2 * Copyright 2011-2018 The OpenSSL Project Authors. All Rights Reserved.
4 * Licensed under the Apache License 2.0 (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 /* Copyright 2011 Google Inc.
12 * Licensed under the Apache License, Version 2.0 (the "License");
14 * you may not use this file except in compliance with the License.
15 * You may obtain a copy of the License at
17 * http://www.apache.org/licenses/LICENSE-2.0
19 * Unless required by applicable law or agreed to in writing, software
20 * distributed under the License is distributed on an "AS IS" BASIS,
21 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
22 * See the License for the specific language governing permissions and
23 * limitations under the License.
27 * A 64-bit implementation of the NIST P-256 elliptic curve point multiplication
29 * OpenSSL integration was taken from Emilia Kasper's work in ecp_nistp224.c.
30 * Otherwise based on Emilia's P224 work, which was inspired by my curve25519
31 * work which got its smarts from Daniel J. Bernstein's work on the same.
34 #include <openssl/opensslconf.h>
35 #ifdef OPENSSL_NO_EC_NISTP_64_GCC_128
36 NON_EMPTY_TRANSLATION_UNIT
41 # include <openssl/err.h>
42 # include "ec_local.h"
44 # if defined(__SIZEOF_INT128__) && __SIZEOF_INT128__==16
45 /* even with gcc, the typedef won't work for 32-bit platforms */
46 typedef __uint128_t uint128_t; /* nonstandard; implemented by gcc on 64-bit
48 typedef __int128_t int128_t;
50 # error "Your compiler doesn't appear to support 128-bit integer types"
58 * The underlying field. P256 operates over GF(2^256-2^224+2^192+2^96-1). We
59 * can serialise an element of this field into 32 bytes. We call this an
63 typedef u8 felem_bytearray[32];
66 * These are the parameters of P256, taken from FIPS 186-3, page 86. These
67 * values are big-endian.
69 static const felem_bytearray nistp256_curve_params[5] = {
70 {0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x01, /* p */
71 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
72 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff,
73 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
74 {0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x01, /* a = -3 */
75 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
76 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff,
77 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xfc}, /* b */
78 {0x5a, 0xc6, 0x35, 0xd8, 0xaa, 0x3a, 0x93, 0xe7,
79 0xb3, 0xeb, 0xbd, 0x55, 0x76, 0x98, 0x86, 0xbc,
80 0x65, 0x1d, 0x06, 0xb0, 0xcc, 0x53, 0xb0, 0xf6,
81 0x3b, 0xce, 0x3c, 0x3e, 0x27, 0xd2, 0x60, 0x4b},
82 {0x6b, 0x17, 0xd1, 0xf2, 0xe1, 0x2c, 0x42, 0x47, /* x */
83 0xf8, 0xbc, 0xe6, 0xe5, 0x63, 0xa4, 0x40, 0xf2,
84 0x77, 0x03, 0x7d, 0x81, 0x2d, 0xeb, 0x33, 0xa0,
85 0xf4, 0xa1, 0x39, 0x45, 0xd8, 0x98, 0xc2, 0x96},
86 {0x4f, 0xe3, 0x42, 0xe2, 0xfe, 0x1a, 0x7f, 0x9b, /* y */
87 0x8e, 0xe7, 0xeb, 0x4a, 0x7c, 0x0f, 0x9e, 0x16,
88 0x2b, 0xce, 0x33, 0x57, 0x6b, 0x31, 0x5e, 0xce,
89 0xcb, 0xb6, 0x40, 0x68, 0x37, 0xbf, 0x51, 0xf5}
93 * The representation of field elements.
94 * ------------------------------------
96 * We represent field elements with either four 128-bit values, eight 128-bit
97 * values, or four 64-bit values. The field element represented is:
98 * v[0]*2^0 + v[1]*2^64 + v[2]*2^128 + v[3]*2^192 (mod p)
100 * v[0]*2^0 + v[1]*2^64 + v[2]*2^128 + ... + v[8]*2^512 (mod p)
102 * 128-bit values are called 'limbs'. Since the limbs are spaced only 64 bits
103 * apart, but are 128-bits wide, the most significant bits of each limb overlap
104 * with the least significant bits of the next.
106 * A field element with four limbs is an 'felem'. One with eight limbs is a
109 * A field element with four, 64-bit values is called a 'smallfelem'. Small
110 * values are used as intermediate values before multiplication.
115 typedef uint128_t limb;
116 typedef limb felem[NLIMBS];
117 typedef limb longfelem[NLIMBS * 2];
118 typedef u64 smallfelem[NLIMBS];
120 /* This is the value of the prime as four 64-bit words, little-endian. */
121 static const u64 kPrime[4] =
122 { 0xfffffffffffffffful, 0xffffffff, 0, 0xffffffff00000001ul };
123 static const u64 bottom63bits = 0x7ffffffffffffffful;
126 * bin32_to_felem takes a little-endian byte array and converts it into felem
127 * form. This assumes that the CPU is little-endian.
129 static void bin32_to_felem(felem out, const u8 in[32])
131 out[0] = *((u64 *)&in[0]);
132 out[1] = *((u64 *)&in[8]);
133 out[2] = *((u64 *)&in[16]);
134 out[3] = *((u64 *)&in[24]);
138 * smallfelem_to_bin32 takes a smallfelem and serialises into a little
139 * endian, 32 byte array. This assumes that the CPU is little-endian.
141 static void smallfelem_to_bin32(u8 out[32], const smallfelem in)
143 *((u64 *)&out[0]) = in[0];
144 *((u64 *)&out[8]) = in[1];
145 *((u64 *)&out[16]) = in[2];
146 *((u64 *)&out[24]) = in[3];
149 /* BN_to_felem converts an OpenSSL BIGNUM into an felem */
150 static int BN_to_felem(felem out, const BIGNUM *bn)
152 felem_bytearray b_out;
155 if (BN_is_negative(bn)) {
156 ECerr(EC_F_BN_TO_FELEM, EC_R_BIGNUM_OUT_OF_RANGE);
159 num_bytes = BN_bn2lebinpad(bn, b_out, sizeof(b_out));
161 ECerr(EC_F_BN_TO_FELEM, EC_R_BIGNUM_OUT_OF_RANGE);
164 bin32_to_felem(out, b_out);
168 /* felem_to_BN converts an felem into an OpenSSL BIGNUM */
169 static BIGNUM *smallfelem_to_BN(BIGNUM *out, const smallfelem in)
171 felem_bytearray b_out;
172 smallfelem_to_bin32(b_out, in);
173 return BN_lebin2bn(b_out, sizeof(b_out), out);
181 static void smallfelem_one(smallfelem out)
189 static void smallfelem_assign(smallfelem out, const smallfelem in)
197 static void felem_assign(felem out, const felem in)
205 /* felem_sum sets out = out + in. */
206 static void felem_sum(felem out, const felem in)
214 /* felem_small_sum sets out = out + in. */
215 static void felem_small_sum(felem out, const smallfelem in)
223 /* felem_scalar sets out = out * scalar */
224 static void felem_scalar(felem out, const u64 scalar)
232 /* longfelem_scalar sets out = out * scalar */
233 static void longfelem_scalar(longfelem out, const u64 scalar)
245 # define two105m41m9 (((limb)1) << 105) - (((limb)1) << 41) - (((limb)1) << 9)
246 # define two105 (((limb)1) << 105)
247 # define two105m41p9 (((limb)1) << 105) - (((limb)1) << 41) + (((limb)1) << 9)
249 /* zero105 is 0 mod p */
250 static const felem zero105 =
251 { two105m41m9, two105, two105m41p9, two105m41p9 };
254 * smallfelem_neg sets |out| to |-small|
256 * out[i] < out[i] + 2^105
258 static void smallfelem_neg(felem out, const smallfelem small)
260 /* In order to prevent underflow, we subtract from 0 mod p. */
261 out[0] = zero105[0] - small[0];
262 out[1] = zero105[1] - small[1];
263 out[2] = zero105[2] - small[2];
264 out[3] = zero105[3] - small[3];
268 * felem_diff subtracts |in| from |out|
272 * out[i] < out[i] + 2^105
274 static void felem_diff(felem out, const felem in)
277 * In order to prevent underflow, we add 0 mod p before subtracting.
279 out[0] += zero105[0];
280 out[1] += zero105[1];
281 out[2] += zero105[2];
282 out[3] += zero105[3];
290 # define two107m43m11 (((limb)1) << 107) - (((limb)1) << 43) - (((limb)1) << 11)
291 # define two107 (((limb)1) << 107)
292 # define two107m43p11 (((limb)1) << 107) - (((limb)1) << 43) + (((limb)1) << 11)
294 /* zero107 is 0 mod p */
295 static const felem zero107 =
296 { two107m43m11, two107, two107m43p11, two107m43p11 };
299 * An alternative felem_diff for larger inputs |in|
300 * felem_diff_zero107 subtracts |in| from |out|
304 * out[i] < out[i] + 2^107
306 static void felem_diff_zero107(felem out, const felem in)
309 * In order to prevent underflow, we add 0 mod p before subtracting.
311 out[0] += zero107[0];
312 out[1] += zero107[1];
313 out[2] += zero107[2];
314 out[3] += zero107[3];
323 * longfelem_diff subtracts |in| from |out|
327 * out[i] < out[i] + 2^70 + 2^40
329 static void longfelem_diff(longfelem out, const longfelem in)
331 static const limb two70m8p6 =
332 (((limb) 1) << 70) - (((limb) 1) << 8) + (((limb) 1) << 6);
333 static const limb two70p40 = (((limb) 1) << 70) + (((limb) 1) << 40);
334 static const limb two70 = (((limb) 1) << 70);
335 static const limb two70m40m38p6 =
336 (((limb) 1) << 70) - (((limb) 1) << 40) - (((limb) 1) << 38) +
338 static const limb two70m6 = (((limb) 1) << 70) - (((limb) 1) << 6);
340 /* add 0 mod p to avoid underflow */
344 out[3] += two70m40m38p6;
350 /* in[i] < 7*2^67 < 2^70 - 2^40 - 2^38 + 2^6 */
361 # define two64m0 (((limb)1) << 64) - 1
362 # define two110p32m0 (((limb)1) << 110) + (((limb)1) << 32) - 1
363 # define two64m46 (((limb)1) << 64) - (((limb)1) << 46)
364 # define two64m32 (((limb)1) << 64) - (((limb)1) << 32)
366 /* zero110 is 0 mod p */
367 static const felem zero110 = { two64m0, two110p32m0, two64m46, two64m32 };
370 * felem_shrink converts an felem into a smallfelem. The result isn't quite
371 * minimal as the value may be greater than p.
378 static void felem_shrink(smallfelem out, const felem in)
383 static const u64 kPrime3Test = 0x7fffffff00000001ul; /* 2^63 - 2^32 + 1 */
386 tmp[3] = zero110[3] + in[3] + ((u64)(in[2] >> 64));
389 tmp[2] = zero110[2] + (u64)in[2];
390 tmp[0] = zero110[0] + in[0];
391 tmp[1] = zero110[1] + in[1];
392 /* tmp[0] < 2**110, tmp[1] < 2^111, tmp[2] < 2**65 */
395 * We perform two partial reductions where we eliminate the high-word of
396 * tmp[3]. We don't update the other words till the end.
398 a = tmp[3] >> 64; /* a < 2^46 */
399 tmp[3] = (u64)tmp[3];
401 tmp[3] += ((limb) a) << 32;
405 a = tmp[3] >> 64; /* a < 2^15 */
406 b += a; /* b < 2^46 + 2^15 < 2^47 */
407 tmp[3] = (u64)tmp[3];
409 tmp[3] += ((limb) a) << 32;
410 /* tmp[3] < 2^64 + 2^47 */
413 * This adjusts the other two words to complete the two partial
417 tmp[1] -= (((limb) b) << 32);
420 * In order to make space in tmp[3] for the carry from 2 -> 3, we
421 * conditionally subtract kPrime if tmp[3] is large enough.
423 high = (u64)(tmp[3] >> 64);
424 /* As tmp[3] < 2^65, high is either 1 or 0 */
428 * all ones if the high word of tmp[3] is 1
429 * all zeros if the high word of tmp[3] if 0
432 mask = 0 - (low >> 63);
435 * all ones if the MSB of low is 1
436 * all zeros if the MSB of low if 0
440 /* if low was greater than kPrime3Test then the MSB is zero */
442 low = 0 - (low >> 63);
445 * all ones if low was > kPrime3Test
446 * all zeros if low was <= kPrime3Test
448 mask = (mask & low) | high;
449 tmp[0] -= mask & kPrime[0];
450 tmp[1] -= mask & kPrime[1];
451 /* kPrime[2] is zero, so omitted */
452 tmp[3] -= mask & kPrime[3];
453 /* tmp[3] < 2**64 - 2**32 + 1 */
455 tmp[1] += ((u64)(tmp[0] >> 64));
456 tmp[0] = (u64)tmp[0];
457 tmp[2] += ((u64)(tmp[1] >> 64));
458 tmp[1] = (u64)tmp[1];
459 tmp[3] += ((u64)(tmp[2] >> 64));
460 tmp[2] = (u64)tmp[2];
469 /* smallfelem_expand converts a smallfelem to an felem */
470 static void smallfelem_expand(felem out, const smallfelem in)
479 * smallfelem_square sets |out| = |small|^2
483 * out[i] < 7 * 2^64 < 2^67
485 static void smallfelem_square(longfelem out, const smallfelem small)
490 a = ((uint128_t) small[0]) * small[0];
496 a = ((uint128_t) small[0]) * small[1];
503 a = ((uint128_t) small[0]) * small[2];
510 a = ((uint128_t) small[0]) * small[3];
516 a = ((uint128_t) small[1]) * small[2];
523 a = ((uint128_t) small[1]) * small[1];
529 a = ((uint128_t) small[1]) * small[3];
536 a = ((uint128_t) small[2]) * small[3];
544 a = ((uint128_t) small[2]) * small[2];
550 a = ((uint128_t) small[3]) * small[3];
558 * felem_square sets |out| = |in|^2
562 * out[i] < 7 * 2^64 < 2^67
564 static void felem_square(longfelem out, const felem in)
567 felem_shrink(small, in);
568 smallfelem_square(out, small);
572 * smallfelem_mul sets |out| = |small1| * |small2|
577 * out[i] < 7 * 2^64 < 2^67
579 static void smallfelem_mul(longfelem out, const smallfelem small1,
580 const smallfelem small2)
585 a = ((uint128_t) small1[0]) * small2[0];
591 a = ((uint128_t) small1[0]) * small2[1];
597 a = ((uint128_t) small1[1]) * small2[0];
603 a = ((uint128_t) small1[0]) * small2[2];
609 a = ((uint128_t) small1[1]) * small2[1];
615 a = ((uint128_t) small1[2]) * small2[0];
621 a = ((uint128_t) small1[0]) * small2[3];
627 a = ((uint128_t) small1[1]) * small2[2];
633 a = ((uint128_t) small1[2]) * small2[1];
639 a = ((uint128_t) small1[3]) * small2[0];
645 a = ((uint128_t) small1[1]) * small2[3];
651 a = ((uint128_t) small1[2]) * small2[2];
657 a = ((uint128_t) small1[3]) * small2[1];
663 a = ((uint128_t) small1[2]) * small2[3];
669 a = ((uint128_t) small1[3]) * small2[2];
675 a = ((uint128_t) small1[3]) * small2[3];
683 * felem_mul sets |out| = |in1| * |in2|
688 * out[i] < 7 * 2^64 < 2^67
690 static void felem_mul(longfelem out, const felem in1, const felem in2)
692 smallfelem small1, small2;
693 felem_shrink(small1, in1);
694 felem_shrink(small2, in2);
695 smallfelem_mul(out, small1, small2);
699 * felem_small_mul sets |out| = |small1| * |in2|
704 * out[i] < 7 * 2^64 < 2^67
706 static void felem_small_mul(longfelem out, const smallfelem small1,
710 felem_shrink(small2, in2);
711 smallfelem_mul(out, small1, small2);
714 # define two100m36m4 (((limb)1) << 100) - (((limb)1) << 36) - (((limb)1) << 4)
715 # define two100 (((limb)1) << 100)
716 # define two100m36p4 (((limb)1) << 100) - (((limb)1) << 36) + (((limb)1) << 4)
717 /* zero100 is 0 mod p */
718 static const felem zero100 =
719 { two100m36m4, two100, two100m36p4, two100m36p4 };
722 * Internal function for the different flavours of felem_reduce.
723 * felem_reduce_ reduces the higher coefficients in[4]-in[7].
725 * out[0] >= in[6] + 2^32*in[6] + in[7] + 2^32*in[7]
726 * out[1] >= in[7] + 2^32*in[4]
727 * out[2] >= in[5] + 2^32*in[5]
728 * out[3] >= in[4] + 2^32*in[5] + 2^32*in[6]
730 * out[0] <= out[0] + in[4] + 2^32*in[5]
731 * out[1] <= out[1] + in[5] + 2^33*in[6]
732 * out[2] <= out[2] + in[7] + 2*in[6] + 2^33*in[7]
733 * out[3] <= out[3] + 2^32*in[4] + 3*in[7]
735 static void felem_reduce_(felem out, const longfelem in)
738 /* combine common terms from below */
739 c = in[4] + (in[5] << 32);
747 /* the remaining terms */
748 /* 256: [(0,1),(96,-1),(192,-1),(224,1)] */
749 out[1] -= (in[4] << 32);
750 out[3] += (in[4] << 32);
752 /* 320: [(32,1),(64,1),(128,-1),(160,-1),(224,-1)] */
753 out[2] -= (in[5] << 32);
755 /* 384: [(0,-1),(32,-1),(96,2),(128,2),(224,-1)] */
757 out[0] -= (in[6] << 32);
758 out[1] += (in[6] << 33);
759 out[2] += (in[6] * 2);
760 out[3] -= (in[6] << 32);
762 /* 448: [(0,-1),(32,-1),(64,-1),(128,1),(160,2),(192,3)] */
764 out[0] -= (in[7] << 32);
765 out[2] += (in[7] << 33);
766 out[3] += (in[7] * 3);
770 * felem_reduce converts a longfelem into an felem.
771 * To be called directly after felem_square or felem_mul.
773 * in[0] < 2^64, in[1] < 3*2^64, in[2] < 5*2^64, in[3] < 7*2^64
774 * in[4] < 7*2^64, in[5] < 5*2^64, in[6] < 3*2^64, in[7] < 2*64
778 static void felem_reduce(felem out, const longfelem in)
780 out[0] = zero100[0] + in[0];
781 out[1] = zero100[1] + in[1];
782 out[2] = zero100[2] + in[2];
783 out[3] = zero100[3] + in[3];
785 felem_reduce_(out, in);
788 * out[0] > 2^100 - 2^36 - 2^4 - 3*2^64 - 3*2^96 - 2^64 - 2^96 > 0
789 * out[1] > 2^100 - 2^64 - 7*2^96 > 0
790 * out[2] > 2^100 - 2^36 + 2^4 - 5*2^64 - 5*2^96 > 0
791 * out[3] > 2^100 - 2^36 + 2^4 - 7*2^64 - 5*2^96 - 3*2^96 > 0
793 * out[0] < 2^100 + 2^64 + 7*2^64 + 5*2^96 < 2^101
794 * out[1] < 2^100 + 3*2^64 + 5*2^64 + 3*2^97 < 2^101
795 * out[2] < 2^100 + 5*2^64 + 2^64 + 3*2^65 + 2^97 < 2^101
796 * out[3] < 2^100 + 7*2^64 + 7*2^96 + 3*2^64 < 2^101
801 * felem_reduce_zero105 converts a larger longfelem into an felem.
807 static void felem_reduce_zero105(felem out, const longfelem in)
809 out[0] = zero105[0] + in[0];
810 out[1] = zero105[1] + in[1];
811 out[2] = zero105[2] + in[2];
812 out[3] = zero105[3] + in[3];
814 felem_reduce_(out, in);
817 * out[0] > 2^105 - 2^41 - 2^9 - 2^71 - 2^103 - 2^71 - 2^103 > 0
818 * out[1] > 2^105 - 2^71 - 2^103 > 0
819 * out[2] > 2^105 - 2^41 + 2^9 - 2^71 - 2^103 > 0
820 * out[3] > 2^105 - 2^41 + 2^9 - 2^71 - 2^103 - 2^103 > 0
822 * out[0] < 2^105 + 2^71 + 2^71 + 2^103 < 2^106
823 * out[1] < 2^105 + 2^71 + 2^71 + 2^103 < 2^106
824 * out[2] < 2^105 + 2^71 + 2^71 + 2^71 + 2^103 < 2^106
825 * out[3] < 2^105 + 2^71 + 2^103 + 2^71 < 2^106
830 * subtract_u64 sets *result = *result - v and *carry to one if the
831 * subtraction underflowed.
833 static void subtract_u64(u64 *result, u64 *carry, u64 v)
835 uint128_t r = *result;
837 *carry = (r >> 64) & 1;
842 * felem_contract converts |in| to its unique, minimal representation. On
843 * entry: in[i] < 2^109
845 static void felem_contract(smallfelem out, const felem in)
848 u64 all_equal_so_far = 0, result = 0, carry;
850 felem_shrink(out, in);
851 /* small is minimal except that the value might be > p */
855 * We are doing a constant time test if out >= kPrime. We need to compare
856 * each u64, from most-significant to least significant. For each one, if
857 * all words so far have been equal (m is all ones) then a non-equal
858 * result is the answer. Otherwise we continue.
860 for (i = 3; i < 4; i--) {
862 uint128_t a = ((uint128_t) kPrime[i]) - out[i];
864 * if out[i] > kPrime[i] then a will underflow and the high 64-bits
867 result |= all_equal_so_far & ((u64)(a >> 64));
870 * if kPrime[i] == out[i] then |equal| will be all zeros and the
871 * decrement will make it all ones.
873 equal = kPrime[i] ^ out[i];
875 equal &= equal << 32;
876 equal &= equal << 16;
881 equal = 0 - (equal >> 63);
883 all_equal_so_far &= equal;
887 * if all_equal_so_far is still all ones then the two values are equal
888 * and so out >= kPrime is true.
890 result |= all_equal_so_far;
892 /* if out >= kPrime then we subtract kPrime. */
893 subtract_u64(&out[0], &carry, result & kPrime[0]);
894 subtract_u64(&out[1], &carry, carry);
895 subtract_u64(&out[2], &carry, carry);
896 subtract_u64(&out[3], &carry, carry);
898 subtract_u64(&out[1], &carry, result & kPrime[1]);
899 subtract_u64(&out[2], &carry, carry);
900 subtract_u64(&out[3], &carry, carry);
902 subtract_u64(&out[2], &carry, result & kPrime[2]);
903 subtract_u64(&out[3], &carry, carry);
905 subtract_u64(&out[3], &carry, result & kPrime[3]);
908 static void smallfelem_square_contract(smallfelem out, const smallfelem in)
913 smallfelem_square(longtmp, in);
914 felem_reduce(tmp, longtmp);
915 felem_contract(out, tmp);
918 static void smallfelem_mul_contract(smallfelem out, const smallfelem in1,
919 const smallfelem in2)
924 smallfelem_mul(longtmp, in1, in2);
925 felem_reduce(tmp, longtmp);
926 felem_contract(out, tmp);
930 * felem_is_zero returns a limb with all bits set if |in| == 0 (mod p) and 0
935 static limb smallfelem_is_zero(const smallfelem small)
940 u64 is_zero = small[0] | small[1] | small[2] | small[3];
942 is_zero &= is_zero << 32;
943 is_zero &= is_zero << 16;
944 is_zero &= is_zero << 8;
945 is_zero &= is_zero << 4;
946 is_zero &= is_zero << 2;
947 is_zero &= is_zero << 1;
948 is_zero = 0 - (is_zero >> 63);
950 is_p = (small[0] ^ kPrime[0]) |
951 (small[1] ^ kPrime[1]) |
952 (small[2] ^ kPrime[2]) | (small[3] ^ kPrime[3]);
960 is_p = 0 - (is_p >> 63);
965 result |= ((limb) is_zero) << 64;
969 static int smallfelem_is_zero_int(const void *small)
971 return (int)(smallfelem_is_zero(small) & ((limb) 1));
975 * felem_inv calculates |out| = |in|^{-1}
977 * Based on Fermat's Little Theorem:
979 * a^{p-1} = 1 (mod p)
980 * a^{p-2} = a^{-1} (mod p)
982 static void felem_inv(felem out, const felem in)
985 /* each e_I will hold |in|^{2^I - 1} */
986 felem e2, e4, e8, e16, e32, e64;
990 felem_square(tmp, in);
991 felem_reduce(ftmp, tmp); /* 2^1 */
992 felem_mul(tmp, in, ftmp);
993 felem_reduce(ftmp, tmp); /* 2^2 - 2^0 */
994 felem_assign(e2, ftmp);
995 felem_square(tmp, ftmp);
996 felem_reduce(ftmp, tmp); /* 2^3 - 2^1 */
997 felem_square(tmp, ftmp);
998 felem_reduce(ftmp, tmp); /* 2^4 - 2^2 */
999 felem_mul(tmp, ftmp, e2);
1000 felem_reduce(ftmp, tmp); /* 2^4 - 2^0 */
1001 felem_assign(e4, ftmp);
1002 felem_square(tmp, ftmp);
1003 felem_reduce(ftmp, tmp); /* 2^5 - 2^1 */
1004 felem_square(tmp, ftmp);
1005 felem_reduce(ftmp, tmp); /* 2^6 - 2^2 */
1006 felem_square(tmp, ftmp);
1007 felem_reduce(ftmp, tmp); /* 2^7 - 2^3 */
1008 felem_square(tmp, ftmp);
1009 felem_reduce(ftmp, tmp); /* 2^8 - 2^4 */
1010 felem_mul(tmp, ftmp, e4);
1011 felem_reduce(ftmp, tmp); /* 2^8 - 2^0 */
1012 felem_assign(e8, ftmp);
1013 for (i = 0; i < 8; i++) {
1014 felem_square(tmp, ftmp);
1015 felem_reduce(ftmp, tmp);
1017 felem_mul(tmp, ftmp, e8);
1018 felem_reduce(ftmp, tmp); /* 2^16 - 2^0 */
1019 felem_assign(e16, ftmp);
1020 for (i = 0; i < 16; i++) {
1021 felem_square(tmp, ftmp);
1022 felem_reduce(ftmp, tmp);
1024 felem_mul(tmp, ftmp, e16);
1025 felem_reduce(ftmp, tmp); /* 2^32 - 2^0 */
1026 felem_assign(e32, ftmp);
1027 for (i = 0; i < 32; i++) {
1028 felem_square(tmp, ftmp);
1029 felem_reduce(ftmp, tmp);
1031 felem_assign(e64, ftmp);
1032 felem_mul(tmp, ftmp, in);
1033 felem_reduce(ftmp, tmp); /* 2^64 - 2^32 + 2^0 */
1034 for (i = 0; i < 192; i++) {
1035 felem_square(tmp, ftmp);
1036 felem_reduce(ftmp, tmp);
1037 } /* 2^256 - 2^224 + 2^192 */
1039 felem_mul(tmp, e64, e32);
1040 felem_reduce(ftmp2, tmp); /* 2^64 - 2^0 */
1041 for (i = 0; i < 16; i++) {
1042 felem_square(tmp, ftmp2);
1043 felem_reduce(ftmp2, tmp);
1045 felem_mul(tmp, ftmp2, e16);
1046 felem_reduce(ftmp2, tmp); /* 2^80 - 2^0 */
1047 for (i = 0; i < 8; i++) {
1048 felem_square(tmp, ftmp2);
1049 felem_reduce(ftmp2, tmp);
1051 felem_mul(tmp, ftmp2, e8);
1052 felem_reduce(ftmp2, tmp); /* 2^88 - 2^0 */
1053 for (i = 0; i < 4; i++) {
1054 felem_square(tmp, ftmp2);
1055 felem_reduce(ftmp2, tmp);
1057 felem_mul(tmp, ftmp2, e4);
1058 felem_reduce(ftmp2, tmp); /* 2^92 - 2^0 */
1059 felem_square(tmp, ftmp2);
1060 felem_reduce(ftmp2, tmp); /* 2^93 - 2^1 */
1061 felem_square(tmp, ftmp2);
1062 felem_reduce(ftmp2, tmp); /* 2^94 - 2^2 */
1063 felem_mul(tmp, ftmp2, e2);
1064 felem_reduce(ftmp2, tmp); /* 2^94 - 2^0 */
1065 felem_square(tmp, ftmp2);
1066 felem_reduce(ftmp2, tmp); /* 2^95 - 2^1 */
1067 felem_square(tmp, ftmp2);
1068 felem_reduce(ftmp2, tmp); /* 2^96 - 2^2 */
1069 felem_mul(tmp, ftmp2, in);
1070 felem_reduce(ftmp2, tmp); /* 2^96 - 3 */
1072 felem_mul(tmp, ftmp2, ftmp);
1073 felem_reduce(out, tmp); /* 2^256 - 2^224 + 2^192 + 2^96 - 3 */
1076 static void smallfelem_inv_contract(smallfelem out, const smallfelem in)
1080 smallfelem_expand(tmp, in);
1081 felem_inv(tmp, tmp);
1082 felem_contract(out, tmp);
1089 * Building on top of the field operations we have the operations on the
1090 * elliptic curve group itself. Points on the curve are represented in Jacobian
1095 * point_double calculates 2*(x_in, y_in, z_in)
1097 * The method is taken from:
1098 * http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#doubling-dbl-2001-b
1100 * Outputs can equal corresponding inputs, i.e., x_out == x_in is allowed.
1101 * while x_out == y_in is not (maybe this works, but it's not tested).
1104 point_double(felem x_out, felem y_out, felem z_out,
1105 const felem x_in, const felem y_in, const felem z_in)
1107 longfelem tmp, tmp2;
1108 felem delta, gamma, beta, alpha, ftmp, ftmp2;
1109 smallfelem small1, small2;
1111 felem_assign(ftmp, x_in);
1112 /* ftmp[i] < 2^106 */
1113 felem_assign(ftmp2, x_in);
1114 /* ftmp2[i] < 2^106 */
1117 felem_square(tmp, z_in);
1118 felem_reduce(delta, tmp);
1119 /* delta[i] < 2^101 */
1122 felem_square(tmp, y_in);
1123 felem_reduce(gamma, tmp);
1124 /* gamma[i] < 2^101 */
1125 felem_shrink(small1, gamma);
1127 /* beta = x*gamma */
1128 felem_small_mul(tmp, small1, x_in);
1129 felem_reduce(beta, tmp);
1130 /* beta[i] < 2^101 */
1132 /* alpha = 3*(x-delta)*(x+delta) */
1133 felem_diff(ftmp, delta);
1134 /* ftmp[i] < 2^105 + 2^106 < 2^107 */
1135 felem_sum(ftmp2, delta);
1136 /* ftmp2[i] < 2^105 + 2^106 < 2^107 */
1137 felem_scalar(ftmp2, 3);
1138 /* ftmp2[i] < 3 * 2^107 < 2^109 */
1139 felem_mul(tmp, ftmp, ftmp2);
1140 felem_reduce(alpha, tmp);
1141 /* alpha[i] < 2^101 */
1142 felem_shrink(small2, alpha);
1144 /* x' = alpha^2 - 8*beta */
1145 smallfelem_square(tmp, small2);
1146 felem_reduce(x_out, tmp);
1147 felem_assign(ftmp, beta);
1148 felem_scalar(ftmp, 8);
1149 /* ftmp[i] < 8 * 2^101 = 2^104 */
1150 felem_diff(x_out, ftmp);
1151 /* x_out[i] < 2^105 + 2^101 < 2^106 */
1153 /* z' = (y + z)^2 - gamma - delta */
1154 felem_sum(delta, gamma);
1155 /* delta[i] < 2^101 + 2^101 = 2^102 */
1156 felem_assign(ftmp, y_in);
1157 felem_sum(ftmp, z_in);
1158 /* ftmp[i] < 2^106 + 2^106 = 2^107 */
1159 felem_square(tmp, ftmp);
1160 felem_reduce(z_out, tmp);
1161 felem_diff(z_out, delta);
1162 /* z_out[i] < 2^105 + 2^101 < 2^106 */
1164 /* y' = alpha*(4*beta - x') - 8*gamma^2 */
1165 felem_scalar(beta, 4);
1166 /* beta[i] < 4 * 2^101 = 2^103 */
1167 felem_diff_zero107(beta, x_out);
1168 /* beta[i] < 2^107 + 2^103 < 2^108 */
1169 felem_small_mul(tmp, small2, beta);
1170 /* tmp[i] < 7 * 2^64 < 2^67 */
1171 smallfelem_square(tmp2, small1);
1172 /* tmp2[i] < 7 * 2^64 */
1173 longfelem_scalar(tmp2, 8);
1174 /* tmp2[i] < 8 * 7 * 2^64 = 7 * 2^67 */
1175 longfelem_diff(tmp, tmp2);
1176 /* tmp[i] < 2^67 + 2^70 + 2^40 < 2^71 */
1177 felem_reduce_zero105(y_out, tmp);
1178 /* y_out[i] < 2^106 */
1182 * point_double_small is the same as point_double, except that it operates on
1186 point_double_small(smallfelem x_out, smallfelem y_out, smallfelem z_out,
1187 const smallfelem x_in, const smallfelem y_in,
1188 const smallfelem z_in)
1190 felem felem_x_out, felem_y_out, felem_z_out;
1191 felem felem_x_in, felem_y_in, felem_z_in;
1193 smallfelem_expand(felem_x_in, x_in);
1194 smallfelem_expand(felem_y_in, y_in);
1195 smallfelem_expand(felem_z_in, z_in);
1196 point_double(felem_x_out, felem_y_out, felem_z_out,
1197 felem_x_in, felem_y_in, felem_z_in);
1198 felem_shrink(x_out, felem_x_out);
1199 felem_shrink(y_out, felem_y_out);
1200 felem_shrink(z_out, felem_z_out);
1203 /* copy_conditional copies in to out iff mask is all ones. */
1204 static void copy_conditional(felem out, const felem in, limb mask)
1207 for (i = 0; i < NLIMBS; ++i) {
1208 const limb tmp = mask & (in[i] ^ out[i]);
1213 /* copy_small_conditional copies in to out iff mask is all ones. */
1214 static void copy_small_conditional(felem out, const smallfelem in, limb mask)
1217 const u64 mask64 = mask;
1218 for (i = 0; i < NLIMBS; ++i) {
1219 out[i] = ((limb) (in[i] & mask64)) | (out[i] & ~mask);
1224 * point_add calculates (x1, y1, z1) + (x2, y2, z2)
1226 * The method is taken from:
1227 * http://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#addition-add-2007-bl,
1228 * adapted for mixed addition (z2 = 1, or z2 = 0 for the point at infinity).
1230 * This function includes a branch for checking whether the two input points
1231 * are equal, (while not equal to the point at infinity). This case never
1232 * happens during single point multiplication, so there is no timing leak for
1233 * ECDH or ECDSA signing.
1235 static void point_add(felem x3, felem y3, felem z3,
1236 const felem x1, const felem y1, const felem z1,
1237 const int mixed, const smallfelem x2,
1238 const smallfelem y2, const smallfelem z2)
1240 felem ftmp, ftmp2, ftmp3, ftmp4, ftmp5, ftmp6, x_out, y_out, z_out;
1241 longfelem tmp, tmp2;
1242 smallfelem small1, small2, small3, small4, small5;
1243 limb x_equal, y_equal, z1_is_zero, z2_is_zero;
1245 felem_shrink(small3, z1);
1247 z1_is_zero = smallfelem_is_zero(small3);
1248 z2_is_zero = smallfelem_is_zero(z2);
1250 /* ftmp = z1z1 = z1**2 */
1251 smallfelem_square(tmp, small3);
1252 felem_reduce(ftmp, tmp);
1253 /* ftmp[i] < 2^101 */
1254 felem_shrink(small1, ftmp);
1257 /* ftmp2 = z2z2 = z2**2 */
1258 smallfelem_square(tmp, z2);
1259 felem_reduce(ftmp2, tmp);
1260 /* ftmp2[i] < 2^101 */
1261 felem_shrink(small2, ftmp2);
1263 felem_shrink(small5, x1);
1265 /* u1 = ftmp3 = x1*z2z2 */
1266 smallfelem_mul(tmp, small5, small2);
1267 felem_reduce(ftmp3, tmp);
1268 /* ftmp3[i] < 2^101 */
1270 /* ftmp5 = z1 + z2 */
1271 felem_assign(ftmp5, z1);
1272 felem_small_sum(ftmp5, z2);
1273 /* ftmp5[i] < 2^107 */
1275 /* ftmp5 = (z1 + z2)**2 - (z1z1 + z2z2) = 2z1z2 */
1276 felem_square(tmp, ftmp5);
1277 felem_reduce(ftmp5, tmp);
1278 /* ftmp2 = z2z2 + z1z1 */
1279 felem_sum(ftmp2, ftmp);
1280 /* ftmp2[i] < 2^101 + 2^101 = 2^102 */
1281 felem_diff(ftmp5, ftmp2);
1282 /* ftmp5[i] < 2^105 + 2^101 < 2^106 */
1284 /* ftmp2 = z2 * z2z2 */
1285 smallfelem_mul(tmp, small2, z2);
1286 felem_reduce(ftmp2, tmp);
1288 /* s1 = ftmp2 = y1 * z2**3 */
1289 felem_mul(tmp, y1, ftmp2);
1290 felem_reduce(ftmp6, tmp);
1291 /* ftmp6[i] < 2^101 */
1294 * We'll assume z2 = 1 (special case z2 = 0 is handled later)
1297 /* u1 = ftmp3 = x1*z2z2 */
1298 felem_assign(ftmp3, x1);
1299 /* ftmp3[i] < 2^106 */
1302 felem_assign(ftmp5, z1);
1303 felem_scalar(ftmp5, 2);
1304 /* ftmp5[i] < 2*2^106 = 2^107 */
1306 /* s1 = ftmp2 = y1 * z2**3 */
1307 felem_assign(ftmp6, y1);
1308 /* ftmp6[i] < 2^106 */
1312 smallfelem_mul(tmp, x2, small1);
1313 felem_reduce(ftmp4, tmp);
1315 /* h = ftmp4 = u2 - u1 */
1316 felem_diff_zero107(ftmp4, ftmp3);
1317 /* ftmp4[i] < 2^107 + 2^101 < 2^108 */
1318 felem_shrink(small4, ftmp4);
1320 x_equal = smallfelem_is_zero(small4);
1322 /* z_out = ftmp5 * h */
1323 felem_small_mul(tmp, small4, ftmp5);
1324 felem_reduce(z_out, tmp);
1325 /* z_out[i] < 2^101 */
1327 /* ftmp = z1 * z1z1 */
1328 smallfelem_mul(tmp, small1, small3);
1329 felem_reduce(ftmp, tmp);
1331 /* s2 = tmp = y2 * z1**3 */
1332 felem_small_mul(tmp, y2, ftmp);
1333 felem_reduce(ftmp5, tmp);
1335 /* r = ftmp5 = (s2 - s1)*2 */
1336 felem_diff_zero107(ftmp5, ftmp6);
1337 /* ftmp5[i] < 2^107 + 2^107 = 2^108 */
1338 felem_scalar(ftmp5, 2);
1339 /* ftmp5[i] < 2^109 */
1340 felem_shrink(small1, ftmp5);
1341 y_equal = smallfelem_is_zero(small1);
1343 if (x_equal && y_equal && !z1_is_zero && !z2_is_zero) {
1344 point_double(x3, y3, z3, x1, y1, z1);
1348 /* I = ftmp = (2h)**2 */
1349 felem_assign(ftmp, ftmp4);
1350 felem_scalar(ftmp, 2);
1351 /* ftmp[i] < 2*2^108 = 2^109 */
1352 felem_square(tmp, ftmp);
1353 felem_reduce(ftmp, tmp);
1355 /* J = ftmp2 = h * I */
1356 felem_mul(tmp, ftmp4, ftmp);
1357 felem_reduce(ftmp2, tmp);
1359 /* V = ftmp4 = U1 * I */
1360 felem_mul(tmp, ftmp3, ftmp);
1361 felem_reduce(ftmp4, tmp);
1363 /* x_out = r**2 - J - 2V */
1364 smallfelem_square(tmp, small1);
1365 felem_reduce(x_out, tmp);
1366 felem_assign(ftmp3, ftmp4);
1367 felem_scalar(ftmp4, 2);
1368 felem_sum(ftmp4, ftmp2);
1369 /* ftmp4[i] < 2*2^101 + 2^101 < 2^103 */
1370 felem_diff(x_out, ftmp4);
1371 /* x_out[i] < 2^105 + 2^101 */
1373 /* y_out = r(V-x_out) - 2 * s1 * J */
1374 felem_diff_zero107(ftmp3, x_out);
1375 /* ftmp3[i] < 2^107 + 2^101 < 2^108 */
1376 felem_small_mul(tmp, small1, ftmp3);
1377 felem_mul(tmp2, ftmp6, ftmp2);
1378 longfelem_scalar(tmp2, 2);
1379 /* tmp2[i] < 2*2^67 = 2^68 */
1380 longfelem_diff(tmp, tmp2);
1381 /* tmp[i] < 2^67 + 2^70 + 2^40 < 2^71 */
1382 felem_reduce_zero105(y_out, tmp);
1383 /* y_out[i] < 2^106 */
1385 copy_small_conditional(x_out, x2, z1_is_zero);
1386 copy_conditional(x_out, x1, z2_is_zero);
1387 copy_small_conditional(y_out, y2, z1_is_zero);
1388 copy_conditional(y_out, y1, z2_is_zero);
1389 copy_small_conditional(z_out, z2, z1_is_zero);
1390 copy_conditional(z_out, z1, z2_is_zero);
1391 felem_assign(x3, x_out);
1392 felem_assign(y3, y_out);
1393 felem_assign(z3, z_out);
1397 * point_add_small is the same as point_add, except that it operates on
1400 static void point_add_small(smallfelem x3, smallfelem y3, smallfelem z3,
1401 smallfelem x1, smallfelem y1, smallfelem z1,
1402 smallfelem x2, smallfelem y2, smallfelem z2)
1404 felem felem_x3, felem_y3, felem_z3;
1405 felem felem_x1, felem_y1, felem_z1;
1406 smallfelem_expand(felem_x1, x1);
1407 smallfelem_expand(felem_y1, y1);
1408 smallfelem_expand(felem_z1, z1);
1409 point_add(felem_x3, felem_y3, felem_z3, felem_x1, felem_y1, felem_z1, 0,
1411 felem_shrink(x3, felem_x3);
1412 felem_shrink(y3, felem_y3);
1413 felem_shrink(z3, felem_z3);
1417 * Base point pre computation
1418 * --------------------------
1420 * Two different sorts of precomputed tables are used in the following code.
1421 * Each contain various points on the curve, where each point is three field
1422 * elements (x, y, z).
1424 * For the base point table, z is usually 1 (0 for the point at infinity).
1425 * This table has 2 * 16 elements, starting with the following:
1426 * index | bits | point
1427 * ------+---------+------------------------------
1430 * 2 | 0 0 1 0 | 2^64G
1431 * 3 | 0 0 1 1 | (2^64 + 1)G
1432 * 4 | 0 1 0 0 | 2^128G
1433 * 5 | 0 1 0 1 | (2^128 + 1)G
1434 * 6 | 0 1 1 0 | (2^128 + 2^64)G
1435 * 7 | 0 1 1 1 | (2^128 + 2^64 + 1)G
1436 * 8 | 1 0 0 0 | 2^192G
1437 * 9 | 1 0 0 1 | (2^192 + 1)G
1438 * 10 | 1 0 1 0 | (2^192 + 2^64)G
1439 * 11 | 1 0 1 1 | (2^192 + 2^64 + 1)G
1440 * 12 | 1 1 0 0 | (2^192 + 2^128)G
1441 * 13 | 1 1 0 1 | (2^192 + 2^128 + 1)G
1442 * 14 | 1 1 1 0 | (2^192 + 2^128 + 2^64)G
1443 * 15 | 1 1 1 1 | (2^192 + 2^128 + 2^64 + 1)G
1444 * followed by a copy of this with each element multiplied by 2^32.
1446 * The reason for this is so that we can clock bits into four different
1447 * locations when doing simple scalar multiplies against the base point,
1448 * and then another four locations using the second 16 elements.
1450 * Tables for other points have table[i] = iG for i in 0 .. 16. */
1452 /* gmul is the table of precomputed base points */
1453 static const smallfelem gmul[2][16][3] = {
1457 {{0xf4a13945d898c296, 0x77037d812deb33a0, 0xf8bce6e563a440f2,
1458 0x6b17d1f2e12c4247},
1459 {0xcbb6406837bf51f5, 0x2bce33576b315ece, 0x8ee7eb4a7c0f9e16,
1460 0x4fe342e2fe1a7f9b},
1462 {{0x90e75cb48e14db63, 0x29493baaad651f7e, 0x8492592e326e25de,
1463 0x0fa822bc2811aaa5},
1464 {0xe41124545f462ee7, 0x34b1a65050fe82f5, 0x6f4ad4bcb3df188b,
1465 0xbff44ae8f5dba80d},
1467 {{0x93391ce2097992af, 0xe96c98fd0d35f1fa, 0xb257c0de95e02789,
1468 0x300a4bbc89d6726f},
1469 {0xaa54a291c08127a0, 0x5bb1eeada9d806a5, 0x7f1ddb25ff1e3c6f,
1470 0x72aac7e0d09b4644},
1472 {{0x57c84fc9d789bd85, 0xfc35ff7dc297eac3, 0xfb982fd588c6766e,
1473 0x447d739beedb5e67},
1474 {0x0c7e33c972e25b32, 0x3d349b95a7fae500, 0xe12e9d953a4aaff7,
1475 0x2d4825ab834131ee},
1477 {{0x13949c932a1d367f, 0xef7fbd2b1a0a11b7, 0xddc6068bb91dfc60,
1478 0xef9519328a9c72ff},
1479 {0x196035a77376d8a8, 0x23183b0895ca1740, 0xc1ee9807022c219c,
1480 0x611e9fc37dbb2c9b},
1482 {{0xcae2b1920b57f4bc, 0x2936df5ec6c9bc36, 0x7dea6482e11238bf,
1483 0x550663797b51f5d8},
1484 {0x44ffe216348a964c, 0x9fb3d576dbdefbe1, 0x0afa40018d9d50e5,
1485 0x157164848aecb851},
1487 {{0xe48ecafffc5cde01, 0x7ccd84e70d715f26, 0xa2e8f483f43e4391,
1488 0xeb5d7745b21141ea},
1489 {0xcac917e2731a3479, 0x85f22cfe2844b645, 0x0990e6a158006cee,
1490 0xeafd72ebdbecc17b},
1492 {{0x6cf20ffb313728be, 0x96439591a3c6b94a, 0x2736ff8344315fc5,
1493 0xa6d39677a7849276},
1494 {0xf2bab833c357f5f4, 0x824a920c2284059b, 0x66b8babd2d27ecdf,
1495 0x674f84749b0b8816},
1497 {{0x2df48c04677c8a3e, 0x74e02f080203a56b, 0x31855f7db8c7fedb,
1498 0x4e769e7672c9ddad},
1499 {0xa4c36165b824bbb0, 0xfb9ae16f3b9122a5, 0x1ec0057206947281,
1500 0x42b99082de830663},
1502 {{0x6ef95150dda868b9, 0xd1f89e799c0ce131, 0x7fdc1ca008a1c478,
1503 0x78878ef61c6ce04d},
1504 {0x9c62b9121fe0d976, 0x6ace570ebde08d4f, 0xde53142c12309def,
1505 0xb6cb3f5d7b72c321},
1507 {{0x7f991ed2c31a3573, 0x5b82dd5bd54fb496, 0x595c5220812ffcae,
1508 0x0c88bc4d716b1287},
1509 {0x3a57bf635f48aca8, 0x7c8181f4df2564f3, 0x18d1b5b39c04e6aa,
1510 0xdd5ddea3f3901dc6},
1512 {{0xe96a79fb3e72ad0c, 0x43a0a28c42ba792f, 0xefe0a423083e49f3,
1513 0x68f344af6b317466},
1514 {0xcdfe17db3fb24d4a, 0x668bfc2271f5c626, 0x604ed93c24d67ff3,
1515 0x31b9c405f8540a20},
1517 {{0xd36b4789a2582e7f, 0x0d1a10144ec39c28, 0x663c62c3edbad7a0,
1518 0x4052bf4b6f461db9},
1519 {0x235a27c3188d25eb, 0xe724f33999bfcc5b, 0x862be6bd71d70cc8,
1520 0xfecf4d5190b0fc61},
1522 {{0x74346c10a1d4cfac, 0xafdf5cc08526a7a4, 0x123202a8f62bff7a,
1523 0x1eddbae2c802e41a},
1524 {0x8fa0af2dd603f844, 0x36e06b7e4c701917, 0x0c45f45273db33a0,
1525 0x43104d86560ebcfc},
1527 {{0x9615b5110d1d78e5, 0x66b0de3225c4744b, 0x0a4a46fb6aaf363a,
1528 0xb48e26b484f7a21c},
1529 {0x06ebb0f621a01b2d, 0xc004e4048b7b0f98, 0x64131bcdfed6f668,
1530 0xfac015404d4d3dab},
1535 {{0x3a5a9e22185a5943, 0x1ab919365c65dfb6, 0x21656b32262c71da,
1536 0x7fe36b40af22af89},
1537 {0xd50d152c699ca101, 0x74b3d5867b8af212, 0x9f09f40407dca6f1,
1538 0xe697d45825b63624},
1540 {{0xa84aa9397512218e, 0xe9a521b074ca0141, 0x57880b3a18a2e902,
1541 0x4a5b506612a677a6},
1542 {0x0beada7a4c4f3840, 0x626db15419e26d9d, 0xc42604fbe1627d40,
1543 0xeb13461ceac089f1},
1545 {{0xf9faed0927a43281, 0x5e52c4144103ecbc, 0xc342967aa815c857,
1546 0x0781b8291c6a220a},
1547 {0x5a8343ceeac55f80, 0x88f80eeee54a05e3, 0x97b2a14f12916434,
1548 0x690cde8df0151593},
1550 {{0xaee9c75df7f82f2a, 0x9e4c35874afdf43a, 0xf5622df437371326,
1551 0x8a535f566ec73617},
1552 {0xc5f9a0ac223094b7, 0xcde533864c8c7669, 0x37e02819085a92bf,
1553 0x0455c08468b08bd7},
1555 {{0x0c0a6e2c9477b5d9, 0xf9a4bf62876dc444, 0x5050a949b6cdc279,
1556 0x06bada7ab77f8276},
1557 {0xc8b4aed1ea48dac9, 0xdebd8a4b7ea1070f, 0x427d49101366eb70,
1558 0x5b476dfd0e6cb18a},
1560 {{0x7c5c3e44278c340a, 0x4d54606812d66f3b, 0x29a751b1ae23c5d8,
1561 0x3e29864e8a2ec908},
1562 {0x142d2a6626dbb850, 0xad1744c4765bd780, 0x1f150e68e322d1ed,
1563 0x239b90ea3dc31e7e},
1565 {{0x78c416527a53322a, 0x305dde6709776f8e, 0xdbcab759f8862ed4,
1566 0x820f4dd949f72ff7},
1567 {0x6cc544a62b5debd4, 0x75be5d937b4e8cc4, 0x1b481b1b215c14d3,
1568 0x140406ec783a05ec},
1570 {{0x6a703f10e895df07, 0xfd75f3fa01876bd8, 0xeb5b06e70ce08ffe,
1571 0x68f6b8542783dfee},
1572 {0x90c76f8a78712655, 0xcf5293d2f310bf7f, 0xfbc8044dfda45028,
1573 0xcbe1feba92e40ce6},
1575 {{0xe998ceea4396e4c1, 0xfc82ef0b6acea274, 0x230f729f2250e927,
1576 0xd0b2f94d2f420109},
1577 {0x4305adddb38d4966, 0x10b838f8624c3b45, 0x7db2636658954e7a,
1578 0x971459828b0719e5},
1580 {{0x4bd6b72623369fc9, 0x57f2929e53d0b876, 0xc2d5cba4f2340687,
1581 0x961610004a866aba},
1582 {0x49997bcd2e407a5e, 0x69ab197d92ddcb24, 0x2cf1f2438fe5131c,
1583 0x7acb9fadcee75e44},
1585 {{0x254e839423d2d4c0, 0xf57f0c917aea685b, 0xa60d880f6f75aaea,
1586 0x24eb9acca333bf5b},
1587 {0xe3de4ccb1cda5dea, 0xfeef9341c51a6b4f, 0x743125f88bac4c4d,
1588 0x69f891c5acd079cc},
1590 {{0xeee44b35702476b5, 0x7ed031a0e45c2258, 0xb422d1e7bd6f8514,
1591 0xe51f547c5972a107},
1592 {0xa25bcd6fc9cf343d, 0x8ca922ee097c184e, 0xa62f98b3a9fe9a06,
1593 0x1c309a2b25bb1387},
1595 {{0x9295dbeb1967c459, 0xb00148833472c98e, 0xc504977708011828,
1596 0x20b87b8aa2c4e503},
1597 {0x3063175de057c277, 0x1bd539338fe582dd, 0x0d11adef5f69a044,
1598 0xf5c6fa49919776be},
1600 {{0x8c944e760fd59e11, 0x3876cba1102fad5f, 0xa454c3fad83faa56,
1601 0x1ed7d1b9332010b9},
1602 {0xa1011a270024b889, 0x05e4d0dcac0cd344, 0x52b520f0eb6a2a24,
1603 0x3a2b03f03217257a},
1605 {{0xf20fc2afdf1d043d, 0xf330240db58d5a62, 0xfc7d229ca0058c3b,
1606 0x15fee545c78dd9f6},
1607 {0x501e82885bc98cda, 0x41ef80e5d046ac04, 0x557d9f49461210fb,
1608 0x4ab5b6b2b8753f81},
1613 * select_point selects the |idx|th point from a precomputation table and
1616 static void select_point(const u64 idx, unsigned int size,
1617 const smallfelem pre_comp[16][3], smallfelem out[3])
1620 u64 *outlimbs = &out[0][0];
1622 memset(out, 0, sizeof(*out) * 3);
1624 for (i = 0; i < size; i++) {
1625 const u64 *inlimbs = (u64 *)&pre_comp[i][0][0];
1632 for (j = 0; j < NLIMBS * 3; j++)
1633 outlimbs[j] |= inlimbs[j] & mask;
1637 /* get_bit returns the |i|th bit in |in| */
1638 static char get_bit(const felem_bytearray in, int i)
1640 if ((i < 0) || (i >= 256))
1642 return (in[i >> 3] >> (i & 7)) & 1;
1646 * Interleaved point multiplication using precomputed point multiples: The
1647 * small point multiples 0*P, 1*P, ..., 17*P are in pre_comp[], the scalars
1648 * in scalars[]. If g_scalar is non-NULL, we also add this multiple of the
1649 * generator, using certain (large) precomputed multiples in g_pre_comp.
1650 * Output point (X, Y, Z) is stored in x_out, y_out, z_out
1652 static void batch_mul(felem x_out, felem y_out, felem z_out,
1653 const felem_bytearray scalars[],
1654 const unsigned num_points, const u8 *g_scalar,
1655 const int mixed, const smallfelem pre_comp[][17][3],
1656 const smallfelem g_pre_comp[2][16][3])
1659 unsigned num, gen_mul = (g_scalar != NULL);
1665 /* set nq to the point at infinity */
1666 memset(nq, 0, sizeof(nq));
1669 * Loop over all scalars msb-to-lsb, interleaving additions of multiples
1670 * of the generator (two in each of the last 32 rounds) and additions of
1671 * other points multiples (every 5th round).
1673 skip = 1; /* save two point operations in the first
1675 for (i = (num_points ? 255 : 31); i >= 0; --i) {
1678 point_double(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2]);
1680 /* add multiples of the generator */
1681 if (gen_mul && (i <= 31)) {
1682 /* first, look 32 bits upwards */
1683 bits = get_bit(g_scalar, i + 224) << 3;
1684 bits |= get_bit(g_scalar, i + 160) << 2;
1685 bits |= get_bit(g_scalar, i + 96) << 1;
1686 bits |= get_bit(g_scalar, i + 32);
1687 /* select the point to add, in constant time */
1688 select_point(bits, 16, g_pre_comp[1], tmp);
1691 /* Arg 1 below is for "mixed" */
1692 point_add(nq[0], nq[1], nq[2],
1693 nq[0], nq[1], nq[2], 1, tmp[0], tmp[1], tmp[2]);
1695 smallfelem_expand(nq[0], tmp[0]);
1696 smallfelem_expand(nq[1], tmp[1]);
1697 smallfelem_expand(nq[2], tmp[2]);
1701 /* second, look at the current position */
1702 bits = get_bit(g_scalar, i + 192) << 3;
1703 bits |= get_bit(g_scalar, i + 128) << 2;
1704 bits |= get_bit(g_scalar, i + 64) << 1;
1705 bits |= get_bit(g_scalar, i);
1706 /* select the point to add, in constant time */
1707 select_point(bits, 16, g_pre_comp[0], tmp);
1708 /* Arg 1 below is for "mixed" */
1709 point_add(nq[0], nq[1], nq[2],
1710 nq[0], nq[1], nq[2], 1, tmp[0], tmp[1], tmp[2]);
1713 /* do other additions every 5 doublings */
1714 if (num_points && (i % 5 == 0)) {
1715 /* loop over all scalars */
1716 for (num = 0; num < num_points; ++num) {
1717 bits = get_bit(scalars[num], i + 4) << 5;
1718 bits |= get_bit(scalars[num], i + 3) << 4;
1719 bits |= get_bit(scalars[num], i + 2) << 3;
1720 bits |= get_bit(scalars[num], i + 1) << 2;
1721 bits |= get_bit(scalars[num], i) << 1;
1722 bits |= get_bit(scalars[num], i - 1);
1723 ec_GFp_nistp_recode_scalar_bits(&sign, &digit, bits);
1726 * select the point to add or subtract, in constant time
1728 select_point(digit, 17, pre_comp[num], tmp);
1729 smallfelem_neg(ftmp, tmp[1]); /* (X, -Y, Z) is the negative
1731 copy_small_conditional(ftmp, tmp[1], (((limb) sign) - 1));
1732 felem_contract(tmp[1], ftmp);
1735 point_add(nq[0], nq[1], nq[2],
1736 nq[0], nq[1], nq[2],
1737 mixed, tmp[0], tmp[1], tmp[2]);
1739 smallfelem_expand(nq[0], tmp[0]);
1740 smallfelem_expand(nq[1], tmp[1]);
1741 smallfelem_expand(nq[2], tmp[2]);
1747 felem_assign(x_out, nq[0]);
1748 felem_assign(y_out, nq[1]);
1749 felem_assign(z_out, nq[2]);
1752 /* Precomputation for the group generator. */
1753 struct nistp256_pre_comp_st {
1754 smallfelem g_pre_comp[2][16][3];
1755 CRYPTO_REF_COUNT references;
1756 CRYPTO_RWLOCK *lock;
1759 const EC_METHOD *EC_GFp_nistp256_method(void)
1761 static const EC_METHOD ret = {
1762 EC_FLAGS_DEFAULT_OCT,
1763 NID_X9_62_prime_field,
1764 ec_GFp_nistp256_group_init,
1765 ec_GFp_simple_group_finish,
1766 ec_GFp_simple_group_clear_finish,
1767 ec_GFp_nist_group_copy,
1768 ec_GFp_nistp256_group_set_curve,
1769 ec_GFp_simple_group_get_curve,
1770 ec_GFp_simple_group_get_degree,
1771 ec_group_simple_order_bits,
1772 ec_GFp_simple_group_check_discriminant,
1773 ec_GFp_simple_point_init,
1774 ec_GFp_simple_point_finish,
1775 ec_GFp_simple_point_clear_finish,
1776 ec_GFp_simple_point_copy,
1777 ec_GFp_simple_point_set_to_infinity,
1778 ec_GFp_simple_set_Jprojective_coordinates_GFp,
1779 ec_GFp_simple_get_Jprojective_coordinates_GFp,
1780 ec_GFp_simple_point_set_affine_coordinates,
1781 ec_GFp_nistp256_point_get_affine_coordinates,
1782 0 /* point_set_compressed_coordinates */ ,
1787 ec_GFp_simple_invert,
1788 ec_GFp_simple_is_at_infinity,
1789 ec_GFp_simple_is_on_curve,
1791 ec_GFp_simple_make_affine,
1792 ec_GFp_simple_points_make_affine,
1793 ec_GFp_nistp256_points_mul,
1794 ec_GFp_nistp256_precompute_mult,
1795 ec_GFp_nistp256_have_precompute_mult,
1796 ec_GFp_nist_field_mul,
1797 ec_GFp_nist_field_sqr,
1799 ec_GFp_simple_field_inv,
1800 0 /* field_encode */ ,
1801 0 /* field_decode */ ,
1802 0, /* field_set_to_one */
1803 ec_key_simple_priv2oct,
1804 ec_key_simple_oct2priv,
1805 0, /* set private */
1806 ec_key_simple_generate_key,
1807 ec_key_simple_check_key,
1808 ec_key_simple_generate_public_key,
1811 ecdh_simple_compute_key,
1812 ecdsa_simple_sign_setup,
1813 ecdsa_simple_sign_sig,
1814 ecdsa_simple_verify_sig,
1815 0, /* field_inverse_mod_ord */
1816 0, /* blind_coordinates */
1818 0, /* ladder_step */
1825 /******************************************************************************/
1827 * FUNCTIONS TO MANAGE PRECOMPUTATION
1830 static NISTP256_PRE_COMP *nistp256_pre_comp_new(void)
1832 NISTP256_PRE_COMP *ret = OPENSSL_zalloc(sizeof(*ret));
1835 ECerr(EC_F_NISTP256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1839 ret->references = 1;
1841 ret->lock = CRYPTO_THREAD_lock_new();
1842 if (ret->lock == NULL) {
1843 ECerr(EC_F_NISTP256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1850 NISTP256_PRE_COMP *EC_nistp256_pre_comp_dup(NISTP256_PRE_COMP *p)
1854 CRYPTO_UP_REF(&p->references, &i, p->lock);
1858 void EC_nistp256_pre_comp_free(NISTP256_PRE_COMP *pre)
1865 CRYPTO_DOWN_REF(&pre->references, &i, pre->lock);
1866 REF_PRINT_COUNT("EC_nistp256", x);
1869 REF_ASSERT_ISNT(i < 0);
1871 CRYPTO_THREAD_lock_free(pre->lock);
1875 /******************************************************************************/
1877 * OPENSSL EC_METHOD FUNCTIONS
1880 int ec_GFp_nistp256_group_init(EC_GROUP *group)
1883 ret = ec_GFp_simple_group_init(group);
1884 group->a_is_minus3 = 1;
1888 int ec_GFp_nistp256_group_set_curve(EC_GROUP *group, const BIGNUM *p,
1889 const BIGNUM *a, const BIGNUM *b,
1893 BIGNUM *curve_p, *curve_a, *curve_b;
1895 BN_CTX *new_ctx = NULL;
1898 ctx = new_ctx = BN_CTX_new();
1904 curve_p = BN_CTX_get(ctx);
1905 curve_a = BN_CTX_get(ctx);
1906 curve_b = BN_CTX_get(ctx);
1907 if (curve_b == NULL)
1909 BN_bin2bn(nistp256_curve_params[0], sizeof(felem_bytearray), curve_p);
1910 BN_bin2bn(nistp256_curve_params[1], sizeof(felem_bytearray), curve_a);
1911 BN_bin2bn(nistp256_curve_params[2], sizeof(felem_bytearray), curve_b);
1912 if ((BN_cmp(curve_p, p)) || (BN_cmp(curve_a, a)) || (BN_cmp(curve_b, b))) {
1913 ECerr(EC_F_EC_GFP_NISTP256_GROUP_SET_CURVE,
1914 EC_R_WRONG_CURVE_PARAMETERS);
1917 group->field_mod_func = BN_nist_mod_256;
1918 ret = ec_GFp_simple_group_set_curve(group, p, a, b, ctx);
1922 BN_CTX_free(new_ctx);
1928 * Takes the Jacobian coordinates (X, Y, Z) of a point and returns (X', Y') =
1931 int ec_GFp_nistp256_point_get_affine_coordinates(const EC_GROUP *group,
1932 const EC_POINT *point,
1933 BIGNUM *x, BIGNUM *y,
1936 felem z1, z2, x_in, y_in;
1937 smallfelem x_out, y_out;
1940 if (EC_POINT_is_at_infinity(group, point)) {
1941 ECerr(EC_F_EC_GFP_NISTP256_POINT_GET_AFFINE_COORDINATES,
1942 EC_R_POINT_AT_INFINITY);
1945 if ((!BN_to_felem(x_in, point->X)) || (!BN_to_felem(y_in, point->Y)) ||
1946 (!BN_to_felem(z1, point->Z)))
1949 felem_square(tmp, z2);
1950 felem_reduce(z1, tmp);
1951 felem_mul(tmp, x_in, z1);
1952 felem_reduce(x_in, tmp);
1953 felem_contract(x_out, x_in);
1955 if (!smallfelem_to_BN(x, x_out)) {
1956 ECerr(EC_F_EC_GFP_NISTP256_POINT_GET_AFFINE_COORDINATES,
1961 felem_mul(tmp, z1, z2);
1962 felem_reduce(z1, tmp);
1963 felem_mul(tmp, y_in, z1);
1964 felem_reduce(y_in, tmp);
1965 felem_contract(y_out, y_in);
1967 if (!smallfelem_to_BN(y, y_out)) {
1968 ECerr(EC_F_EC_GFP_NISTP256_POINT_GET_AFFINE_COORDINATES,
1976 /* points below is of size |num|, and tmp_smallfelems is of size |num+1| */
1977 static void make_points_affine(size_t num, smallfelem points[][3],
1978 smallfelem tmp_smallfelems[])
1981 * Runs in constant time, unless an input is the point at infinity (which
1982 * normally shouldn't happen).
1984 ec_GFp_nistp_points_make_affine_internal(num,
1988 (void (*)(void *))smallfelem_one,
1989 smallfelem_is_zero_int,
1990 (void (*)(void *, const void *))
1992 (void (*)(void *, const void *))
1993 smallfelem_square_contract,
1995 (void *, const void *,
1997 smallfelem_mul_contract,
1998 (void (*)(void *, const void *))
1999 smallfelem_inv_contract,
2000 /* nothing to contract */
2001 (void (*)(void *, const void *))
2006 * Computes scalar*generator + \sum scalars[i]*points[i], ignoring NULL
2007 * values Result is stored in r (r can equal one of the inputs).
2009 int ec_GFp_nistp256_points_mul(const EC_GROUP *group, EC_POINT *r,
2010 const BIGNUM *scalar, size_t num,
2011 const EC_POINT *points[],
2012 const BIGNUM *scalars[], BN_CTX *ctx)
2017 BIGNUM *x, *y, *z, *tmp_scalar;
2018 felem_bytearray g_secret;
2019 felem_bytearray *secrets = NULL;
2020 smallfelem (*pre_comp)[17][3] = NULL;
2021 smallfelem *tmp_smallfelems = NULL;
2024 int have_pre_comp = 0;
2025 size_t num_points = num;
2026 smallfelem x_in, y_in, z_in;
2027 felem x_out, y_out, z_out;
2028 NISTP256_PRE_COMP *pre = NULL;
2029 const smallfelem(*g_pre_comp)[16][3] = NULL;
2030 EC_POINT *generator = NULL;
2031 const EC_POINT *p = NULL;
2032 const BIGNUM *p_scalar = NULL;
2035 x = BN_CTX_get(ctx);
2036 y = BN_CTX_get(ctx);
2037 z = BN_CTX_get(ctx);
2038 tmp_scalar = BN_CTX_get(ctx);
2039 if (tmp_scalar == NULL)
2042 if (scalar != NULL) {
2043 pre = group->pre_comp.nistp256;
2045 /* we have precomputation, try to use it */
2046 g_pre_comp = (const smallfelem(*)[16][3])pre->g_pre_comp;
2048 /* try to use the standard precomputation */
2049 g_pre_comp = &gmul[0];
2050 generator = EC_POINT_new(group);
2051 if (generator == NULL)
2053 /* get the generator from precomputation */
2054 if (!smallfelem_to_BN(x, g_pre_comp[0][1][0]) ||
2055 !smallfelem_to_BN(y, g_pre_comp[0][1][1]) ||
2056 !smallfelem_to_BN(z, g_pre_comp[0][1][2])) {
2057 ECerr(EC_F_EC_GFP_NISTP256_POINTS_MUL, ERR_R_BN_LIB);
2060 if (!EC_POINT_set_Jprojective_coordinates_GFp(group,
2064 if (0 == EC_POINT_cmp(group, generator, group->generator, ctx))
2065 /* precomputation matches generator */
2069 * we don't have valid precomputation: treat the generator as a
2074 if (num_points > 0) {
2075 if (num_points >= 3) {
2077 * unless we precompute multiples for just one or two points,
2078 * converting those into affine form is time well spent
2082 secrets = OPENSSL_malloc(sizeof(*secrets) * num_points);
2083 pre_comp = OPENSSL_malloc(sizeof(*pre_comp) * num_points);
2086 OPENSSL_malloc(sizeof(*tmp_smallfelems) * (num_points * 17 + 1));
2087 if ((secrets == NULL) || (pre_comp == NULL)
2088 || (mixed && (tmp_smallfelems == NULL))) {
2089 ECerr(EC_F_EC_GFP_NISTP256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
2094 * we treat NULL scalars as 0, and NULL points as points at infinity,
2095 * i.e., they contribute nothing to the linear combination
2097 memset(secrets, 0, sizeof(*secrets) * num_points);
2098 memset(pre_comp, 0, sizeof(*pre_comp) * num_points);
2099 for (i = 0; i < num_points; ++i) {
2102 * we didn't have a valid precomputation, so we pick the
2105 p = EC_GROUP_get0_generator(group);
2108 /* the i^th point */
2110 p_scalar = scalars[i];
2112 if ((p_scalar != NULL) && (p != NULL)) {
2113 /* reduce scalar to 0 <= scalar < 2^256 */
2114 if ((BN_num_bits(p_scalar) > 256)
2115 || (BN_is_negative(p_scalar))) {
2117 * this is an unusual input, and we don't guarantee
2120 if (!BN_nnmod(tmp_scalar, p_scalar, group->order, ctx)) {
2121 ECerr(EC_F_EC_GFP_NISTP256_POINTS_MUL, ERR_R_BN_LIB);
2124 num_bytes = BN_bn2lebinpad(tmp_scalar,
2125 secrets[i], sizeof(secrets[i]));
2127 num_bytes = BN_bn2lebinpad(p_scalar,
2128 secrets[i], sizeof(secrets[i]));
2130 if (num_bytes < 0) {
2131 ECerr(EC_F_EC_GFP_NISTP256_POINTS_MUL, ERR_R_BN_LIB);
2134 /* precompute multiples */
2135 if ((!BN_to_felem(x_out, p->X)) ||
2136 (!BN_to_felem(y_out, p->Y)) ||
2137 (!BN_to_felem(z_out, p->Z)))
2139 felem_shrink(pre_comp[i][1][0], x_out);
2140 felem_shrink(pre_comp[i][1][1], y_out);
2141 felem_shrink(pre_comp[i][1][2], z_out);
2142 for (j = 2; j <= 16; ++j) {
2144 point_add_small(pre_comp[i][j][0], pre_comp[i][j][1],
2145 pre_comp[i][j][2], pre_comp[i][1][0],
2146 pre_comp[i][1][1], pre_comp[i][1][2],
2147 pre_comp[i][j - 1][0],
2148 pre_comp[i][j - 1][1],
2149 pre_comp[i][j - 1][2]);
2151 point_double_small(pre_comp[i][j][0],
2154 pre_comp[i][j / 2][0],
2155 pre_comp[i][j / 2][1],
2156 pre_comp[i][j / 2][2]);
2162 make_points_affine(num_points * 17, pre_comp[0], tmp_smallfelems);
2165 /* the scalar for the generator */
2166 if ((scalar != NULL) && (have_pre_comp)) {
2167 memset(g_secret, 0, sizeof(g_secret));
2168 /* reduce scalar to 0 <= scalar < 2^256 */
2169 if ((BN_num_bits(scalar) > 256) || (BN_is_negative(scalar))) {
2171 * this is an unusual input, and we don't guarantee
2174 if (!BN_nnmod(tmp_scalar, scalar, group->order, ctx)) {
2175 ECerr(EC_F_EC_GFP_NISTP256_POINTS_MUL, ERR_R_BN_LIB);
2178 num_bytes = BN_bn2lebinpad(tmp_scalar, g_secret, sizeof(g_secret));
2180 num_bytes = BN_bn2lebinpad(scalar, g_secret, sizeof(g_secret));
2182 /* do the multiplication with generator precomputation */
2183 batch_mul(x_out, y_out, z_out,
2184 (const felem_bytearray(*))secrets, num_points,
2186 mixed, (const smallfelem(*)[17][3])pre_comp, g_pre_comp);
2188 /* do the multiplication without generator precomputation */
2189 batch_mul(x_out, y_out, z_out,
2190 (const felem_bytearray(*))secrets, num_points,
2191 NULL, mixed, (const smallfelem(*)[17][3])pre_comp, NULL);
2193 /* reduce the output to its unique minimal representation */
2194 felem_contract(x_in, x_out);
2195 felem_contract(y_in, y_out);
2196 felem_contract(z_in, z_out);
2197 if ((!smallfelem_to_BN(x, x_in)) || (!smallfelem_to_BN(y, y_in)) ||
2198 (!smallfelem_to_BN(z, z_in))) {
2199 ECerr(EC_F_EC_GFP_NISTP256_POINTS_MUL, ERR_R_BN_LIB);
2202 ret = EC_POINT_set_Jprojective_coordinates_GFp(group, r, x, y, z, ctx);
2206 EC_POINT_free(generator);
2207 OPENSSL_free(secrets);
2208 OPENSSL_free(pre_comp);
2209 OPENSSL_free(tmp_smallfelems);
2213 int ec_GFp_nistp256_precompute_mult(EC_GROUP *group, BN_CTX *ctx)
2216 NISTP256_PRE_COMP *pre = NULL;
2219 EC_POINT *generator = NULL;
2220 smallfelem tmp_smallfelems[32];
2221 felem x_tmp, y_tmp, z_tmp;
2223 BN_CTX *new_ctx = NULL;
2226 /* throw away old precomputation */
2227 EC_pre_comp_free(group);
2231 ctx = new_ctx = BN_CTX_new();
2237 x = BN_CTX_get(ctx);
2238 y = BN_CTX_get(ctx);
2241 /* get the generator */
2242 if (group->generator == NULL)
2244 generator = EC_POINT_new(group);
2245 if (generator == NULL)
2247 BN_bin2bn(nistp256_curve_params[3], sizeof(felem_bytearray), x);
2248 BN_bin2bn(nistp256_curve_params[4], sizeof(felem_bytearray), y);
2249 if (!EC_POINT_set_affine_coordinates(group, generator, x, y, ctx))
2251 if ((pre = nistp256_pre_comp_new()) == NULL)
2254 * if the generator is the standard one, use built-in precomputation
2256 if (0 == EC_POINT_cmp(group, generator, group->generator, ctx)) {
2257 memcpy(pre->g_pre_comp, gmul, sizeof(pre->g_pre_comp));
2260 if ((!BN_to_felem(x_tmp, group->generator->X)) ||
2261 (!BN_to_felem(y_tmp, group->generator->Y)) ||
2262 (!BN_to_felem(z_tmp, group->generator->Z)))
2264 felem_shrink(pre->g_pre_comp[0][1][0], x_tmp);
2265 felem_shrink(pre->g_pre_comp[0][1][1], y_tmp);
2266 felem_shrink(pre->g_pre_comp[0][1][2], z_tmp);
2268 * compute 2^64*G, 2^128*G, 2^192*G for the first table, 2^32*G, 2^96*G,
2269 * 2^160*G, 2^224*G for the second one
2271 for (i = 1; i <= 8; i <<= 1) {
2272 point_double_small(pre->g_pre_comp[1][i][0], pre->g_pre_comp[1][i][1],
2273 pre->g_pre_comp[1][i][2], pre->g_pre_comp[0][i][0],
2274 pre->g_pre_comp[0][i][1],
2275 pre->g_pre_comp[0][i][2]);
2276 for (j = 0; j < 31; ++j) {
2277 point_double_small(pre->g_pre_comp[1][i][0],
2278 pre->g_pre_comp[1][i][1],
2279 pre->g_pre_comp[1][i][2],
2280 pre->g_pre_comp[1][i][0],
2281 pre->g_pre_comp[1][i][1],
2282 pre->g_pre_comp[1][i][2]);
2286 point_double_small(pre->g_pre_comp[0][2 * i][0],
2287 pre->g_pre_comp[0][2 * i][1],
2288 pre->g_pre_comp[0][2 * i][2],
2289 pre->g_pre_comp[1][i][0], pre->g_pre_comp[1][i][1],
2290 pre->g_pre_comp[1][i][2]);
2291 for (j = 0; j < 31; ++j) {
2292 point_double_small(pre->g_pre_comp[0][2 * i][0],
2293 pre->g_pre_comp[0][2 * i][1],
2294 pre->g_pre_comp[0][2 * i][2],
2295 pre->g_pre_comp[0][2 * i][0],
2296 pre->g_pre_comp[0][2 * i][1],
2297 pre->g_pre_comp[0][2 * i][2]);
2300 for (i = 0; i < 2; i++) {
2301 /* g_pre_comp[i][0] is the point at infinity */
2302 memset(pre->g_pre_comp[i][0], 0, sizeof(pre->g_pre_comp[i][0]));
2303 /* the remaining multiples */
2304 /* 2^64*G + 2^128*G resp. 2^96*G + 2^160*G */
2305 point_add_small(pre->g_pre_comp[i][6][0], pre->g_pre_comp[i][6][1],
2306 pre->g_pre_comp[i][6][2], pre->g_pre_comp[i][4][0],
2307 pre->g_pre_comp[i][4][1], pre->g_pre_comp[i][4][2],
2308 pre->g_pre_comp[i][2][0], pre->g_pre_comp[i][2][1],
2309 pre->g_pre_comp[i][2][2]);
2310 /* 2^64*G + 2^192*G resp. 2^96*G + 2^224*G */
2311 point_add_small(pre->g_pre_comp[i][10][0], pre->g_pre_comp[i][10][1],
2312 pre->g_pre_comp[i][10][2], pre->g_pre_comp[i][8][0],
2313 pre->g_pre_comp[i][8][1], pre->g_pre_comp[i][8][2],
2314 pre->g_pre_comp[i][2][0], pre->g_pre_comp[i][2][1],
2315 pre->g_pre_comp[i][2][2]);
2316 /* 2^128*G + 2^192*G resp. 2^160*G + 2^224*G */
2317 point_add_small(pre->g_pre_comp[i][12][0], pre->g_pre_comp[i][12][1],
2318 pre->g_pre_comp[i][12][2], pre->g_pre_comp[i][8][0],
2319 pre->g_pre_comp[i][8][1], pre->g_pre_comp[i][8][2],
2320 pre->g_pre_comp[i][4][0], pre->g_pre_comp[i][4][1],
2321 pre->g_pre_comp[i][4][2]);
2323 * 2^64*G + 2^128*G + 2^192*G resp. 2^96*G + 2^160*G + 2^224*G
2325 point_add_small(pre->g_pre_comp[i][14][0], pre->g_pre_comp[i][14][1],
2326 pre->g_pre_comp[i][14][2], pre->g_pre_comp[i][12][0],
2327 pre->g_pre_comp[i][12][1], pre->g_pre_comp[i][12][2],
2328 pre->g_pre_comp[i][2][0], pre->g_pre_comp[i][2][1],
2329 pre->g_pre_comp[i][2][2]);
2330 for (j = 1; j < 8; ++j) {
2331 /* odd multiples: add G resp. 2^32*G */
2332 point_add_small(pre->g_pre_comp[i][2 * j + 1][0],
2333 pre->g_pre_comp[i][2 * j + 1][1],
2334 pre->g_pre_comp[i][2 * j + 1][2],
2335 pre->g_pre_comp[i][2 * j][0],
2336 pre->g_pre_comp[i][2 * j][1],
2337 pre->g_pre_comp[i][2 * j][2],
2338 pre->g_pre_comp[i][1][0],
2339 pre->g_pre_comp[i][1][1],
2340 pre->g_pre_comp[i][1][2]);
2343 make_points_affine(31, &(pre->g_pre_comp[0][1]), tmp_smallfelems);
2346 SETPRECOMP(group, nistp256, pre);
2352 EC_POINT_free(generator);
2354 BN_CTX_free(new_ctx);
2356 EC_nistp256_pre_comp_free(pre);
2360 int ec_GFp_nistp256_have_precompute_mult(const EC_GROUP *group)
2362 return HAVEPRECOMP(group, nistp256);