2 * Copyright 2014-2022 The OpenSSL Project Authors. All Rights Reserved.
3 * Copyright (c) 2014, Intel Corporation. All Rights Reserved.
4 * Copyright (c) 2015, CloudFlare, Inc.
6 * Licensed under the Apache License 2.0 (the "License"). You may not use
7 * this file except in compliance with the License. You can obtain a copy
8 * in the file LICENSE in the source distribution or at
9 * https://www.openssl.org/source/license.html
11 * Originally written by Shay Gueron (1, 2), and Vlad Krasnov (1, 3)
12 * (1) Intel Corporation, Israel Development Center, Haifa, Israel
13 * (2) University of Haifa, Israel
14 * (3) CloudFlare, Inc.
17 * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with
22 * ECDSA low level APIs are deprecated for public use, but still ok for
25 #include "internal/deprecated.h"
29 #include "internal/cryptlib.h"
30 #include "crypto/bn.h"
32 #include "internal/refcount.h"
35 # define TOBN(hi,lo) lo,hi
37 # define TOBN(hi,lo) ((BN_ULONG)hi<<32|lo)
41 # define ALIGN32 __attribute((aligned(32)))
42 #elif defined(_MSC_VER)
43 # define ALIGN32 __declspec(align(32))
48 #define ALIGNPTR(p,N) ((unsigned char *)p+N-(size_t)p%N)
49 #define P256_LIMBS (256/BN_BITS2)
51 typedef unsigned short u16;
54 BN_ULONG X[P256_LIMBS];
55 BN_ULONG Y[P256_LIMBS];
56 BN_ULONG Z[P256_LIMBS];
60 BN_ULONG X[P256_LIMBS];
61 BN_ULONG Y[P256_LIMBS];
64 typedef P256_POINT_AFFINE PRECOMP256_ROW[64];
66 /* structure for precomputed multiples of the generator */
67 struct nistz256_pre_comp_st {
68 const EC_GROUP *group; /* Parent EC_GROUP object */
69 size_t w; /* Window size */
71 * Constant time access to the X and Y coordinates of the pre-computed,
72 * generator multiplies, in the Montgomery domain. Pre-calculated
73 * multiplies are stored in affine form.
75 PRECOMP256_ROW *precomp;
76 void *precomp_storage;
77 CRYPTO_REF_COUNT references;
80 /* Functions implemented in assembly */
82 * Most of below mentioned functions *preserve* the property of inputs
83 * being fully reduced, i.e. being in [0, modulus) range. Simply put if
84 * inputs are fully reduced, then output is too. Note that reverse is
85 * not true, in sense that given partially reduced inputs output can be
86 * either, not unlikely reduced. And "most" in first sentence refers to
87 * the fact that given the calculations flow one can tolerate that
88 * addition, 1st function below, produces partially reduced result *if*
89 * multiplications by 2 and 3, which customarily use addition, fully
90 * reduce it. This effectively gives two options: a) addition produces
91 * fully reduced result [as long as inputs are, just like remaining
92 * functions]; b) addition is allowed to produce partially reduced
93 * result, but multiplications by 2 and 3 perform additional reduction
94 * step. Choice between the two can be platform-specific, but it was a)
95 * in all cases so far...
97 /* Modular add: res = a+b mod P */
98 void ecp_nistz256_add(BN_ULONG res[P256_LIMBS],
99 const BN_ULONG a[P256_LIMBS],
100 const BN_ULONG b[P256_LIMBS]);
101 /* Modular mul by 2: res = 2*a mod P */
102 void ecp_nistz256_mul_by_2(BN_ULONG res[P256_LIMBS],
103 const BN_ULONG a[P256_LIMBS]);
104 /* Modular mul by 3: res = 3*a mod P */
105 void ecp_nistz256_mul_by_3(BN_ULONG res[P256_LIMBS],
106 const BN_ULONG a[P256_LIMBS]);
108 /* Modular div by 2: res = a/2 mod P */
109 void ecp_nistz256_div_by_2(BN_ULONG res[P256_LIMBS],
110 const BN_ULONG a[P256_LIMBS]);
111 /* Modular sub: res = a-b mod P */
112 void ecp_nistz256_sub(BN_ULONG res[P256_LIMBS],
113 const BN_ULONG a[P256_LIMBS],
114 const BN_ULONG b[P256_LIMBS]);
115 /* Modular neg: res = -a mod P */
116 void ecp_nistz256_neg(BN_ULONG res[P256_LIMBS], const BN_ULONG a[P256_LIMBS]);
117 /* Montgomery mul: res = a*b*2^-256 mod P */
118 void ecp_nistz256_mul_mont(BN_ULONG res[P256_LIMBS],
119 const BN_ULONG a[P256_LIMBS],
120 const BN_ULONG b[P256_LIMBS]);
121 /* Montgomery sqr: res = a*a*2^-256 mod P */
122 void ecp_nistz256_sqr_mont(BN_ULONG res[P256_LIMBS],
123 const BN_ULONG a[P256_LIMBS]);
124 /* Convert a number from Montgomery domain, by multiplying with 1 */
125 void ecp_nistz256_from_mont(BN_ULONG res[P256_LIMBS],
126 const BN_ULONG in[P256_LIMBS]);
127 /* Convert a number to Montgomery domain, by multiplying with 2^512 mod P*/
128 void ecp_nistz256_to_mont(BN_ULONG res[P256_LIMBS],
129 const BN_ULONG in[P256_LIMBS]);
130 /* Functions that perform constant time access to the precomputed tables */
131 void ecp_nistz256_scatter_w5(P256_POINT *val,
132 const P256_POINT *in_t, int idx);
133 void ecp_nistz256_gather_w5(P256_POINT *val,
134 const P256_POINT *in_t, int idx);
135 void ecp_nistz256_scatter_w7(P256_POINT_AFFINE *val,
136 const P256_POINT_AFFINE *in_t, int idx);
137 void ecp_nistz256_gather_w7(P256_POINT_AFFINE *val,
138 const P256_POINT_AFFINE *in_t, int idx);
140 /* One converted into the Montgomery domain */
141 static const BN_ULONG ONE[P256_LIMBS] = {
142 TOBN(0x00000000, 0x00000001), TOBN(0xffffffff, 0x00000000),
143 TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe)
146 static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group);
148 /* Precomputed tables for the default generator */
149 extern const PRECOMP256_ROW ecp_nistz256_precomputed[37];
151 /* Recode window to a signed digit, see ecp_nistputil.c for details */
152 static unsigned int _booth_recode_w5(unsigned int in)
156 s = ~((in >> 5) - 1);
157 d = (1 << 6) - in - 1;
158 d = (d & s) | (in & ~s);
159 d = (d >> 1) + (d & 1);
161 return (d << 1) + (s & 1);
164 static unsigned int _booth_recode_w7(unsigned int in)
168 s = ~((in >> 7) - 1);
169 d = (1 << 8) - in - 1;
170 d = (d & s) | (in & ~s);
171 d = (d >> 1) + (d & 1);
173 return (d << 1) + (s & 1);
176 static void copy_conditional(BN_ULONG dst[P256_LIMBS],
177 const BN_ULONG src[P256_LIMBS], BN_ULONG move)
179 BN_ULONG mask1 = 0-move;
180 BN_ULONG mask2 = ~mask1;
182 dst[0] = (src[0] & mask1) ^ (dst[0] & mask2);
183 dst[1] = (src[1] & mask1) ^ (dst[1] & mask2);
184 dst[2] = (src[2] & mask1) ^ (dst[2] & mask2);
185 dst[3] = (src[3] & mask1) ^ (dst[3] & mask2);
186 if (P256_LIMBS == 8) {
187 dst[4] = (src[4] & mask1) ^ (dst[4] & mask2);
188 dst[5] = (src[5] & mask1) ^ (dst[5] & mask2);
189 dst[6] = (src[6] & mask1) ^ (dst[6] & mask2);
190 dst[7] = (src[7] & mask1) ^ (dst[7] & mask2);
194 static BN_ULONG is_zero(BN_ULONG in)
202 static BN_ULONG is_equal(const BN_ULONG a[P256_LIMBS],
203 const BN_ULONG b[P256_LIMBS])
211 if (P256_LIMBS == 8) {
221 static BN_ULONG is_one(const BIGNUM *z)
224 BN_ULONG *a = bn_get_words(z);
226 if (bn_get_top(z) == (P256_LIMBS - P256_LIMBS / 8)) {
228 res |= a[1] ^ ONE[1];
229 res |= a[2] ^ ONE[2];
230 res |= a[3] ^ ONE[3];
231 if (P256_LIMBS == 8) {
232 res |= a[4] ^ ONE[4];
233 res |= a[5] ^ ONE[5];
234 res |= a[6] ^ ONE[6];
236 * no check for a[7] (being zero) on 32-bit platforms,
237 * because value of "one" takes only 7 limbs.
247 * For reference, this macro is used only when new ecp_nistz256 assembly
248 * module is being developed. For example, configure with
249 * -DECP_NISTZ256_REFERENCE_IMPLEMENTATION and implement only functions
250 * performing simplest arithmetic operations on 256-bit vectors. Then
251 * work on implementation of higher-level functions performing point
252 * operations. Then remove ECP_NISTZ256_REFERENCE_IMPLEMENTATION
253 * and never define it again. (The correct macro denoting presence of
254 * ecp_nistz256 module is ECP_NISTZ256_ASM.)
256 #ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
257 void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a);
258 void ecp_nistz256_point_add(P256_POINT *r,
259 const P256_POINT *a, const P256_POINT *b);
260 void ecp_nistz256_point_add_affine(P256_POINT *r,
262 const P256_POINT_AFFINE *b);
264 /* Point double: r = 2*a */
265 static void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a)
267 BN_ULONG S[P256_LIMBS];
268 BN_ULONG M[P256_LIMBS];
269 BN_ULONG Zsqr[P256_LIMBS];
270 BN_ULONG tmp0[P256_LIMBS];
272 const BN_ULONG *in_x = a->X;
273 const BN_ULONG *in_y = a->Y;
274 const BN_ULONG *in_z = a->Z;
276 BN_ULONG *res_x = r->X;
277 BN_ULONG *res_y = r->Y;
278 BN_ULONG *res_z = r->Z;
280 ecp_nistz256_mul_by_2(S, in_y);
282 ecp_nistz256_sqr_mont(Zsqr, in_z);
284 ecp_nistz256_sqr_mont(S, S);
286 ecp_nistz256_mul_mont(res_z, in_z, in_y);
287 ecp_nistz256_mul_by_2(res_z, res_z);
289 ecp_nistz256_add(M, in_x, Zsqr);
290 ecp_nistz256_sub(Zsqr, in_x, Zsqr);
292 ecp_nistz256_sqr_mont(res_y, S);
293 ecp_nistz256_div_by_2(res_y, res_y);
295 ecp_nistz256_mul_mont(M, M, Zsqr);
296 ecp_nistz256_mul_by_3(M, M);
298 ecp_nistz256_mul_mont(S, S, in_x);
299 ecp_nistz256_mul_by_2(tmp0, S);
301 ecp_nistz256_sqr_mont(res_x, M);
303 ecp_nistz256_sub(res_x, res_x, tmp0);
304 ecp_nistz256_sub(S, S, res_x);
306 ecp_nistz256_mul_mont(S, S, M);
307 ecp_nistz256_sub(res_y, S, res_y);
310 /* Point addition: r = a+b */
311 static void ecp_nistz256_point_add(P256_POINT *r,
312 const P256_POINT *a, const P256_POINT *b)
314 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
315 BN_ULONG U1[P256_LIMBS], S1[P256_LIMBS];
316 BN_ULONG Z1sqr[P256_LIMBS];
317 BN_ULONG Z2sqr[P256_LIMBS];
318 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
319 BN_ULONG Hsqr[P256_LIMBS];
320 BN_ULONG Rsqr[P256_LIMBS];
321 BN_ULONG Hcub[P256_LIMBS];
323 BN_ULONG res_x[P256_LIMBS];
324 BN_ULONG res_y[P256_LIMBS];
325 BN_ULONG res_z[P256_LIMBS];
327 BN_ULONG in1infty, in2infty;
329 const BN_ULONG *in1_x = a->X;
330 const BN_ULONG *in1_y = a->Y;
331 const BN_ULONG *in1_z = a->Z;
333 const BN_ULONG *in2_x = b->X;
334 const BN_ULONG *in2_y = b->Y;
335 const BN_ULONG *in2_z = b->Z;
338 * Infinity in encoded as (,,0)
340 in1infty = (in1_z[0] | in1_z[1] | in1_z[2] | in1_z[3]);
342 in1infty |= (in1_z[4] | in1_z[5] | in1_z[6] | in1_z[7]);
344 in2infty = (in2_z[0] | in2_z[1] | in2_z[2] | in2_z[3]);
346 in2infty |= (in2_z[4] | in2_z[5] | in2_z[6] | in2_z[7]);
348 in1infty = is_zero(in1infty);
349 in2infty = is_zero(in2infty);
351 ecp_nistz256_sqr_mont(Z2sqr, in2_z); /* Z2^2 */
352 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
354 ecp_nistz256_mul_mont(S1, Z2sqr, in2_z); /* S1 = Z2^3 */
355 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
357 ecp_nistz256_mul_mont(S1, S1, in1_y); /* S1 = Y1*Z2^3 */
358 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
359 ecp_nistz256_sub(R, S2, S1); /* R = S2 - S1 */
361 ecp_nistz256_mul_mont(U1, in1_x, Z2sqr); /* U1 = X1*Z2^2 */
362 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
363 ecp_nistz256_sub(H, U2, U1); /* H = U2 - U1 */
366 * The formulae are incorrect if the points are equal so we check for
367 * this and do doubling if this happens.
369 * Points here are in Jacobian projective coordinates (Xi, Yi, Zi)
370 * that are bound to the affine coordinates (xi, yi) by the following
375 * For the sake of optimization, the algorithm operates over
376 * intermediate variables U1, U2 and S1, S2 that are derived from
377 * the projective coordinates:
378 * - U1 = X1 * (Z2)^2 ; U2 = X2 * (Z1)^2
379 * - S1 = Y1 * (Z2)^3 ; S2 = Y2 * (Z1)^3
381 * It is easy to prove that is_equal(U1, U2) implies that the affine
382 * x-coordinates are equal, or either point is at infinity.
383 * Likewise is_equal(S1, S2) implies that the affine y-coordinates are
384 * equal, or either point is at infinity.
386 * The special case of either point being the point at infinity (Z1 or Z2
387 * is zero), is handled separately later on in this function, so we avoid
388 * jumping to point_double here in those special cases.
390 * When both points are inverse of each other, we know that the affine
391 * x-coordinates are equal, and the y-coordinates have different sign.
392 * Therefore since U1 = U2, we know H = 0, and therefore Z3 = H*Z1*Z2
393 * will equal 0, thus the result is infinity, if we simply let this
394 * function continue normally.
396 * We use bitwise operations to avoid potential side-channels introduced by
397 * the short-circuiting behaviour of boolean operators.
399 if (is_equal(U1, U2) & ~in1infty & ~in2infty & is_equal(S1, S2)) {
401 * This is obviously not constant-time but it should never happen during
402 * single point multiplication, so there is no timing leak for ECDH or
405 ecp_nistz256_point_double(r, a);
409 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
410 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
411 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
412 ecp_nistz256_mul_mont(res_z, res_z, in2_z); /* Z3 = H*Z1*Z2 */
413 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
415 ecp_nistz256_mul_mont(U2, U1, Hsqr); /* U1*H^2 */
416 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
418 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
419 ecp_nistz256_sub(res_x, res_x, Hcub);
421 ecp_nistz256_sub(res_y, U2, res_x);
423 ecp_nistz256_mul_mont(S2, S1, Hcub);
424 ecp_nistz256_mul_mont(res_y, R, res_y);
425 ecp_nistz256_sub(res_y, res_y, S2);
427 copy_conditional(res_x, in2_x, in1infty);
428 copy_conditional(res_y, in2_y, in1infty);
429 copy_conditional(res_z, in2_z, in1infty);
431 copy_conditional(res_x, in1_x, in2infty);
432 copy_conditional(res_y, in1_y, in2infty);
433 copy_conditional(res_z, in1_z, in2infty);
435 memcpy(r->X, res_x, sizeof(res_x));
436 memcpy(r->Y, res_y, sizeof(res_y));
437 memcpy(r->Z, res_z, sizeof(res_z));
440 /* Point addition when b is known to be affine: r = a+b */
441 static void ecp_nistz256_point_add_affine(P256_POINT *r,
443 const P256_POINT_AFFINE *b)
445 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
446 BN_ULONG Z1sqr[P256_LIMBS];
447 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
448 BN_ULONG Hsqr[P256_LIMBS];
449 BN_ULONG Rsqr[P256_LIMBS];
450 BN_ULONG Hcub[P256_LIMBS];
452 BN_ULONG res_x[P256_LIMBS];
453 BN_ULONG res_y[P256_LIMBS];
454 BN_ULONG res_z[P256_LIMBS];
456 BN_ULONG in1infty, in2infty;
458 const BN_ULONG *in1_x = a->X;
459 const BN_ULONG *in1_y = a->Y;
460 const BN_ULONG *in1_z = a->Z;
462 const BN_ULONG *in2_x = b->X;
463 const BN_ULONG *in2_y = b->Y;
466 * Infinity in encoded as (,,0)
468 in1infty = (in1_z[0] | in1_z[1] | in1_z[2] | in1_z[3]);
470 in1infty |= (in1_z[4] | in1_z[5] | in1_z[6] | in1_z[7]);
473 * In affine representation we encode infinity as (0,0), which is
474 * not on the curve, so it is OK
476 in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
477 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
479 in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
480 in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
482 in1infty = is_zero(in1infty);
483 in2infty = is_zero(in2infty);
485 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
487 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
488 ecp_nistz256_sub(H, U2, in1_x); /* H = U2 - U1 */
490 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
492 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
494 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
495 ecp_nistz256_sub(R, S2, in1_y); /* R = S2 - S1 */
497 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
498 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
499 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
501 ecp_nistz256_mul_mont(U2, in1_x, Hsqr); /* U1*H^2 */
502 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
504 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
505 ecp_nistz256_sub(res_x, res_x, Hcub);
506 ecp_nistz256_sub(H, U2, res_x);
508 ecp_nistz256_mul_mont(S2, in1_y, Hcub);
509 ecp_nistz256_mul_mont(H, H, R);
510 ecp_nistz256_sub(res_y, H, S2);
512 copy_conditional(res_x, in2_x, in1infty);
513 copy_conditional(res_x, in1_x, in2infty);
515 copy_conditional(res_y, in2_y, in1infty);
516 copy_conditional(res_y, in1_y, in2infty);
518 copy_conditional(res_z, ONE, in1infty);
519 copy_conditional(res_z, in1_z, in2infty);
521 memcpy(r->X, res_x, sizeof(res_x));
522 memcpy(r->Y, res_y, sizeof(res_y));
523 memcpy(r->Z, res_z, sizeof(res_z));
527 /* r = in^-1 mod p */
528 static void ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS],
529 const BN_ULONG in[P256_LIMBS])
532 * The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff
533 * ffffffff ffffffff We use FLT and used poly-2 as exponent
535 BN_ULONG p2[P256_LIMBS];
536 BN_ULONG p4[P256_LIMBS];
537 BN_ULONG p8[P256_LIMBS];
538 BN_ULONG p16[P256_LIMBS];
539 BN_ULONG p32[P256_LIMBS];
540 BN_ULONG res[P256_LIMBS];
543 ecp_nistz256_sqr_mont(res, in);
544 ecp_nistz256_mul_mont(p2, res, in); /* 3*p */
546 ecp_nistz256_sqr_mont(res, p2);
547 ecp_nistz256_sqr_mont(res, res);
548 ecp_nistz256_mul_mont(p4, res, p2); /* f*p */
550 ecp_nistz256_sqr_mont(res, p4);
551 ecp_nistz256_sqr_mont(res, res);
552 ecp_nistz256_sqr_mont(res, res);
553 ecp_nistz256_sqr_mont(res, res);
554 ecp_nistz256_mul_mont(p8, res, p4); /* ff*p */
556 ecp_nistz256_sqr_mont(res, p8);
557 for (i = 0; i < 7; i++)
558 ecp_nistz256_sqr_mont(res, res);
559 ecp_nistz256_mul_mont(p16, res, p8); /* ffff*p */
561 ecp_nistz256_sqr_mont(res, p16);
562 for (i = 0; i < 15; i++)
563 ecp_nistz256_sqr_mont(res, res);
564 ecp_nistz256_mul_mont(p32, res, p16); /* ffffffff*p */
566 ecp_nistz256_sqr_mont(res, p32);
567 for (i = 0; i < 31; i++)
568 ecp_nistz256_sqr_mont(res, res);
569 ecp_nistz256_mul_mont(res, res, in);
571 for (i = 0; i < 32 * 4; i++)
572 ecp_nistz256_sqr_mont(res, res);
573 ecp_nistz256_mul_mont(res, res, p32);
575 for (i = 0; i < 32; i++)
576 ecp_nistz256_sqr_mont(res, res);
577 ecp_nistz256_mul_mont(res, res, p32);
579 for (i = 0; i < 16; i++)
580 ecp_nistz256_sqr_mont(res, res);
581 ecp_nistz256_mul_mont(res, res, p16);
583 for (i = 0; i < 8; i++)
584 ecp_nistz256_sqr_mont(res, res);
585 ecp_nistz256_mul_mont(res, res, p8);
587 ecp_nistz256_sqr_mont(res, res);
588 ecp_nistz256_sqr_mont(res, res);
589 ecp_nistz256_sqr_mont(res, res);
590 ecp_nistz256_sqr_mont(res, res);
591 ecp_nistz256_mul_mont(res, res, p4);
593 ecp_nistz256_sqr_mont(res, res);
594 ecp_nistz256_sqr_mont(res, res);
595 ecp_nistz256_mul_mont(res, res, p2);
597 ecp_nistz256_sqr_mont(res, res);
598 ecp_nistz256_sqr_mont(res, res);
599 ecp_nistz256_mul_mont(res, res, in);
601 memcpy(r, res, sizeof(res));
605 * ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
606 * returns one if it fits. Otherwise it returns zero.
608 __owur static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],
611 return bn_copy_words(out, in, P256_LIMBS);
614 /* r = sum(scalar[i]*point[i]) */
615 __owur static int ecp_nistz256_windowed_mul(const EC_GROUP *group,
617 const BIGNUM **scalar,
618 const EC_POINT **point,
619 size_t num, BN_CTX *ctx)
624 unsigned char (*p_str)[33] = NULL;
625 const unsigned int window_size = 5;
626 const unsigned int mask = (1 << (window_size + 1)) - 1;
628 P256_POINT *temp; /* place for 5 temporary points */
629 const BIGNUM **scalars = NULL;
630 P256_POINT (*table)[16] = NULL;
631 void *table_storage = NULL;
633 if ((num * 16 + 6) > OPENSSL_MALLOC_MAX_NELEMS(P256_POINT)
635 OPENSSL_malloc((num * 16 + 5) * sizeof(P256_POINT) + 64)) == NULL
637 OPENSSL_malloc(num * 33 * sizeof(unsigned char))) == NULL
638 || (scalars = OPENSSL_malloc(num * sizeof(BIGNUM *))) == NULL)
641 table = (void *)ALIGNPTR(table_storage, 64);
642 temp = (P256_POINT *)(table + num);
644 for (i = 0; i < num; i++) {
645 P256_POINT *row = table[i];
647 /* This is an unusual input, we don't guarantee constant-timeness. */
648 if ((BN_num_bits(scalar[i]) > 256) || BN_is_negative(scalar[i])) {
651 if ((mod = BN_CTX_get(ctx)) == NULL)
653 if (!BN_nnmod(mod, scalar[i], group->order, ctx)) {
654 ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
659 scalars[i] = scalar[i];
661 for (j = 0; j < bn_get_top(scalars[i]) * BN_BYTES; j += BN_BYTES) {
662 BN_ULONG d = bn_get_words(scalars[i])[j / BN_BYTES];
664 p_str[i][j + 0] = (unsigned char)d;
665 p_str[i][j + 1] = (unsigned char)(d >> 8);
666 p_str[i][j + 2] = (unsigned char)(d >> 16);
667 p_str[i][j + 3] = (unsigned char)(d >>= 24);
670 p_str[i][j + 4] = (unsigned char)d;
671 p_str[i][j + 5] = (unsigned char)(d >> 8);
672 p_str[i][j + 6] = (unsigned char)(d >> 16);
673 p_str[i][j + 7] = (unsigned char)(d >> 24);
679 if (!ecp_nistz256_bignum_to_field_elem(temp[0].X, point[i]->X)
680 || !ecp_nistz256_bignum_to_field_elem(temp[0].Y, point[i]->Y)
681 || !ecp_nistz256_bignum_to_field_elem(temp[0].Z, point[i]->Z)) {
682 ERR_raise(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
687 * row[0] is implicitly (0,0,0) (the point at infinity), therefore it
688 * is not stored. All other values are actually stored with an offset
692 ecp_nistz256_scatter_w5 (row, &temp[0], 1);
693 ecp_nistz256_point_double(&temp[1], &temp[0]); /*1+1=2 */
694 ecp_nistz256_scatter_w5 (row, &temp[1], 2);
695 ecp_nistz256_point_add (&temp[2], &temp[1], &temp[0]); /*2+1=3 */
696 ecp_nistz256_scatter_w5 (row, &temp[2], 3);
697 ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*2=4 */
698 ecp_nistz256_scatter_w5 (row, &temp[1], 4);
699 ecp_nistz256_point_double(&temp[2], &temp[2]); /*2*3=6 */
700 ecp_nistz256_scatter_w5 (row, &temp[2], 6);
701 ecp_nistz256_point_add (&temp[3], &temp[1], &temp[0]); /*4+1=5 */
702 ecp_nistz256_scatter_w5 (row, &temp[3], 5);
703 ecp_nistz256_point_add (&temp[4], &temp[2], &temp[0]); /*6+1=7 */
704 ecp_nistz256_scatter_w5 (row, &temp[4], 7);
705 ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*4=8 */
706 ecp_nistz256_scatter_w5 (row, &temp[1], 8);
707 ecp_nistz256_point_double(&temp[2], &temp[2]); /*2*6=12 */
708 ecp_nistz256_scatter_w5 (row, &temp[2], 12);
709 ecp_nistz256_point_double(&temp[3], &temp[3]); /*2*5=10 */
710 ecp_nistz256_scatter_w5 (row, &temp[3], 10);
711 ecp_nistz256_point_double(&temp[4], &temp[4]); /*2*7=14 */
712 ecp_nistz256_scatter_w5 (row, &temp[4], 14);
713 ecp_nistz256_point_add (&temp[2], &temp[2], &temp[0]); /*12+1=13*/
714 ecp_nistz256_scatter_w5 (row, &temp[2], 13);
715 ecp_nistz256_point_add (&temp[3], &temp[3], &temp[0]); /*10+1=11*/
716 ecp_nistz256_scatter_w5 (row, &temp[3], 11);
717 ecp_nistz256_point_add (&temp[4], &temp[4], &temp[0]); /*14+1=15*/
718 ecp_nistz256_scatter_w5 (row, &temp[4], 15);
719 ecp_nistz256_point_add (&temp[2], &temp[1], &temp[0]); /*8+1=9 */
720 ecp_nistz256_scatter_w5 (row, &temp[2], 9);
721 ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*8=16 */
722 ecp_nistz256_scatter_w5 (row, &temp[1], 16);
727 wvalue = p_str[0][(idx - 1) / 8];
728 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
731 * We gather to temp[0], because we know it's position relative
734 ecp_nistz256_gather_w5(&temp[0], table[0], _booth_recode_w5(wvalue) >> 1);
735 memcpy(r, &temp[0], sizeof(temp[0]));
738 for (i = (idx == 255 ? 1 : 0); i < num; i++) {
739 unsigned int off = (idx - 1) / 8;
741 wvalue = p_str[i][off] | p_str[i][off + 1] << 8;
742 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
744 wvalue = _booth_recode_w5(wvalue);
746 ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1);
748 ecp_nistz256_neg(temp[1].Y, temp[0].Y);
749 copy_conditional(temp[0].Y, temp[1].Y, (wvalue & 1));
751 ecp_nistz256_point_add(r, r, &temp[0]);
756 ecp_nistz256_point_double(r, r);
757 ecp_nistz256_point_double(r, r);
758 ecp_nistz256_point_double(r, r);
759 ecp_nistz256_point_double(r, r);
760 ecp_nistz256_point_double(r, r);
764 for (i = 0; i < num; i++) {
765 wvalue = p_str[i][0];
766 wvalue = (wvalue << 1) & mask;
768 wvalue = _booth_recode_w5(wvalue);
770 ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1);
772 ecp_nistz256_neg(temp[1].Y, temp[0].Y);
773 copy_conditional(temp[0].Y, temp[1].Y, wvalue & 1);
775 ecp_nistz256_point_add(r, r, &temp[0]);
780 OPENSSL_free(table_storage);
782 OPENSSL_free(scalars);
786 /* Coordinates of G, for which we have precomputed tables */
787 static const BN_ULONG def_xG[P256_LIMBS] = {
788 TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
789 TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
792 static const BN_ULONG def_yG[P256_LIMBS] = {
793 TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
794 TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
798 * ecp_nistz256_is_affine_G returns one if |generator| is the standard, P-256
801 static int ecp_nistz256_is_affine_G(const EC_POINT *generator)
803 return (bn_get_top(generator->X) == P256_LIMBS) &&
804 (bn_get_top(generator->Y) == P256_LIMBS) &&
805 is_equal(bn_get_words(generator->X), def_xG) &&
806 is_equal(bn_get_words(generator->Y), def_yG) &&
807 is_one(generator->Z);
810 __owur static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx)
813 * We precompute a table for a Booth encoded exponent (wNAF) based
814 * computation. Each table holds 64 values for safe access, with an
815 * implicit value of infinity at index zero. We use window of size 7, and
816 * therefore require ceil(256/7) = 37 tables.
819 EC_POINT *P = NULL, *T = NULL;
820 const EC_POINT *generator;
821 NISTZ256_PRE_COMP *pre_comp;
822 BN_CTX *new_ctx = NULL;
823 int i, j, k, ret = 0;
826 PRECOMP256_ROW *preComputedTable = NULL;
827 unsigned char *precomp_storage = NULL;
829 /* if there is an old NISTZ256_PRE_COMP object, throw it away */
830 EC_pre_comp_free(group);
831 generator = EC_GROUP_get0_generator(group);
832 if (generator == NULL) {
833 ERR_raise(ERR_LIB_EC, EC_R_UNDEFINED_GENERATOR);
837 if (ecp_nistz256_is_affine_G(generator)) {
839 * No need to calculate tables for the standard generator because we
840 * have them statically.
845 if ((pre_comp = ecp_nistz256_pre_comp_new(group)) == NULL)
849 ctx = new_ctx = BN_CTX_new_ex(group->libctx);
856 order = EC_GROUP_get0_order(group);
860 if (BN_is_zero(order)) {
861 ERR_raise(ERR_LIB_EC, EC_R_UNKNOWN_ORDER);
867 if ((precomp_storage =
868 OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE) + 64)) == NULL)
871 preComputedTable = (void *)ALIGNPTR(precomp_storage, 64);
873 P = EC_POINT_new(group);
874 T = EC_POINT_new(group);
875 if (P == NULL || T == NULL)
879 * The zero entry is implicitly infinity, and we skip it, storing other
880 * values with -1 offset.
882 if (!EC_POINT_copy(T, generator))
885 for (k = 0; k < 64; k++) {
886 if (!EC_POINT_copy(P, T))
888 for (j = 0; j < 37; j++) {
889 P256_POINT_AFFINE temp;
891 * It would be faster to use EC_POINTs_make_affine and
892 * make multiple points affine at the same time.
894 if (group->meth->make_affine == NULL
895 || !group->meth->make_affine(group, P, ctx))
897 if (!ecp_nistz256_bignum_to_field_elem(temp.X, P->X) ||
898 !ecp_nistz256_bignum_to_field_elem(temp.Y, P->Y)) {
899 ERR_raise(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
902 ecp_nistz256_scatter_w7(preComputedTable[j], &temp, k);
903 for (i = 0; i < 7; i++) {
904 if (!EC_POINT_dbl(group, P, P, ctx))
908 if (!EC_POINT_add(group, T, T, generator, ctx))
912 pre_comp->group = group;
914 pre_comp->precomp = preComputedTable;
915 pre_comp->precomp_storage = precomp_storage;
916 precomp_storage = NULL;
917 SETPRECOMP(group, nistz256, pre_comp);
923 BN_CTX_free(new_ctx);
925 EC_nistz256_pre_comp_free(pre_comp);
926 OPENSSL_free(precomp_storage);
932 __owur static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group,
933 const P256_POINT_AFFINE *in,
938 if ((ret = bn_set_words(out->X, in->X, P256_LIMBS))
939 && (ret = bn_set_words(out->Y, in->Y, P256_LIMBS))
940 && (ret = bn_set_words(out->Z, ONE, P256_LIMBS)))
946 /* r = scalar*G + sum(scalars[i]*points[i]) */
947 __owur static int ecp_nistz256_points_mul(const EC_GROUP *group,
949 const BIGNUM *scalar,
951 const EC_POINT *points[],
952 const BIGNUM *scalars[], BN_CTX *ctx)
954 int i = 0, ret = 0, no_precomp_for_generator = 0, p_is_infinity = 0;
955 unsigned char p_str[33] = { 0 };
956 const PRECOMP256_ROW *preComputedTable = NULL;
957 const NISTZ256_PRE_COMP *pre_comp = NULL;
958 const EC_POINT *generator = NULL;
959 const BIGNUM **new_scalars = NULL;
960 const EC_POINT **new_points = NULL;
961 unsigned int idx = 0;
962 const unsigned int window_size = 7;
963 const unsigned int mask = (1 << (window_size + 1)) - 1;
971 if ((num + 1) == 0 || (num + 1) > OPENSSL_MALLOC_MAX_NELEMS(void *)) {
972 ERR_raise(ERR_LIB_EC, ERR_R_PASSED_INVALID_ARGUMENT);
976 memset(&p, 0, sizeof(p));
980 generator = EC_GROUP_get0_generator(group);
981 if (generator == NULL) {
982 ERR_raise(ERR_LIB_EC, EC_R_UNDEFINED_GENERATOR);
986 /* look if we can use precomputed multiples of generator */
987 pre_comp = group->pre_comp.nistz256;
991 * If there is a precomputed table for the generator, check that
992 * it was generated with the same generator.
994 EC_POINT *pre_comp_generator = EC_POINT_new(group);
995 if (pre_comp_generator == NULL)
998 ecp_nistz256_gather_w7(&p.a, pre_comp->precomp[0], 1);
999 if (!ecp_nistz256_set_from_affine(pre_comp_generator,
1000 group, &p.a, ctx)) {
1001 EC_POINT_free(pre_comp_generator);
1005 if (0 == EC_POINT_cmp(group, generator, pre_comp_generator, ctx))
1006 preComputedTable = (const PRECOMP256_ROW *)pre_comp->precomp;
1008 EC_POINT_free(pre_comp_generator);
1011 if (preComputedTable == NULL && ecp_nistz256_is_affine_G(generator)) {
1013 * If there is no precomputed data, but the generator is the
1014 * default, a hardcoded table of precomputed data is used. This
1015 * is because applications, such as Apache, do not use
1016 * EC_KEY_precompute_mult.
1018 preComputedTable = ecp_nistz256_precomputed;
1021 if (preComputedTable) {
1024 if ((BN_num_bits(scalar) > 256)
1025 || BN_is_negative(scalar)) {
1026 if ((tmp_scalar = BN_CTX_get(ctx)) == NULL)
1029 if (!BN_nnmod(tmp_scalar, scalar, group->order, ctx)) {
1030 ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1033 scalar = tmp_scalar;
1036 for (i = 0; i < bn_get_top(scalar) * BN_BYTES; i += BN_BYTES) {
1037 BN_ULONG d = bn_get_words(scalar)[i / BN_BYTES];
1039 p_str[i + 0] = (unsigned char)d;
1040 p_str[i + 1] = (unsigned char)(d >> 8);
1041 p_str[i + 2] = (unsigned char)(d >> 16);
1042 p_str[i + 3] = (unsigned char)(d >>= 24);
1043 if (BN_BYTES == 8) {
1045 p_str[i + 4] = (unsigned char)d;
1046 p_str[i + 5] = (unsigned char)(d >> 8);
1047 p_str[i + 6] = (unsigned char)(d >> 16);
1048 p_str[i + 7] = (unsigned char)(d >> 24);
1056 wvalue = (p_str[0] << 1) & mask;
1059 wvalue = _booth_recode_w7(wvalue);
1061 ecp_nistz256_gather_w7(&p.a, preComputedTable[0],
1064 ecp_nistz256_neg(p.p.Z, p.p.Y);
1065 copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
1068 * Since affine infinity is encoded as (0,0) and
1069 * Jacobian is (,,0), we need to harmonize them
1070 * by assigning "one" or zero to Z.
1072 infty = (p.p.X[0] | p.p.X[1] | p.p.X[2] | p.p.X[3] |
1073 p.p.Y[0] | p.p.Y[1] | p.p.Y[2] | p.p.Y[3]);
1074 if (P256_LIMBS == 8)
1075 infty |= (p.p.X[4] | p.p.X[5] | p.p.X[6] | p.p.X[7] |
1076 p.p.Y[4] | p.p.Y[5] | p.p.Y[6] | p.p.Y[7]);
1078 infty = 0 - is_zero(infty);
1081 p.p.Z[0] = ONE[0] & infty;
1082 p.p.Z[1] = ONE[1] & infty;
1083 p.p.Z[2] = ONE[2] & infty;
1084 p.p.Z[3] = ONE[3] & infty;
1085 if (P256_LIMBS == 8) {
1086 p.p.Z[4] = ONE[4] & infty;
1087 p.p.Z[5] = ONE[5] & infty;
1088 p.p.Z[6] = ONE[6] & infty;
1089 p.p.Z[7] = ONE[7] & infty;
1092 for (i = 1; i < 37; i++) {
1093 unsigned int off = (idx - 1) / 8;
1094 wvalue = p_str[off] | p_str[off + 1] << 8;
1095 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1098 wvalue = _booth_recode_w7(wvalue);
1100 ecp_nistz256_gather_w7(&t.a,
1101 preComputedTable[i], wvalue >> 1);
1103 ecp_nistz256_neg(t.p.Z, t.a.Y);
1104 copy_conditional(t.a.Y, t.p.Z, wvalue & 1);
1106 ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a);
1110 no_precomp_for_generator = 1;
1115 if (no_precomp_for_generator) {
1117 * Without a precomputed table for the generator, it has to be
1118 * handled like a normal point.
1120 new_scalars = OPENSSL_malloc((num + 1) * sizeof(BIGNUM *));
1121 if (new_scalars == NULL)
1124 new_points = OPENSSL_malloc((num + 1) * sizeof(EC_POINT *));
1125 if (new_points == NULL)
1128 memcpy(new_scalars, scalars, num * sizeof(BIGNUM *));
1129 new_scalars[num] = scalar;
1130 memcpy(new_points, points, num * sizeof(EC_POINT *));
1131 new_points[num] = generator;
1133 scalars = new_scalars;
1134 points = new_points;
1139 P256_POINT *out = &t.p;
1143 if (!ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx))
1147 ecp_nistz256_point_add(&p.p, &p.p, out);
1150 /* Not constant-time, but we're only operating on the public output. */
1151 if (!bn_set_words(r->X, p.p.X, P256_LIMBS) ||
1152 !bn_set_words(r->Y, p.p.Y, P256_LIMBS) ||
1153 !bn_set_words(r->Z, p.p.Z, P256_LIMBS)) {
1156 r->Z_is_one = is_one(r->Z) & 1;
1162 OPENSSL_free(new_points);
1163 OPENSSL_free(new_scalars);
1167 __owur static int ecp_nistz256_get_affine(const EC_GROUP *group,
1168 const EC_POINT *point,
1169 BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
1171 BN_ULONG z_inv2[P256_LIMBS];
1172 BN_ULONG z_inv3[P256_LIMBS];
1173 BN_ULONG x_aff[P256_LIMBS];
1174 BN_ULONG y_aff[P256_LIMBS];
1175 BN_ULONG point_x[P256_LIMBS], point_y[P256_LIMBS], point_z[P256_LIMBS];
1176 BN_ULONG x_ret[P256_LIMBS], y_ret[P256_LIMBS];
1178 if (EC_POINT_is_at_infinity(group, point)) {
1179 ERR_raise(ERR_LIB_EC, EC_R_POINT_AT_INFINITY);
1183 if (!ecp_nistz256_bignum_to_field_elem(point_x, point->X) ||
1184 !ecp_nistz256_bignum_to_field_elem(point_y, point->Y) ||
1185 !ecp_nistz256_bignum_to_field_elem(point_z, point->Z)) {
1186 ERR_raise(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
1190 ecp_nistz256_mod_inverse(z_inv3, point_z);
1191 ecp_nistz256_sqr_mont(z_inv2, z_inv3);
1192 ecp_nistz256_mul_mont(x_aff, z_inv2, point_x);
1195 ecp_nistz256_from_mont(x_ret, x_aff);
1196 if (!bn_set_words(x, x_ret, P256_LIMBS))
1201 ecp_nistz256_mul_mont(z_inv3, z_inv3, z_inv2);
1202 ecp_nistz256_mul_mont(y_aff, z_inv3, point_y);
1203 ecp_nistz256_from_mont(y_ret, y_aff);
1204 if (!bn_set_words(y, y_ret, P256_LIMBS))
1211 static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group)
1213 NISTZ256_PRE_COMP *ret = NULL;
1218 ret = OPENSSL_zalloc(sizeof(*ret));
1224 ret->w = 6; /* default */
1226 if (!CRYPTO_NEW_REF(&ret->references, 1)) {
1233 NISTZ256_PRE_COMP *EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP *p)
1237 CRYPTO_UP_REF(&p->references, &i);
1241 void EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP *pre)
1248 CRYPTO_DOWN_REF(&pre->references, &i);
1249 REF_PRINT_COUNT("EC_nistz256", pre);
1252 REF_ASSERT_ISNT(i < 0);
1254 OPENSSL_free(pre->precomp_storage);
1255 CRYPTO_FREE_REF(&pre->references);
1260 static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP *group)
1262 /* There is a hard-coded table for the default generator. */
1263 const EC_POINT *generator = EC_GROUP_get0_generator(group);
1265 if (generator != NULL && ecp_nistz256_is_affine_G(generator)) {
1266 /* There is a hard-coded table for the default generator. */
1270 return HAVEPRECOMP(group, nistz256);
1273 #if defined(__x86_64) || defined(__x86_64__) || \
1274 defined(_M_AMD64) || defined(_M_X64) || \
1275 defined(__powerpc64__) || defined(_ARCH_PP64) || \
1276 defined(__aarch64__)
1278 * Montgomery mul modulo Order(P): res = a*b*2^-256 mod Order(P)
1280 void ecp_nistz256_ord_mul_mont(BN_ULONG res[P256_LIMBS],
1281 const BN_ULONG a[P256_LIMBS],
1282 const BN_ULONG b[P256_LIMBS]);
1283 void ecp_nistz256_ord_sqr_mont(BN_ULONG res[P256_LIMBS],
1284 const BN_ULONG a[P256_LIMBS],
1287 static int ecp_nistz256_inv_mod_ord(const EC_GROUP *group, BIGNUM *r,
1288 const BIGNUM *x, BN_CTX *ctx)
1290 /* RR = 2^512 mod ord(p256) */
1291 static const BN_ULONG RR[P256_LIMBS] = {
1292 TOBN(0x83244c95,0xbe79eea2), TOBN(0x4699799c,0x49bd6fa6),
1293 TOBN(0x2845b239,0x2b6bec59), TOBN(0x66e12d94,0xf3d95620)
1295 /* The constant 1 (unlike ONE that is one in Montgomery representation) */
1296 static const BN_ULONG one[P256_LIMBS] = {
1297 TOBN(0,1), TOBN(0,0), TOBN(0,0), TOBN(0,0)
1300 * We don't use entry 0 in the table, so we omit it and address
1303 BN_ULONG table[15][P256_LIMBS];
1304 BN_ULONG out[P256_LIMBS], t[P256_LIMBS];
1307 i_1 = 0, i_10, i_11, i_101, i_111, i_1010, i_1111,
1308 i_10101, i_101010, i_101111, i_x6, i_x8, i_x16, i_x32
1312 * Catch allocation failure early.
1314 if (bn_wexpand(r, P256_LIMBS) == NULL) {
1315 ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1319 if ((BN_num_bits(x) > 256) || BN_is_negative(x)) {
1322 if ((tmp = BN_CTX_get(ctx)) == NULL
1323 || !BN_nnmod(tmp, x, group->order, ctx)) {
1324 ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1330 if (!ecp_nistz256_bignum_to_field_elem(t, x)) {
1331 ERR_raise(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
1335 ecp_nistz256_ord_mul_mont(table[0], t, RR);
1338 * Original sparse-then-fixed-window algorithm, retained for reference.
1340 for (i = 2; i < 16; i += 2) {
1341 ecp_nistz256_ord_sqr_mont(table[i-1], table[i/2-1], 1);
1342 ecp_nistz256_ord_mul_mont(table[i], table[i-1], table[0]);
1346 * The top 128bit of the exponent are highly redudndant, so we
1347 * perform an optimized flow
1349 ecp_nistz256_ord_sqr_mont(t, table[15-1], 4); /* f0 */
1350 ecp_nistz256_ord_mul_mont(t, t, table[15-1]); /* ff */
1352 ecp_nistz256_ord_sqr_mont(out, t, 8); /* ff00 */
1353 ecp_nistz256_ord_mul_mont(out, out, t); /* ffff */
1355 ecp_nistz256_ord_sqr_mont(t, out, 16); /* ffff0000 */
1356 ecp_nistz256_ord_mul_mont(t, t, out); /* ffffffff */
1358 ecp_nistz256_ord_sqr_mont(out, t, 64); /* ffffffff0000000000000000 */
1359 ecp_nistz256_ord_mul_mont(out, out, t); /* ffffffff00000000ffffffff */
1361 ecp_nistz256_ord_sqr_mont(out, out, 32); /* ffffffff00000000ffffffff00000000 */
1362 ecp_nistz256_ord_mul_mont(out, out, t); /* ffffffff00000000ffffffffffffffff */
1365 * The bottom 128 bit of the exponent are processed with fixed 4-bit window
1367 for (i = 0; i < 32; i++) {
1368 /* expLo - the low 128 bits of the exponent we use (ord(p256) - 2),
1369 * split into nibbles */
1370 static const unsigned char expLo[32] = {
1371 0xb,0xc,0xe,0x6,0xf,0xa,0xa,0xd,0xa,0x7,0x1,0x7,0x9,0xe,0x8,0x4,
1372 0xf,0x3,0xb,0x9,0xc,0xa,0xc,0x2,0xf,0xc,0x6,0x3,0x2,0x5,0x4,0xf
1375 ecp_nistz256_ord_sqr_mont(out, out, 4);
1376 /* The exponent is public, no need in constant-time access */
1377 ecp_nistz256_ord_mul_mont(out, out, table[expLo[i]-1]);
1381 * https://briansmith.org/ecc-inversion-addition-chains-01#p256_scalar_inversion
1383 * Even though this code path spares 12 squarings, 4.5%, and 13
1384 * multiplications, 25%, on grand scale sign operation is not that
1385 * much faster, not more that 2%...
1388 /* pre-calculate powers */
1389 ecp_nistz256_ord_sqr_mont(table[i_10], table[i_1], 1);
1391 ecp_nistz256_ord_mul_mont(table[i_11], table[i_1], table[i_10]);
1393 ecp_nistz256_ord_mul_mont(table[i_101], table[i_11], table[i_10]);
1395 ecp_nistz256_ord_mul_mont(table[i_111], table[i_101], table[i_10]);
1397 ecp_nistz256_ord_sqr_mont(table[i_1010], table[i_101], 1);
1399 ecp_nistz256_ord_mul_mont(table[i_1111], table[i_1010], table[i_101]);
1401 ecp_nistz256_ord_sqr_mont(table[i_10101], table[i_1010], 1);
1402 ecp_nistz256_ord_mul_mont(table[i_10101], table[i_10101], table[i_1]);
1404 ecp_nistz256_ord_sqr_mont(table[i_101010], table[i_10101], 1);
1406 ecp_nistz256_ord_mul_mont(table[i_101111], table[i_101010], table[i_101]);
1408 ecp_nistz256_ord_mul_mont(table[i_x6], table[i_101010], table[i_10101]);
1410 ecp_nistz256_ord_sqr_mont(table[i_x8], table[i_x6], 2);
1411 ecp_nistz256_ord_mul_mont(table[i_x8], table[i_x8], table[i_11]);
1413 ecp_nistz256_ord_sqr_mont(table[i_x16], table[i_x8], 8);
1414 ecp_nistz256_ord_mul_mont(table[i_x16], table[i_x16], table[i_x8]);
1416 ecp_nistz256_ord_sqr_mont(table[i_x32], table[i_x16], 16);
1417 ecp_nistz256_ord_mul_mont(table[i_x32], table[i_x32], table[i_x16]);
1420 ecp_nistz256_ord_sqr_mont(out, table[i_x32], 64);
1421 ecp_nistz256_ord_mul_mont(out, out, table[i_x32]);
1423 for (i = 0; i < 27; i++) {
1424 static const struct { unsigned char p, i; } chain[27] = {
1425 { 32, i_x32 }, { 6, i_101111 }, { 5, i_111 },
1426 { 4, i_11 }, { 5, i_1111 }, { 5, i_10101 },
1427 { 4, i_101 }, { 3, i_101 }, { 3, i_101 },
1428 { 5, i_111 }, { 9, i_101111 }, { 6, i_1111 },
1429 { 2, i_1 }, { 5, i_1 }, { 6, i_1111 },
1430 { 5, i_111 }, { 4, i_111 }, { 5, i_111 },
1431 { 5, i_101 }, { 3, i_11 }, { 10, i_101111 },
1432 { 2, i_11 }, { 5, i_11 }, { 5, i_11 },
1433 { 3, i_1 }, { 7, i_10101 }, { 6, i_1111 }
1436 ecp_nistz256_ord_sqr_mont(out, out, chain[i].p);
1437 ecp_nistz256_ord_mul_mont(out, out, table[chain[i].i]);
1440 ecp_nistz256_ord_mul_mont(out, out, one);
1443 * Can't fail, but check return code to be consistent anyway.
1445 if (!bn_set_words(r, out, P256_LIMBS))
1453 # define ecp_nistz256_inv_mod_ord NULL
1456 const EC_METHOD *EC_GFp_nistz256_method(void)
1458 static const EC_METHOD ret = {
1459 EC_FLAGS_DEFAULT_OCT,
1460 NID_X9_62_prime_field,
1461 ossl_ec_GFp_mont_group_init,
1462 ossl_ec_GFp_mont_group_finish,
1463 ossl_ec_GFp_mont_group_clear_finish,
1464 ossl_ec_GFp_mont_group_copy,
1465 ossl_ec_GFp_mont_group_set_curve,
1466 ossl_ec_GFp_simple_group_get_curve,
1467 ossl_ec_GFp_simple_group_get_degree,
1468 ossl_ec_group_simple_order_bits,
1469 ossl_ec_GFp_simple_group_check_discriminant,
1470 ossl_ec_GFp_simple_point_init,
1471 ossl_ec_GFp_simple_point_finish,
1472 ossl_ec_GFp_simple_point_clear_finish,
1473 ossl_ec_GFp_simple_point_copy,
1474 ossl_ec_GFp_simple_point_set_to_infinity,
1475 ossl_ec_GFp_simple_point_set_affine_coordinates,
1476 ecp_nistz256_get_affine,
1478 ossl_ec_GFp_simple_add,
1479 ossl_ec_GFp_simple_dbl,
1480 ossl_ec_GFp_simple_invert,
1481 ossl_ec_GFp_simple_is_at_infinity,
1482 ossl_ec_GFp_simple_is_on_curve,
1483 ossl_ec_GFp_simple_cmp,
1484 ossl_ec_GFp_simple_make_affine,
1485 ossl_ec_GFp_simple_points_make_affine,
1486 ecp_nistz256_points_mul, /* mul */
1487 ecp_nistz256_mult_precompute, /* precompute_mult */
1488 ecp_nistz256_window_have_precompute_mult, /* have_precompute_mult */
1489 ossl_ec_GFp_mont_field_mul,
1490 ossl_ec_GFp_mont_field_sqr,
1492 ossl_ec_GFp_mont_field_inv,
1493 ossl_ec_GFp_mont_field_encode,
1494 ossl_ec_GFp_mont_field_decode,
1495 ossl_ec_GFp_mont_field_set_to_one,
1496 ossl_ec_key_simple_priv2oct,
1497 ossl_ec_key_simple_oct2priv,
1498 0, /* set private */
1499 ossl_ec_key_simple_generate_key,
1500 ossl_ec_key_simple_check_key,
1501 ossl_ec_key_simple_generate_public_key,
1504 ossl_ecdh_simple_compute_key,
1505 ossl_ecdsa_simple_sign_setup,
1506 ossl_ecdsa_simple_sign_sig,
1507 ossl_ecdsa_simple_verify_sig,
1508 ecp_nistz256_inv_mod_ord, /* can be #define-d NULL */
1509 0, /* blind_coordinates */
1511 0, /* ladder_step */