ULONG tmp0[P256_LIMBS];
256 const BN_ULONG *in_x = a->X;
257 const BN_ULONG *in_y = a->Y;
258 const BN_ULONG *in_z = a->Z;
260 BN_ULONG *res_x = r->X;
261 BN_ULONG *res_y = r->Y;
262 BN_ULONG *res_z = r->Z;
264 ecp_nistz256_mul_by_2(S, in_y);
266 ecp_nistz256_sqr_mont(Zsqr, in_z);
268 ecp_nistz256_sqr_mont(S, S);
270 ecp_nistz256_mul_mont(res_z, in_z, in_y);
271 ecp_nistz256_mul_by_2(res_z, res_z);
273 ecp_nistz256_add(M, in_x, Zsqr);
274 ecp_nistz256_sub(Zsqr, in_x, Zsqr);
276 ecp_nistz256_sqr_mont(res_y, S);
277 ecp_nistz256_div_by_2(res_y, res_y);
279 ecp_nistz256_mul_mont(M, M, Zsqr);
280 ecp_nistz256_mul_by_3(M, M);
282 ecp_nistz256_mul_mont(S, S, in_x);
283 ecp_nistz256_mul_by_2(tmp0, S);
285 ecp_nistz256_sqr_mont(res_x, M);
287 ecp_nistz256_sub(res_x, res_x, tmp0);
288 ecp_nistz256_sub(S, S, res_x);
290 ecp_nistz256_mul_mont(S, S, M);
291 ecp_nistz256_sub(res_y, S, res_y);
294 /* Point addition: r = a+b */
295 static void ecp_nistz256_point_add(P256_POINT *r,
296 const P256_POINT *a, const P256_POINT *b)
298 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
299 BN_ULONG U1[P256_LIMBS], S1[P256_LIMBS];
300 BN_ULONG Z1sqr[P256_LIMBS];
301 BN_ULONG Z2sqr[P256_LIMBS];
302 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
303 BN_ULONG Hsqr[P256_LIMBS];
304 BN_ULONG Rsqr[P256_LIMBS];
305 BN_ULONG Hcub[P256_LIMBS];
307 BN_ULONG res_x[P256_LIMBS];
308 BN_ULONG res_y[P256_LIMBS];
309 BN_ULONG res_z[P256_LIMBS];
311 BN_ULONG in1infty, in2infty;
313 const BN_ULONG *in1_x = a->X;
314 const BN_ULONG *in1_y = a->Y;
315 const BN_ULONG *in1_z = a->Z;
317 const BN_ULONG *in2_x = b->X;
318 const BN_ULONG *in2_y = b->Y;
319 const BN_ULONG *in2_z = b->Z;
321 /* We encode infinity as (0,0), which is not on the curve,
323 in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
324 in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]);
326 in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
327 in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]);
329 in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
330 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
332 in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
333 in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
335 in1infty = is_zero(in1infty);
336 in2infty = is_zero(in2infty);
338 ecp_nistz256_sqr_mont(Z2sqr, in2_z); /* Z2^2 */
339 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
341 ecp_nistz256_mul_mont(S1, Z2sqr, in2_z); /* S1 = Z2^3 */
342 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
344 ecp_nistz256_mul_mont(S1, S1, in1_y); /* S1 = Y1*Z2^3 */
345 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
346 ecp_nistz256_sub(R, S2, S1); /* R = S2 - S1 */
348 ecp_nistz256_mul_mont(U1, in1_x, Z2sqr); /* U1 = X1*Z2^2 */
349 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
350 ecp_nistz256_sub(H, U2, U1); /* H = U2 - U1 */
353 * This should not happen during sign/ecdh, so no constant time violation
355 if (is_equal(U1, U2) && !in1infty && !in2infty) {
356 if (is_equal(S1, S2)) {
357 ecp_nistz256_point_double(r, a);
360 memset(r, 0, sizeof(*r));
365 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
366 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
367 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
368 ecp_nistz256_mul_mont(res_z, res_z, in2_z); /* Z3 = H*Z1*Z2 */
369 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
371 ecp_nistz256_mul_mont(U2, U1, Hsqr); /* U1*H^2 */
372 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
374 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
375 ecp_nistz256_sub(res_x, res_x, Hcub);
377 ecp_nistz256_sub(res_y, U2, res_x);
379 ecp_nistz256_mul_mont(S2, S1, Hcub);
380 ecp_nistz256_mul_mont(res_y, R, res_y);
381 ecp_nistz256_sub(res_y, res_y, S2);
383 copy_conditional(res_x, in2_x, in1infty);
384 copy_conditional(res_y, in2_y, in1infty);
385 copy_conditional(res_z, in2_z, in1infty);
387 copy_conditional(res_x, in1_x, in2infty);
388 copy_conditional(res_y, in1_y, in2infty);
389 copy_conditional(res_z, in1_z, in2infty);
391 memcpy(r->X, res_x, sizeof(res_x));
392 memcpy(r->Y, res_y, sizeof(res_y));
393 memcpy(r->Z, res_z, sizeof(res_z));
396 /* Point addition when b is known to be affine: r = a+b */
397 static void ecp_nistz256_point_add_affine(P256_POINT *r,
399 const P256_POINT_AFFINE *b)
401 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
402 BN_ULONG Z1sqr[P256_LIMBS];
403 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
404 BN_ULONG Hsqr[P256_LIMBS];
405 BN_ULONG Rsqr[P256_LIMBS];
406 BN_ULONG Hcub[P256_LIMBS];
408 BN_ULONG res_x[P256_LIMBS];
409 BN_ULONG res_y[P256_LIMBS];
410 BN_ULONG res_z[P256_LIMBS];
412 BN_ULONG in1infty, in2infty;
414 const BN_ULONG *in1_x = a->X;
415 const BN_ULONG *in1_y = a->Y;
416 const BN_ULONG *in1_z = a->Z;
418 const BN_ULONG *in2_x = b->X;
419 const BN_ULONG *in2_y = b->Y;
422 * In affine representation we encode infty as (0,0), which is not on the
425 in1infty = (in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
426 in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3]);
428 in1infty |= (in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
429 in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7]);
431 in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
432 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
434 in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
435 in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
437 in1infty = is_zero(in1infty);
438 in2infty = is_zero(in2infty);
440 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
442 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
443 ecp_nistz256_sub(H, U2, in1_x); /* H = U2 - U1 */
445 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
447 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
449 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
450 ecp_nistz256_sub(R, S2, in1_y); /* R = S2 - S1 */
452 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
453 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
454 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
456 ecp_nistz256_mul_mont(U2, in1_x, Hsqr); /* U1*H^2 */
457 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
459 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
460 ecp_nistz256_sub(res_x, res_x, Hcub);
461 ecp_nistz256_sub(H, U2, res_x);
463 ecp_nistz256_mul_mont(S2, in1_y, Hcub);
464 ecp_nistz256_mul_mont(H, H, R);
465 ecp_nistz256_sub(res_y, H, S2);
467 copy_conditional(res_x, in2_x, in1infty);
468 copy_conditional(res_x, in1_x, in2infty);
470 copy_conditional(res_y, in2_y, in1infty);
471 copy_conditional(res_y, in1_y, in2infty);
473 copy_conditional(res_z, ONE, in1infty);
474 copy_conditional(res_z, in1_z, in2infty);
476 memcpy(r->X, res_x, sizeof(res_x));
477 memcpy(r->Y, res_y, sizeof(res_y));
478 memcpy(r->Z, res_z, sizeof(res_z));
482 /* r = in^-1 mod p */
483 static void ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS],
484 const BN_ULONG in[P256_LIMBS])
487 * The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff
488 * ffffffff ffffffff We use FLT and used poly-2 as exponent
490 BN_ULONG p2[P256_LIMBS];
491 BN_ULONG p4[P256_LIMBS];
492 BN_ULONG p8[P256_LIMBS];
493 BN_ULONG p16[P256_LIMBS];
494 BN_ULONG p32[P256_LIMBS];
495 BN_ULONG res[P256_LIMBS];
498 ecp_nistz256_sqr_mont(res, in);
499 ecp_nistz256_mul_mont(p2, res, in); /* 3*p */
501 ecp_nistz256_sqr_mont(res, p2);
502 ecp_nistz256_sqr_mont(res, res);
503 ecp_nistz256_mul_mont(p4, res, p2); /* f*p */
505 ecp_nistz256_sqr_mont(res, p4);
506 ecp_nistz256_sqr_mont(res, res);
507 ecp_nistz256_sqr_mont(res, res);
508 ecp_nistz256_sqr_mont(res, res);
509 ecp_nistz256_mul_mont(p8, res, p4); /* ff*p */
511 ecp_nistz256_sqr_mont(res, p8);
512 for (i = 0; i < 7; i++)
513 ecp_nistz256_sqr_mont(res, res);
514 ecp_nistz256_mul_mont(p16, res, p8); /* ffff*p */
516 ecp_nistz256_sqr_mont(res, p16);
517 for (i = 0; i < 15; i++)
518 ecp_nistz256_sqr_mont(res, res);
519 ecp_nistz256_mul_mont(p32, res, p16); /* ffffffff*p */
521 ecp_nistz256_sqr_mont(res, p32);
522 for (i = 0; i < 31; i++)
523 ecp_nistz256_sqr_mont(res, res);
524 ecp_nistz256_mul_mont(res, res, in);
526 for (i = 0; i < 32 * 4; i++)
527 ecp_nistz256_sqr_mont(res, res);
528 ecp_nistz256_mul_mont(res, res, p32);
530 for (i = 0; i < 32; i++)
531 ecp_nistz256_sqr_mont(res, res);
532 ecp_nistz256_mul_mont(res, res, p32);
534 for (i = 0; i < 16; i++)
535 ecp_nistz256_sqr_mont(res, res);
536 ecp_nistz256_mul_mont(res, res, p16);
538 for (i = 0; i < 8; i++)
539 ecp_nistz256_sqr_mont(res, res);
540 ecp_nistz256_mul_mont(res, res, p8);
542 ecp_nistz256_sqr_mont(res, res);
543 ecp_nistz256_sqr_mont(res, res);
544 ecp_nistz256_sqr_mont(res, res);
545 ecp_nistz256_sqr_mont(res, res);
546 ecp_nistz256_mul_mont(res, res, p4);
548 ecp_nistz256_sqr_mont(res, res);
549 ecp_nistz256_sqr_mont(res, res);
550 ecp_nistz256_mul_mont(res, res, p2);
552 ecp_nistz256_sqr_mont(res, res);
553 ecp_nistz256_sqr_mont(res, res);
554 ecp_nistz256_mul_mont(res, res, in);
556 memcpy(r, res, sizeof(res));
560 * ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
561 * returns one if it fits. Otherwise it returns zero.
563 __owur static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],
566 return bn_copy_words(out, in, P256_LIMBS);
569 /* r = sum(scalar[i]*point[i]) */
570 __owur static int ecp_nistz256_windowed_mul(const EC_GROUP *group,
572 const BIGNUM **scalar,
573 const EC_POINT **point,
574 size_t num, BN_CTX *ctx)
579 unsigned char (*p_str)[33] = NULL;
580 const unsigned int window_size = 5;
581 const unsigned int mask = (1 << (window_size + 1)) - 1;
583 P256_POINT *temp; /* place for 5 temporary points */
584 const BIGNUM **scalars = NULL;
585 P256_POINT (*table)[16] = NULL;
586 void *table_storage = NULL;
588 if ((num * 16 + 6) > OPENSSL_MALLOC_MAX_NELEMS(P256_POINT)
590 OPENSSL_malloc((num * 16 + 5) * sizeof(P256_POINT) + 64)) == NULL
592 OPENSSL_malloc(num * 33 * sizeof(unsigned char))) == NULL
593 || (scalars = OPENSSL_malloc(num * sizeof(BIGNUM *))) == NULL) {
594 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_MALLOC_FAILURE);
598 table = (void *)ALIGNPTR(table_storage, 64);
599 temp = (P256_POINT *)(table + num);
601 for (i = 0; i < num; i++) {
602 P256_POINT *row = table[i];
604 /* This is an unusual input, we don't guarantee constant-timeness. */
605 if ((BN_num_bits(scalar[i]) > 256) || BN_is_negative(scalar[i])) {
608 if ((mod = BN_CTX_get(ctx)) == NULL)
610 if (!BN_nnmod(mod, scalar[i], group->order, ctx)) {
611 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_BN_LIB);
616 scalars[i] = scalar[i];
618 for (j = 0; j < bn_get_top(scalars[i]) * BN_BYTES; j += BN_BYTES) {
619 BN_ULONG d = bn_get_words(scalars[i])[j / BN_BYTES];
621 p_str[i][j + 0] = (unsigned char)d;
622 p_str[i][j + 1] = (unsigned char)(d >> 8);
623 p_str[i][j + 2] = (unsigned char)(d >> 16);
624 p_str[i][j + 3] = (unsigned char)(d >>= 24);
627 p_str[i][j + 4] = (unsigned char)d;
628 p_str[i][j + 5] = (unsigned char)(d >> 8);
629 p_str[i][j + 6] = (unsigned char)(d >> 16);
630 p_str[i][j + 7] = (unsigned char)(d >> 24);
636 if (!ecp_nistz256_bignum_to_field_elem(temp[0].X, point[i]->X)
637 || !ecp_nistz256_bignum_to_field_elem(temp[0].Y, point[i]->Y)
638 || !ecp_nistz256_bignum_to_field_elem(temp[0].Z, point[i]->Z)) {
639 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL,
640 EC_R_COORDINATES_OUT_OF_RANGE);
645 * row[0] is implicitly (0,0,0) (the point at infinity), therefore it
646 * is not stored. All other values are actually stored with an offset
650 ecp_nistz256_scatter_w5 (row, &temp[0], 1);
651 ecp_nistz256_point_double(&temp[1], &temp[0]); /*1+1=2 */
652 ecp_nistz256_scatter_w5 (row, &temp[1], 2);
653 ecp_nistz256_point_add (&temp[2], &temp[1], &temp[0]); /*2+1=3 */
654 ecp_nistz256_scatter_w5 (row, &temp[2], 3);
655 ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*2=4 */
656 ecp_nistz256_scatter_w5 (row, &temp[1], 4);
657 ecp_nistz256_point_double(&temp[2], &temp[2]); /*2*3=6 */
658 ecp_nistz256_scatter_w5 (row, &temp[2], 6);
659 ecp_nistz256_point_add (&temp[3], &temp[1], &temp[0]); /*4+1=5 */
660 ecp_nistz256_scatter_w5 (row, &temp[3], 5);
661 ecp_nistz256_point_add (&temp[4], &temp[2], &temp[0]); /*6+1=7 */
662 ecp_nistz256_scatter_w5 (row, &temp[4], 7);
663 ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*4=8 */
664 ecp_nistz256_scatter_w5 (row, &temp[1], 8);
665 ecp_nistz256_point_double(&temp[2], &temp[2]); /*2*6=12 */
666 ecp_nistz256_scatter_w5 (row, &temp[2], 12);
667 ecp_nistz256_point_double(&temp[3], &temp[3]); /*2*5=10 */
668 ecp_nistz256_scatter_w5 (row, &temp[3], 10);
669 ecp_nistz256_point_double(&temp[4], &temp[4]); /*2*7=14 */
670 ecp_nistz256_scatter_w5 (row, &temp[4], 14);
671 ecp_nistz256_point_add (&temp[2], &temp[2], &temp[0]); /*12+1=13*/
672 ecp_nistz256_scatter_w5 (row, &temp[2], 13);
673 ecp_nistz256_point_add (&temp[3], &temp[3], &temp[0]); /*10+1=11*/
674 ecp_nistz256_scatter_w5 (row, &temp[3], 11);
675 ecp_nistz256_point_add (&temp[4], &temp[4], &temp[0]); /*14+1=15*/
676 ecp_nistz256_scatter_w5 (row, &temp[4], 15);
677 ecp_nistz256_point_add (&temp[2], &temp[1], &temp[0]); /*8+1=9 */
678 ecp_nistz256_scatter_w5 (row, &temp[2], 9);
679 ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*8=16 */
680 ecp_nistz256_scatter_w5 (row, &temp[1], 16);
685 wvalue = p_str[0][(idx - 1) / 8];
686 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
689 * We gather to temp[0], because we know it's position relative
692 ecp_nistz256_gather_w5(&temp[0], table[0], _booth_recode_w5(wvalue) >> 1);
693 memcpy(r, &temp[0], sizeof(temp[0]));
696 for (i = (idx == 255 ? 1 : 0); i < num; i++) {
697 unsigned int off = (idx - 1) / 8;
699 wvalue = p_str[i][off] | p_str[i][off + 1] << 8;
700 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
702 wvalue = _booth_recode_w5(wvalue);
704 ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1);
706 ecp_nistz256_neg(temp[1].Y, temp[0].Y);
707 copy_conditional(temp[0].Y, temp[1].Y, (wvalue & 1));
709 ecp_nistz256_point_add(r, r, &temp[0]);
714 ecp_nistz256_point_double(r, r);
715 ecp_nistz256_point_double(r, r);
716 ecp_nistz256_point_double(r, r);
717 ecp_nistz256_point_double(r, r);
718 ecp_nistz256_point_double(r, r);
722 for (i = 0; i < num; i++) {
723 wvalue = p_str[i][0];
724 wvalue = (wvalue << 1) & mask;
726 wvalue = _booth_recode_w5(wvalue);
728 ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1);
730 ecp_nistz256_neg(temp[1].Y, temp[0].Y);
731 copy_conditional(temp[0].Y, temp[1].Y, wvalue & 1);
733 ecp_nistz256_point_add(r, r, &temp[0]);
738 OPENSSL_free(table_storage);
740 OPENSSL_free(scalars);
744 /* Coordinates of G, for which we have precomputed tables */
745 const static BN_ULONG def_xG[P256_LIMBS] = {
746 TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
747 TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
750 const static BN_ULONG def_yG[P256_LIMBS] = {
751 TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
752 TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
756 * ecp_nistz256_is_affine_G returns one if |generator| is the standard, P-256
759 static int ecp_nistz256_is_affine_G(const EC_POINT *generator)
761 return (bn_get_top(generator->X) == P256_LIMBS) &&
762 (bn_get_top(generator->Y) == P256_LIMBS) &&
763 is_equal(bn_get_words(generator->X), def_xG) &&
764 is_equal(bn_get_words(generator->Y), def_yG) &&
765 is_one(generator->Z);
768 __owur static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx)
771 * We precompute a table for a Booth encoded exponent (wNAF) based
772 * computation. Each table holds 64 values for safe access, with an
773 * implicit value of infinity at index zero. We use window of size 7, and
774 * therefore require ceil(256/7) = 37 tables.
777 EC_POINT *P = NULL, *T = NULL;
778 const EC_POINT *generator;
779 NISTZ256_PRE_COMP *pre_comp;
780 BN_CTX *new_ctx = NULL;
781 int i, j, k, ret = 0;
784 PRECOMP256_ROW *preComputedTable = NULL;
785 unsigned char *precomp_storage = NULL;
787 /* if there is an old NISTZ256_PRE_COMP object, throw it away */
788 EC_pre_comp_free(group);
789 generator = EC_GROUP_get0_generator(group);
790 if (generator == NULL) {
791 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNDEFINED_GENERATOR);
795 if (ecp_nistz256_is_affine_G(generator)) {
797 * No need to calculate tables for the standard generator because we
798 * have them statically.
803 if ((pre_comp = ecp_nistz256_pre_comp_new(group)) == NULL)
807 ctx = new_ctx = BN_CTX_new();
814 order = EC_GROUP_get0_order(group);
818 if (BN_is_zero(order)) {
819 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNKNOWN_ORDER);
825 if ((precomp_storage =
826 OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE) + 64)) == NULL) {
827 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, ERR_R_MALLOC_FAILURE);
831 preComputedTable = (void *)ALIGNPTR(precomp_storage, 64);
833 P = EC_POINT_new(group);
834 T = EC_POINT_new(group);
835 if (P == NULL || T == NULL)
839 * The zero entry is implicitly infinity, and we skip it, storing other
840 * values with -1 offset.
842 if (!EC_POINT_copy(T, generator))
845 for (k = 0; k < 64; k++) {
846 if (!EC_POINT_copy(P, T))
848 for (j = 0; j < 37; j++) {
849 P256_POINT_AFFINE temp;
851 * It would be faster to use EC_POINTs_make_affine and
852 * make multiple points affine at the same time.
854 if (!EC_POINT_make_affine(group, P, ctx))
856 if (!ecp_nistz256_bignum_to_field_elem(temp.X, P->X) ||
857 !ecp_nistz256_bignum_to_field_elem(temp.Y, P->Y)) {
858 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE,
859 EC_R_COORDINATES_OUT_OF_RANGE);
862 ecp_nistz256_scatter_w7(preComputedTable[j], &temp, k);
863 for (i = 0; i < 7; i++) {
864 if (!EC_POINT_dbl(group, P, P, ctx))
868 if (!EC_POINT_add(group, T, T, generator, ctx))
872 pre_comp->group = group;
874 pre_comp->precomp = preComputedTable;
875 pre_comp->precomp_storage = precomp_storage;
876 precomp_storage = NULL;
877 SETPRECOMP(group, nistz256, pre_comp);
884 BN_CTX_free(new_ctx);
886 EC_nistz256_pre_comp_free(pre_comp);
887 OPENSSL_free(precomp_storage);
894 * Note that by default ECP_NISTZ256_AVX2 is undefined. While it's great
895 * code processing 4 points in parallel, corresponding serial operation
896 * is several times slower, because it uses 29x29=58-bit multiplication
897 * as opposite to 64x64=128-bit in integer-only scalar case. As result
898 * it doesn't provide *significant* performance improvement. Note that
899 * just defining ECP_NISTZ256_AVX2 is not sufficient to make it work,
900 * you'd need to compile even asm/ecp_nistz256-avx.pl module.
902 #if defined(ECP_NISTZ256_AVX2)
903 # if !(defined(__x86_64) || defined(__x86_64__) || \
904 defined(_M_AMD64) || defined(_MX64)) || \
905 !(defined(__GNUC__) || defined(_MSC_VER)) /* this is for ALIGN32 */
906 # undef ECP_NISTZ256_AVX2
908 /* Constant time access, loading four values, from four consecutive tables */
909 void ecp_nistz256_avx2_multi_gather_w7(void *result, const void *in,
910 int index0, int index1, int index2,
912 void ecp_nistz256_avx2_transpose_convert(void *RESULTx4, const void *in);
913 void ecp_nistz256_avx2_convert_transpose_back(void *result, const void *Ax4);
914 void ecp_nistz256_avx2_point_add_affine_x4(void *RESULTx4, const void *Ax4,
916 void ecp_nistz256_avx2_point_add_affines_x4(void *RESULTx4, const void *Ax4,
918 void ecp_nistz256_avx2_to_mont(void *RESULTx4, const void *Ax4);
919 void ecp_nistz256_avx2_from_mont(void *RESULTx4, const void *Ax4);
920 void ecp_nistz256_avx2_set1(void *RESULTx4);
921 int ecp_nistz_avx2_eligible(void);
923 static void booth_recode_w7(unsigned char *sign,
924 unsigned char *digit, unsigned char in)
928 s = ~((in >> 7) - 1);
929 d = (1 << 8) - in - 1;
930 d = (d & s) | (in & ~s);
931 d = (d >> 1) + (d & 1);
938 * ecp_nistz256_avx2_mul_g performs multiplication by G, using only the
939 * precomputed table. It does 4 affine point additions in parallel,
940 * significantly speeding up point multiplication for a fixed value.
942 static void ecp_nistz256_avx2_mul_g(P256_POINT *r,
943 unsigned char p_str[33],
944 const P256_POINT_AFFINE(*preComputedTable)[64])
946 const unsigned int window_size = 7;
947 const unsigned int mask = (1 << (window_size + 1)) - 1;
949 /* Using 4 windows at a time */
950 unsigned char sign0, digit0;
951 unsigned char sign1, digit1;
952 unsigned char sign2, digit2;
953 unsigned char sign3, digit3;
954 unsigned int idx = 0;
955 BN_ULONG tmp[P256_LIMBS];
958 ALIGN32 BN_ULONG aX4[4 * 9 * 3] = { 0 };
959 ALIGN32 BN_ULONG bX4[4 * 9 * 2] = { 0 };
960 ALIGN32 P256_POINT_AFFINE point_arr[4];
961 ALIGN32 P256_POINT res_point_arr[4];
963 /* Initial four windows */
964 wvalue = *((u16 *) & p_str[0]);
965 wvalue = (wvalue << 1) & mask;
967 booth_recode_w7(&sign0, &digit0, wvalue);
968 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
969 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
971 booth_recode_w7(&sign1, &digit1, wvalue);
972 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
973 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
975 booth_recode_w7(&sign2, &digit2, wvalue);
976 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
977 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
979 booth_recode_w7(&sign3, &digit3, wvalue);
981 ecp_nistz256_avx2_multi_gather_w7(point_arr, preComputedTable[0],
982 digit0, digit1, digit2, digit3);
984 ecp_nistz256_neg(tmp, point_arr[0].Y);
985 copy_conditional(point_arr[0].Y, tmp, sign0);
986 ecp_nistz256_neg(tmp, point_arr[1].Y);
987 copy_conditional(point_arr[1].Y, tmp, sign1);
988 ecp_nistz256_neg(tmp, point_arr[2].Y);
989 copy_conditional(point_arr[2].Y, tmp, sign2);
990 ecp_nistz256_neg(tmp, point_arr[3].Y);
991 copy_conditional(point_arr[3].Y, tmp, sign3);
993 ecp_nistz256_avx2_transpose_convert(aX4, point_arr);
994 ecp_nistz256_avx2_to_mont(aX4, aX4);
995 ecp_nistz256_avx2_to_mont(&aX4[4 * 9], &aX4[4 * 9]);
996 ecp_nistz256_avx2_set1(&aX4[4 * 9 * 2]);
998 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
999 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1001 booth_recode_w7(&sign0, &digit0, wvalue);
1002 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
1003 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1005 booth_recode_w7(&sign1, &digit1, wvalue);
1006 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
1007 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1009 booth_recode_w7(&sign2, &digit2, wvalue);
1010 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
1011 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1013 booth_recode_w7(&sign3, &digit3, wvalue);
1015 ecp_nistz256_avx2_multi_gather_w7(point_arr, preComputedTable[4 * 1],
1016 digit0, digit1, digit2, digit3);
1018 ecp_nistz256_neg(tmp, point_arr[0].Y);
1019 copy_conditional(point_arr[0].Y, tmp, sign0);
1020 ecp_nistz256_neg(tmp, point_arr[1].Y);
1021 copy_conditional(point_arr[1].Y, tmp, sign1);
1022 ecp_nistz256_neg(tmp, point_arr[2].Y);
1023 copy_conditional(point_arr[2].Y, tmp, sign2);
1024 ecp_nistz256_neg(tmp, point_arr[3].Y);
1025 copy_conditional(point_arr[3].Y, tmp, sign3);
1027 ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
1028 ecp_nistz256_avx2_to_mont(bX4, bX4);
1029 ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
1030 /* Optimized when both inputs are affine */
1031 ecp_nistz256_avx2_point_add_affines_x4(aX4, aX4, bX4);
1033 for (i = 2; i < 9; i++) {
1034 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
1035 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1037 booth_recode_w7(&sign0, &digit0, wvalue);
1038 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
1039 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1041 booth_recode_w7(&sign1, &digit1, wvalue);
1042 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
1043 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1045 booth_recode_w7(&sign2, &digit2, wvalue);
1046 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
1047 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1049 booth_recode_w7(&sign3, &digit3, wvalue);
1051 ecp_nistz256_avx2_multi_gather_w7(point_arr,
1052 preComputedTable[4 * i],
1053 digit0, digit1, digit2, digit3);
1055 ecp_nistz256_neg(tmp, point_arr[0].Y);
1056 copy_conditional(point_arr[0].Y, tmp, sign0);
1057 ecp_nistz256_neg(tmp, point_arr[1].Y);
1058 copy_conditional(point_arr[1].Y, tmp, sign1);
1059 ecp_nistz256_neg(tmp, point_arr[2].Y);
1060 copy_conditional(point_arr[2].Y, tmp, sign2);
1061 ecp_nistz256_neg(tmp, point_arr[3].Y);
1062 copy_conditional(point_arr[3].Y, tmp, sign3);
1064 ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
1065 ecp_nistz256_avx2_to_mont(bX4, bX4);
1066 ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
1068 ecp_nistz256_avx2_point_add_affine_x4(aX4, aX4, bX4);
1071 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 0], &aX4[4 * 9 * 0]);
1072 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 1], &aX4[4 * 9 * 1]);
1073 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 2], &aX4[4 * 9 * 2]);
1075 ecp_nistz256_avx2_convert_transpose_back(res_point_arr, aX4);
1076 /* Last window is performed serially */
1077 wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
1078 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1079 booth_recode_w7(&sign0, &digit0, wvalue);
1080 ecp_nistz256_gather_w7((P256_POINT_AFFINE *)r,
1081 preComputedTable[36], digit0);
1082 ecp_nistz256_neg(tmp, r->Y);
1083 copy_conditional(r->Y, tmp, sign0);
1084 memcpy(r->Z, ONE, sizeof(ONE));
1085 /* Sum the four windows */
1086 ecp_nistz256_point_add(r, r, &res_point_arr[0]);
1087 ecp_nistz256_point_add(r, r, &res_point_arr[1]);
1088 ecp_nistz256_point_add(r, r, &res_point_arr[2]);
1089 ecp_nistz256_point_add(r, r, &res_point_arr[3]);
1094 __owur static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group,
1095 const P256_POINT_AFFINE *in,
1099 BN_ULONG d_x[P256_LIMBS], d_y[P256_LIMBS];
1110 memcpy(d_x, in->X, sizeof(d_x));
1111 bn_set_static_words(x, d_x, P256_LIMBS);
1113 memcpy(d_y, in->Y, sizeof(d_y));
1114 bn_set_static_words(y, d_y, P256_LIMBS);
1116 ret = EC_POINT_set_affine_coordinates_GFp(group, out, x, y, ctx);
1124 /* r = scalar*G + sum(scalars[i]*points[i]) */
1125 __owur static int ecp_nistz256_points_mul(const EC_GROUP *group,
1127 const BIGNUM *scalar,
1129 const EC_POINT *points[],
1130 const BIGNUM *scalars[], BN_CTX *ctx)
1132 int i = 0, ret = 0, no_precomp_for_generator = 0, p_is_infinity = 0;
1134 unsigned char p_str[33] = { 0 };
1135 const PRECOMP256_ROW *preComputedTable = NULL;
1136 const NISTZ256_PRE_COMP *pre_comp = NULL;
1137 const EC_POINT *generator = NULL;
1138 BN_CTX *new_ctx = NULL;
1139 const BIGNUM **new_scalars = NULL;
1140 const EC_POINT **new_points = NULL;
1141 unsigned int idx = 0;
1142 const unsigned int window_size = 7;
1143 const unsigned int mask = (1 << (window_size + 1)) - 1;
1144 unsigned int wvalue;
1147 P256_POINT_AFFINE a;
1151 if ((num + 1) == 0 || (num + 1) > OPENSSL_MALLOC_MAX_NELEMS(void *)) {
1152 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1156 if (group->meth != r->meth) {
1157 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
1161 if ((scalar == NULL) && (num == 0))
1162 return EC_POINT_set_to_infinity(group, r);
1164 for (j = 0; j < num; j++) {
1165 if (group->meth != points[j]->meth) {
1166 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
1172 ctx = new_ctx = BN_CTX_new();
1180 generator = EC_GROUP_get0_generator(group);
1181 if (generator == NULL) {
1182 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_UNDEFINED_GENERATOR);
1186 /* look if we can use precomputed multiples of generator */
1187 pre_comp = group->pre_comp.nistz256;
1191 * If there is a precomputed table for the generator, check that
1192 * it was generated with the same generator.
1194 EC_POINT *pre_comp_generator = EC_POINT_new(group);
1195 if (pre_comp_generator == NULL)
1198 if (!ecp_nistz256_set_from_affine(pre_comp_generator,
1199 group, pre_comp->precomp[0],
1201 EC_POINT_free(pre_comp_generator);
1205 if (0 == EC_POINT_cmp(group, generator, pre_comp_generator, ctx))
1206 preComputedTable = (const PRECOMP256_ROW *)pre_comp->precomp;
1208 EC_POINT_free(pre_comp_generator);
1211 if (preComputedTable == NULL && ecp_nistz256_is_affine_G(generator)) {
1213 * If there is no precomputed data, but the generator is the
1214 * default, a hardcoded table of precomputed data is used. This
1215 * is because applications, such as Apache, do not use
1216 * EC_KEY_precompute_mult.
1218 preComputedTable = ecp_nistz256_precomputed;
1221 if (preComputedTable) {
1222 if ((BN_num_bits(scalar) > 256)
1223 || BN_is_negative(scalar)) {
1224 if ((tmp_scalar = BN_CTX_get(ctx)) == NULL)
1227 if (!BN_nnmod(tmp_scalar, scalar, group->order, ctx)) {
1228 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_BN_LIB);
1231 scalar = tmp_scalar;
1234 for (i = 0; i < bn_get_top(scalar) * BN_BYTES; i += BN_BYTES) {
1235 BN_ULONG d = bn_get_words(scalar)[i / BN_BYTES];
1237 p_str[i + 0] = (unsigned char)d;
1238 p_str[i + 1] = (unsigned char)(d >> 8);
1239 p_str[i + 2] = (unsigned char)(d >> 16);
1240 p_str[i + 3] = (unsigned char)(d >>= 24);
1241 if (BN_BYTES == 8) {
1243 p_str[i + 4] = (unsigned char)d;
1244 p_str[i + 5] = (unsigned char)(d >> 8);
1245 p_str[i + 6] = (unsigned char)(d >> 16);
1246 p_str[i + 7] = (unsigned char)(d >> 24);
1253 #if defined(ECP_NISTZ256_AVX2)
1254 if (ecp_nistz_avx2_eligible()) {
1255 ecp_nistz256_avx2_mul_g(&p.p, p_str, preComputedTable);
1260 wvalue = (p_str[0] << 1) & mask;
1263 wvalue = _booth_recode_w7(wvalue);
1265 ecp_nistz256_gather_w7(&p.a, preComputedTable[0],
1268 ecp_nistz256_neg(p.p.Z, p.p.Y);
1269 copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
1271 memcpy(p.p.Z, ONE, sizeof(ONE));
1273 for (i = 1; i < 37; i++) {
1274 unsigned int off = (idx - 1) / 8;
1275 wvalue = p_str[off] | p_str[off + 1] << 8;
1276 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1279 wvalue = _booth_recode_w7(wvalue);
1281 ecp_nistz256_gather_w7(&t.a,
1282 preComputedTable[i], wvalue >> 1);
1284 ecp_nistz256_neg(t.p.Z, t.a.Y);
1285 copy_conditional(t.a.Y, t.p.Z, wvalue & 1);
1287 ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a);
1292 no_precomp_for_generator = 1;
1297 if (no_precomp_for_generator) {
1299 * Without a precomputed table for the generator, it has to be
1300 * handled like a normal point.
1302 new_scalars = OPENSSL_malloc((num + 1) * sizeof(BIGNUM *));
1303 if (new_scalars == NULL) {
1304 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1308 new_points = OPENSSL_malloc((num + 1) * sizeof(EC_POINT *));
1309 if (new_points == NULL) {
1310 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1314 memcpy(new_scalars, scalars, num * sizeof(BIGNUM *));
1315 new_scalars[num] = scalar;
1316 memcpy(new_points, points, num * sizeof(EC_POINT *));
1317 new_points[num] = generator;
1319 scalars = new_scalars;
1320 points = new_points;
1325 P256_POINT *out = &t.p;
1329 if (!ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx))
1333 ecp_nistz256_point_add(&p.p, &p.p, out);
1336 /* Not constant-time, but we're only operating on the public output. */
1337 if (!bn_set_words(r->X, p.p.X, P256_LIMBS) ||
1338 !bn_set_words(r->Y, p.p.Y, P256_LIMBS) ||
1339 !bn_set_words(r->Z, p.p.Z, P256_LIMBS)) {
1342 r->Z_is_one = is_one(r->Z) & 1;
1349 BN_CTX_free(new_ctx);
1350 OPENSSL_free(new_points);
1351 OPENSSL_free(new_scalars);
1355 __owur static int ecp_nistz256_get_affine(const EC_GROUP *group,
1356 const EC_POINT *point,
1357 BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
1359 BN_ULONG z_inv2[P256_LIMBS];
1360 BN_ULONG z_inv3[P256_LIMBS];
1361 BN_ULONG x_aff[P256_LIMBS];
1362 BN_ULONG y_aff[P256_LIMBS];
1363 BN_ULONG point_x[P256_LIMBS], point_y[P256_LIMBS], point_z[P256_LIMBS];
1364 BN_ULONG x_ret[P256_LIMBS], y_ret[P256_LIMBS];
1366 if (EC_POINT_is_at_infinity(group, point)) {
1367 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_POINT_AT_INFINITY);
1371 if (!ecp_nistz256_bignum_to_field_elem(point_x, point->X) ||
1372 !ecp_nistz256_bignum_to_field_elem(point_y, point->Y) ||
1373 !ecp_nistz256_bignum_to_field_elem(point_z, point->Z)) {
1374 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_COORDINATES_OUT_OF_RANGE);
1378 ecp_nistz256_mod_inverse(z_inv3, point_z);
1379 ecp_nistz256_sqr_mont(z_inv2, z_inv3);
1380 ecp_nistz256_mul_mont(x_aff, z_inv2, point_x);
1383 ecp_nistz256_from_mont(x_ret, x_aff);
1384 if (!bn_set_words(x, x_ret, P256_LIMBS))
1389 ecp_nistz256_mul_mont(z_inv3, z_inv3, z_inv2);
1390 ecp_nistz256_mul_mont(y_aff, z_inv3, point_y);
1391 ecp_nistz256_from_mont(y_ret, y_aff);
1392 if (!bn_set_words(y, y_ret, P256_LIMBS))
1399 static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group)
1401 NISTZ256_PRE_COMP *ret = NULL;
1406 ret = OPENSSL_zalloc(sizeof(*ret));
1409 ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1414 ret->w = 6; /* default */
1415 ret->references = 1;
1417 ret->lock = CRYPTO_THREAD_lock_new();
1418 if (ret->lock == NULL) {
1419 ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1426 NISTZ256_PRE_COMP *EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP *p)
1430 CRYPTO_atomic_add(&p->references, 1, &i, p->lock);
1434 void EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP *pre)
1441 CRYPTO_atomic_add(&pre->references, -1, &i, pre->lock);
1442 REF_PRINT_COUNT("EC_nistz256", x);
1445 REF_ASSERT_ISNT(i < 0);
1447 OPENSSL_free(pre->precomp_storage);
1448 CRYPTO_THREAD_lock_free(pre->lock);
1453 static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP *group)
1455 /* There is a hard-coded table for the default generator. */
1456 const EC_POINT *generator = EC_GROUP_get0_generator(group);
1458 if (generator != NULL && ecp_nistz256_is_affine_G(generator)) {
1459 /* There is a hard-coded table for the default generator. */
1463 return HAVEPRECOMP(group, nistz256);
1466 const EC_METHOD *EC_GFp_nistz256_method(void)
1468 static const EC_METHOD ret = {
1469 EC_FLAGS_DEFAULT_OCT,
1470 NID_X9_62_prime_field,
1471 ec_GFp_mont_group_init,
1472 ec_GFp_mont_group_finish,
1473 ec_GFp_mont_group_clear_finish,
1474 ec_GFp_mont_group_copy,
1475 ec_GFp_mont_group_set_curve,
1476 ec_GFp_simple_group_get_curve,
1477 ec_GFp_simple_group_get_degree,
1478 ec_group_simple_order_bits,
1479 ec_GFp_simple_group_check_discriminant,
1480 ec_GFp_simple_point_init,
1481 ec_GFp_simple_point_finish,
1482 ec_GFp_simple_point_clear_finish,
1483 ec_GFp_simple_point_copy,
1484 ec_GFp_simple_point_set_to_infinity,
1485 ec_GFp_simple_set_Jprojective_coordinates_GFp,
1486 ec_GFp_simple_get_Jprojective_coordinates_GFp,
1487 ec_GFp_simple_point_set_affine_coordinates,
1488 ecp_nistz256_get_affine,
1492 ec_GFp_simple_invert,
1493 ec_GFp_simple_is_at_infinity,
1494 ec_GFp_simple_is_on_curve,
1496 ec_GFp_simple_make_affine,
1497 ec_GFp_simple_points_make_affine,
1498 ecp_nistz256_points_mul, /* mul */
1499 ecp_nistz256_mult_precompute, /* precompute_mult */
1500 ecp_nistz256_window_have_precompute_mult, /* have_precompute_mult */
1501 ec_GFp_mont_field_mul,
1502 ec_GFp_mont_field_sqr,
1504 ec_GFp_mont_field_encode,
1505 ec_GFp_mont_field_decode,
1506 ec_GFp_mont_field_set_to_one,
1507 ec_key_simple_priv2oct,
1508 ec_key_simple_oct2priv,
1509 0, /* set private */
1510 ec_key_simple_generate_key,
1511 ec_key_simple_check_key,
1512 ec_key_simple_generate_public_key,
1515 ecdh_simple_compute_key