xref: /freebsd/crypto/openssl/crypto/ec/ecp_nistz256.c (revision f25b8c9fb4f58cf61adb47d7570abe7caa6d385d)
1 /*
2  * Copyright 2014-2025 The OpenSSL Project Authors. All Rights Reserved.
3  * Copyright (c) 2014, Intel Corporation. All Rights Reserved.
4  * Copyright (c) 2015, CloudFlare, Inc.
5  *
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
10  *
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.
15  *
16  * Reference:
17  * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with
18  *                          256 Bit Primes"
19  */
20 
21 /*
22  * ECDSA low level APIs are deprecated for public use, but still ok for
23  * internal use.
24  */
25 #include "internal/deprecated.h"
26 
27 #include <string.h>
28 
29 #include "internal/cryptlib.h"
30 #include "crypto/bn.h"
31 #include "ec_local.h"
32 #include "internal/refcount.h"
33 
34 #if BN_BITS2 != 64
35 #define TOBN(hi, lo) lo, hi
36 #else
37 #define TOBN(hi, lo) ((BN_ULONG)hi << 32 | lo)
38 #endif
39 
40 #define ALIGNPTR(p, N) ((unsigned char *)p + N - (size_t)p % N)
41 #define P256_LIMBS (256 / BN_BITS2)
42 
43 typedef unsigned short u16;
44 
45 typedef struct {
46     BN_ULONG X[P256_LIMBS];
47     BN_ULONG Y[P256_LIMBS];
48     BN_ULONG Z[P256_LIMBS];
49 } P256_POINT;
50 
51 typedef struct {
52     BN_ULONG X[P256_LIMBS];
53     BN_ULONG Y[P256_LIMBS];
54 } P256_POINT_AFFINE;
55 
56 typedef P256_POINT_AFFINE PRECOMP256_ROW[64];
57 
58 /* structure for precomputed multiples of the generator */
59 struct nistz256_pre_comp_st {
60     const EC_GROUP *group; /* Parent EC_GROUP object */
61     size_t w; /* Window size */
62     /*
63      * Constant time access to the X and Y coordinates of the pre-computed,
64      * generator multiplies, in the Montgomery domain. Pre-calculated
65      * multiplies are stored in affine form.
66      */
67     PRECOMP256_ROW *precomp;
68     void *precomp_storage;
69     CRYPTO_REF_COUNT references;
70 };
71 
72 /* Functions implemented in assembly */
73 /*
74  * Most of below mentioned functions *preserve* the property of inputs
75  * being fully reduced, i.e. being in [0, modulus) range. Simply put if
76  * inputs are fully reduced, then output is too. Note that reverse is
77  * not true, in sense that given partially reduced inputs output can be
78  * either, not unlikely reduced. And "most" in first sentence refers to
79  * the fact that given the calculations flow one can tolerate that
80  * addition, 1st function below, produces partially reduced result *if*
81  * multiplications by 2 and 3, which customarily use addition, fully
82  * reduce it. This effectively gives two options: a) addition produces
83  * fully reduced result [as long as inputs are, just like remaining
84  * functions]; b) addition is allowed to produce partially reduced
85  * result, but multiplications by 2 and 3 perform additional reduction
86  * step. Choice between the two can be platform-specific, but it was a)
87  * in all cases so far...
88  */
89 /* Modular add: res = a+b mod P   */
90 void ecp_nistz256_add(BN_ULONG res[P256_LIMBS],
91     const BN_ULONG a[P256_LIMBS],
92     const BN_ULONG b[P256_LIMBS]);
93 /* Modular mul by 2: res = 2*a mod P */
94 void ecp_nistz256_mul_by_2(BN_ULONG res[P256_LIMBS],
95     const BN_ULONG a[P256_LIMBS]);
96 /* Modular mul by 3: res = 3*a mod P */
97 void ecp_nistz256_mul_by_3(BN_ULONG res[P256_LIMBS],
98     const BN_ULONG a[P256_LIMBS]);
99 
100 /* Modular div by 2: res = a/2 mod P */
101 void ecp_nistz256_div_by_2(BN_ULONG res[P256_LIMBS],
102     const BN_ULONG a[P256_LIMBS]);
103 /* Modular sub: res = a-b mod P   */
104 void ecp_nistz256_sub(BN_ULONG res[P256_LIMBS],
105     const BN_ULONG a[P256_LIMBS],
106     const BN_ULONG b[P256_LIMBS]);
107 /* Modular neg: res = -a mod P    */
108 void ecp_nistz256_neg(BN_ULONG res[P256_LIMBS], const BN_ULONG a[P256_LIMBS]);
109 /* Montgomery mul: res = a*b*2^-256 mod P */
110 void ecp_nistz256_mul_mont(BN_ULONG res[P256_LIMBS],
111     const BN_ULONG a[P256_LIMBS],
112     const BN_ULONG b[P256_LIMBS]);
113 /* Montgomery sqr: res = a*a*2^-256 mod P */
114 void ecp_nistz256_sqr_mont(BN_ULONG res[P256_LIMBS],
115     const BN_ULONG a[P256_LIMBS]);
116 /* Convert a number from Montgomery domain, by multiplying with 1 */
117 void ecp_nistz256_from_mont(BN_ULONG res[P256_LIMBS],
118     const BN_ULONG in[P256_LIMBS]);
119 /* Convert a number to Montgomery domain, by multiplying with 2^512 mod P*/
120 void ecp_nistz256_to_mont(BN_ULONG res[P256_LIMBS],
121     const BN_ULONG in[P256_LIMBS]);
122 /* Functions that perform constant time access to the precomputed tables */
123 void ecp_nistz256_scatter_w5(P256_POINT *val,
124     const P256_POINT *in_t, int idx);
125 void ecp_nistz256_gather_w5(P256_POINT *val,
126     const P256_POINT *in_t, int idx);
127 void ecp_nistz256_scatter_w7(P256_POINT_AFFINE *val,
128     const P256_POINT_AFFINE *in_t, int idx);
129 void ecp_nistz256_gather_w7(P256_POINT_AFFINE *val,
130     const P256_POINT_AFFINE *in_t, int idx);
131 
132 /* One converted into the Montgomery domain */
133 static const BN_ULONG ONE[P256_LIMBS] = {
134     TOBN(0x00000000, 0x00000001), TOBN(0xffffffff, 0x00000000),
135     TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe)
136 };
137 
138 static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group);
139 
140 /* Precomputed tables for the default generator */
141 extern const PRECOMP256_ROW ecp_nistz256_precomputed[37];
142 
143 /* Recode window to a signed digit, see ecp_nistputil.c for details */
_booth_recode_w5(unsigned int in)144 static unsigned int _booth_recode_w5(unsigned int in)
145 {
146     unsigned int s, d;
147 
148     s = ~((in >> 5) - 1);
149     d = (1 << 6) - in - 1;
150     d = (d & s) | (in & ~s);
151     d = (d >> 1) + (d & 1);
152 
153     return (d << 1) + (s & 1);
154 }
155 
_booth_recode_w7(unsigned int in)156 static unsigned int _booth_recode_w7(unsigned int in)
157 {
158     unsigned int s, d;
159 
160     s = ~((in >> 7) - 1);
161     d = (1 << 8) - in - 1;
162     d = (d & s) | (in & ~s);
163     d = (d >> 1) + (d & 1);
164 
165     return (d << 1) + (s & 1);
166 }
167 
copy_conditional(BN_ULONG dst[P256_LIMBS],const BN_ULONG src[P256_LIMBS],BN_ULONG move)168 static void copy_conditional(BN_ULONG dst[P256_LIMBS],
169     const BN_ULONG src[P256_LIMBS], BN_ULONG move)
170 {
171     BN_ULONG mask1 = 0 - move;
172     BN_ULONG mask2 = ~mask1;
173 
174     dst[0] = (src[0] & mask1) ^ (dst[0] & mask2);
175     dst[1] = (src[1] & mask1) ^ (dst[1] & mask2);
176     dst[2] = (src[2] & mask1) ^ (dst[2] & mask2);
177     dst[3] = (src[3] & mask1) ^ (dst[3] & mask2);
178     if (P256_LIMBS == 8) {
179         dst[4] = (src[4] & mask1) ^ (dst[4] & mask2);
180         dst[5] = (src[5] & mask1) ^ (dst[5] & mask2);
181         dst[6] = (src[6] & mask1) ^ (dst[6] & mask2);
182         dst[7] = (src[7] & mask1) ^ (dst[7] & mask2);
183     }
184 }
185 
is_zero(BN_ULONG in)186 static BN_ULONG is_zero(BN_ULONG in)
187 {
188     in |= (0 - in);
189     in = ~in;
190     in >>= BN_BITS2 - 1;
191     return in;
192 }
193 
is_equal(const BN_ULONG a[P256_LIMBS],const BN_ULONG b[P256_LIMBS])194 static BN_ULONG is_equal(const BN_ULONG a[P256_LIMBS],
195     const BN_ULONG b[P256_LIMBS])
196 {
197     BN_ULONG res;
198 
199     res = a[0] ^ b[0];
200     res |= a[1] ^ b[1];
201     res |= a[2] ^ b[2];
202     res |= a[3] ^ b[3];
203     if (P256_LIMBS == 8) {
204         res |= a[4] ^ b[4];
205         res |= a[5] ^ b[5];
206         res |= a[6] ^ b[6];
207         res |= a[7] ^ b[7];
208     }
209 
210     return is_zero(res);
211 }
212 
is_one(const BIGNUM * z)213 static BN_ULONG is_one(const BIGNUM *z)
214 {
215     BN_ULONG res = 0;
216     BN_ULONG *a = bn_get_words(z);
217 
218     if (bn_get_top(z) == (P256_LIMBS - P256_LIMBS / 8)) {
219         res = a[0] ^ ONE[0];
220         res |= a[1] ^ ONE[1];
221         res |= a[2] ^ ONE[2];
222         res |= a[3] ^ ONE[3];
223         if (P256_LIMBS == 8) {
224             res |= a[4] ^ ONE[4];
225             res |= a[5] ^ ONE[5];
226             res |= a[6] ^ ONE[6];
227             /*
228              * no check for a[7] (being zero) on 32-bit platforms,
229              * because value of "one" takes only 7 limbs.
230              */
231         }
232         res = is_zero(res);
233     }
234 
235     return res;
236 }
237 
238 /*
239  * For reference, this macro is used only when new ecp_nistz256 assembly
240  * module is being developed.  For example, configure with
241  * -DECP_NISTZ256_REFERENCE_IMPLEMENTATION and implement only functions
242  * performing simplest arithmetic operations on 256-bit vectors. Then
243  * work on implementation of higher-level functions performing point
244  * operations. Then remove ECP_NISTZ256_REFERENCE_IMPLEMENTATION
245  * and never define it again. (The correct macro denoting presence of
246  * ecp_nistz256 module is ECP_NISTZ256_ASM.)
247  */
248 #ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
249 void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a);
250 void ecp_nistz256_point_add(P256_POINT *r,
251     const P256_POINT *a, const P256_POINT *b);
252 void ecp_nistz256_point_add_affine(P256_POINT *r,
253     const P256_POINT *a,
254     const P256_POINT_AFFINE *b);
255 #else
256 /* Point double: r = 2*a */
ecp_nistz256_point_double(P256_POINT * r,const P256_POINT * a)257 static void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a)
258 {
259     BN_ULONG S[P256_LIMBS];
260     BN_ULONG M[P256_LIMBS];
261     BN_ULONG Zsqr[P256_LIMBS];
262     BN_ULONG tmp0[P256_LIMBS];
263 
264     const BN_ULONG *in_x = a->X;
265     const BN_ULONG *in_y = a->Y;
266     const BN_ULONG *in_z = a->Z;
267 
268     BN_ULONG *res_x = r->X;
269     BN_ULONG *res_y = r->Y;
270     BN_ULONG *res_z = r->Z;
271 
272     ecp_nistz256_mul_by_2(S, in_y);
273 
274     ecp_nistz256_sqr_mont(Zsqr, in_z);
275 
276     ecp_nistz256_sqr_mont(S, S);
277 
278     ecp_nistz256_mul_mont(res_z, in_z, in_y);
279     ecp_nistz256_mul_by_2(res_z, res_z);
280 
281     ecp_nistz256_add(M, in_x, Zsqr);
282     ecp_nistz256_sub(Zsqr, in_x, Zsqr);
283 
284     ecp_nistz256_sqr_mont(res_y, S);
285     ecp_nistz256_div_by_2(res_y, res_y);
286 
287     ecp_nistz256_mul_mont(M, M, Zsqr);
288     ecp_nistz256_mul_by_3(M, M);
289 
290     ecp_nistz256_mul_mont(S, S, in_x);
291     ecp_nistz256_mul_by_2(tmp0, S);
292 
293     ecp_nistz256_sqr_mont(res_x, M);
294 
295     ecp_nistz256_sub(res_x, res_x, tmp0);
296     ecp_nistz256_sub(S, S, res_x);
297 
298     ecp_nistz256_mul_mont(S, S, M);
299     ecp_nistz256_sub(res_y, S, res_y);
300 }
301 
302 /* Point addition: r = a+b */
ecp_nistz256_point_add(P256_POINT * r,const P256_POINT * a,const P256_POINT * b)303 static void ecp_nistz256_point_add(P256_POINT *r,
304     const P256_POINT *a, const P256_POINT *b)
305 {
306     BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
307     BN_ULONG U1[P256_LIMBS], S1[P256_LIMBS];
308     BN_ULONG Z1sqr[P256_LIMBS];
309     BN_ULONG Z2sqr[P256_LIMBS];
310     BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
311     BN_ULONG Hsqr[P256_LIMBS];
312     BN_ULONG Rsqr[P256_LIMBS];
313     BN_ULONG Hcub[P256_LIMBS];
314 
315     BN_ULONG res_x[P256_LIMBS];
316     BN_ULONG res_y[P256_LIMBS];
317     BN_ULONG res_z[P256_LIMBS];
318 
319     BN_ULONG in1infty, in2infty;
320 
321     const BN_ULONG *in1_x = a->X;
322     const BN_ULONG *in1_y = a->Y;
323     const BN_ULONG *in1_z = a->Z;
324 
325     const BN_ULONG *in2_x = b->X;
326     const BN_ULONG *in2_y = b->Y;
327     const BN_ULONG *in2_z = b->Z;
328 
329     /*
330      * Infinity in encoded as (,,0)
331      */
332     in1infty = (in1_z[0] | in1_z[1] | in1_z[2] | in1_z[3]);
333     if (P256_LIMBS == 8)
334         in1infty |= (in1_z[4] | in1_z[5] | in1_z[6] | in1_z[7]);
335 
336     in2infty = (in2_z[0] | in2_z[1] | in2_z[2] | in2_z[3]);
337     if (P256_LIMBS == 8)
338         in2infty |= (in2_z[4] | in2_z[5] | in2_z[6] | in2_z[7]);
339 
340     in1infty = is_zero(in1infty);
341     in2infty = is_zero(in2infty);
342 
343     ecp_nistz256_sqr_mont(Z2sqr, in2_z); /* Z2^2 */
344     ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
345 
346     ecp_nistz256_mul_mont(S1, Z2sqr, in2_z); /* S1 = Z2^3 */
347     ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
348 
349     ecp_nistz256_mul_mont(S1, S1, in1_y); /* S1 = Y1*Z2^3 */
350     ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
351     ecp_nistz256_sub(R, S2, S1); /* R = S2 - S1 */
352 
353     ecp_nistz256_mul_mont(U1, in1_x, Z2sqr); /* U1 = X1*Z2^2 */
354     ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
355     ecp_nistz256_sub(H, U2, U1); /* H = U2 - U1 */
356 
357     /*
358      * The formulae are incorrect if the points are equal so we check for
359      * this and do doubling if this happens.
360      *
361      * Points here are in Jacobian projective coordinates (Xi, Yi, Zi)
362      * that are bound to the affine coordinates (xi, yi) by the following
363      * equations:
364      *     - xi = Xi / (Zi)^2
365      *     - y1 = Yi / (Zi)^3
366      *
367      * For the sake of optimization, the algorithm operates over
368      * intermediate variables U1, U2 and S1, S2 that are derived from
369      * the projective coordinates:
370      *     - U1 = X1 * (Z2)^2 ; U2 = X2 * (Z1)^2
371      *     - S1 = Y1 * (Z2)^3 ; S2 = Y2 * (Z1)^3
372      *
373      * It is easy to prove that is_equal(U1, U2) implies that the affine
374      * x-coordinates are equal, or either point is at infinity.
375      * Likewise is_equal(S1, S2) implies that the affine y-coordinates are
376      * equal, or either point is at infinity.
377      *
378      * The special case of either point being the point at infinity (Z1 or Z2
379      * is zero), is handled separately later on in this function, so we avoid
380      * jumping to point_double here in those special cases.
381      *
382      * When both points are inverse of each other, we know that the affine
383      * x-coordinates are equal, and the y-coordinates have different sign.
384      * Therefore since U1 = U2, we know H = 0, and therefore Z3 = H*Z1*Z2
385      * will equal 0, thus the result is infinity, if we simply let this
386      * function continue normally.
387      *
388      * We use bitwise operations to avoid potential side-channels introduced by
389      * the short-circuiting behaviour of boolean operators.
390      */
391     if (is_equal(U1, U2) & ~in1infty & ~in2infty & is_equal(S1, S2)) {
392         /*
393          * This is obviously not constant-time but it should never happen during
394          * single point multiplication, so there is no timing leak for ECDH or
395          * ECDSA signing.
396          */
397         ecp_nistz256_point_double(r, a);
398         return;
399     }
400 
401     ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
402     ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
403     ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
404     ecp_nistz256_mul_mont(res_z, res_z, in2_z); /* Z3 = H*Z1*Z2 */
405     ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
406 
407     ecp_nistz256_mul_mont(U2, U1, Hsqr); /* U1*H^2 */
408     ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
409 
410     ecp_nistz256_sub(res_x, Rsqr, Hsqr);
411     ecp_nistz256_sub(res_x, res_x, Hcub);
412 
413     ecp_nistz256_sub(res_y, U2, res_x);
414 
415     ecp_nistz256_mul_mont(S2, S1, Hcub);
416     ecp_nistz256_mul_mont(res_y, R, res_y);
417     ecp_nistz256_sub(res_y, res_y, S2);
418 
419     copy_conditional(res_x, in2_x, in1infty);
420     copy_conditional(res_y, in2_y, in1infty);
421     copy_conditional(res_z, in2_z, in1infty);
422 
423     copy_conditional(res_x, in1_x, in2infty);
424     copy_conditional(res_y, in1_y, in2infty);
425     copy_conditional(res_z, in1_z, in2infty);
426 
427     memcpy(r->X, res_x, sizeof(res_x));
428     memcpy(r->Y, res_y, sizeof(res_y));
429     memcpy(r->Z, res_z, sizeof(res_z));
430 }
431 
432 /* Point addition when b is known to be affine: r = a+b */
ecp_nistz256_point_add_affine(P256_POINT * r,const P256_POINT * a,const P256_POINT_AFFINE * b)433 static void ecp_nistz256_point_add_affine(P256_POINT *r,
434     const P256_POINT *a,
435     const P256_POINT_AFFINE *b)
436 {
437     BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
438     BN_ULONG Z1sqr[P256_LIMBS];
439     BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
440     BN_ULONG Hsqr[P256_LIMBS];
441     BN_ULONG Rsqr[P256_LIMBS];
442     BN_ULONG Hcub[P256_LIMBS];
443 
444     BN_ULONG res_x[P256_LIMBS];
445     BN_ULONG res_y[P256_LIMBS];
446     BN_ULONG res_z[P256_LIMBS];
447 
448     BN_ULONG in1infty, in2infty;
449 
450     const BN_ULONG *in1_x = a->X;
451     const BN_ULONG *in1_y = a->Y;
452     const BN_ULONG *in1_z = a->Z;
453 
454     const BN_ULONG *in2_x = b->X;
455     const BN_ULONG *in2_y = b->Y;
456 
457     /*
458      * Infinity in encoded as (,,0)
459      */
460     in1infty = (in1_z[0] | in1_z[1] | in1_z[2] | in1_z[3]);
461     if (P256_LIMBS == 8)
462         in1infty |= (in1_z[4] | in1_z[5] | in1_z[6] | in1_z[7]);
463 
464     /*
465      * In affine representation we encode infinity as (0,0), which is
466      * not on the curve, so it is OK
467      */
468     in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] | in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
469     if (P256_LIMBS == 8)
470         in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] | in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
471 
472     in1infty = is_zero(in1infty);
473     in2infty = is_zero(in2infty);
474 
475     ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
476 
477     ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
478     ecp_nistz256_sub(H, U2, in1_x); /* H = U2 - U1 */
479 
480     ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
481 
482     ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
483 
484     ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
485     ecp_nistz256_sub(R, S2, in1_y); /* R = S2 - S1 */
486 
487     ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
488     ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
489     ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
490 
491     ecp_nistz256_mul_mont(U2, in1_x, Hsqr); /* U1*H^2 */
492     ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
493 
494     ecp_nistz256_sub(res_x, Rsqr, Hsqr);
495     ecp_nistz256_sub(res_x, res_x, Hcub);
496     ecp_nistz256_sub(H, U2, res_x);
497 
498     ecp_nistz256_mul_mont(S2, in1_y, Hcub);
499     ecp_nistz256_mul_mont(H, H, R);
500     ecp_nistz256_sub(res_y, H, S2);
501 
502     copy_conditional(res_x, in2_x, in1infty);
503     copy_conditional(res_x, in1_x, in2infty);
504 
505     copy_conditional(res_y, in2_y, in1infty);
506     copy_conditional(res_y, in1_y, in2infty);
507 
508     copy_conditional(res_z, ONE, in1infty);
509     copy_conditional(res_z, in1_z, in2infty);
510 
511     memcpy(r->X, res_x, sizeof(res_x));
512     memcpy(r->Y, res_y, sizeof(res_y));
513     memcpy(r->Z, res_z, sizeof(res_z));
514 }
515 #endif
516 
517 /* r = in^-1 mod p */
ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS],const BN_ULONG in[P256_LIMBS])518 static void ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS],
519     const BN_ULONG in[P256_LIMBS])
520 {
521     /*
522      * The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff
523      * ffffffff ffffffff We use FLT and used poly-2 as exponent
524      */
525     BN_ULONG p2[P256_LIMBS];
526     BN_ULONG p4[P256_LIMBS];
527     BN_ULONG p8[P256_LIMBS];
528     BN_ULONG p16[P256_LIMBS];
529     BN_ULONG p32[P256_LIMBS];
530     BN_ULONG res[P256_LIMBS];
531     int i;
532 
533     ecp_nistz256_sqr_mont(res, in);
534     ecp_nistz256_mul_mont(p2, res, in); /* 3*p */
535 
536     ecp_nistz256_sqr_mont(res, p2);
537     ecp_nistz256_sqr_mont(res, res);
538     ecp_nistz256_mul_mont(p4, res, p2); /* f*p */
539 
540     ecp_nistz256_sqr_mont(res, p4);
541     ecp_nistz256_sqr_mont(res, res);
542     ecp_nistz256_sqr_mont(res, res);
543     ecp_nistz256_sqr_mont(res, res);
544     ecp_nistz256_mul_mont(p8, res, p4); /* ff*p */
545 
546     ecp_nistz256_sqr_mont(res, p8);
547     for (i = 0; i < 7; i++)
548         ecp_nistz256_sqr_mont(res, res);
549     ecp_nistz256_mul_mont(p16, res, p8); /* ffff*p */
550 
551     ecp_nistz256_sqr_mont(res, p16);
552     for (i = 0; i < 15; i++)
553         ecp_nistz256_sqr_mont(res, res);
554     ecp_nistz256_mul_mont(p32, res, p16); /* ffffffff*p */
555 
556     ecp_nistz256_sqr_mont(res, p32);
557     for (i = 0; i < 31; i++)
558         ecp_nistz256_sqr_mont(res, res);
559     ecp_nistz256_mul_mont(res, res, in);
560 
561     for (i = 0; i < 32 * 4; i++)
562         ecp_nistz256_sqr_mont(res, res);
563     ecp_nistz256_mul_mont(res, res, p32);
564 
565     for (i = 0; i < 32; i++)
566         ecp_nistz256_sqr_mont(res, res);
567     ecp_nistz256_mul_mont(res, res, p32);
568 
569     for (i = 0; i < 16; i++)
570         ecp_nistz256_sqr_mont(res, res);
571     ecp_nistz256_mul_mont(res, res, p16);
572 
573     for (i = 0; i < 8; i++)
574         ecp_nistz256_sqr_mont(res, res);
575     ecp_nistz256_mul_mont(res, res, p8);
576 
577     ecp_nistz256_sqr_mont(res, res);
578     ecp_nistz256_sqr_mont(res, res);
579     ecp_nistz256_sqr_mont(res, res);
580     ecp_nistz256_sqr_mont(res, res);
581     ecp_nistz256_mul_mont(res, res, p4);
582 
583     ecp_nistz256_sqr_mont(res, res);
584     ecp_nistz256_sqr_mont(res, res);
585     ecp_nistz256_mul_mont(res, res, p2);
586 
587     ecp_nistz256_sqr_mont(res, res);
588     ecp_nistz256_sqr_mont(res, res);
589     ecp_nistz256_mul_mont(res, res, in);
590 
591     memcpy(r, res, sizeof(res));
592 }
593 
594 /*
595  * ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
596  * returns one if it fits. Otherwise it returns zero.
597  */
ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],const BIGNUM * in)598 __owur static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],
599     const BIGNUM *in)
600 {
601     return bn_copy_words(out, in, P256_LIMBS);
602 }
603 
604 /* r = sum(scalar[i]*point[i]) */
ecp_nistz256_windowed_mul(const EC_GROUP * group,P256_POINT * r,const BIGNUM ** scalar,const EC_POINT ** point,size_t num,BN_CTX * ctx)605 __owur static int ecp_nistz256_windowed_mul(const EC_GROUP *group,
606     P256_POINT *r,
607     const BIGNUM **scalar,
608     const EC_POINT **point,
609     size_t num, BN_CTX *ctx)
610 {
611     size_t i;
612     int j, ret = 0;
613     unsigned int idx;
614     unsigned char (*p_str)[33] = NULL;
615     const unsigned int window_size = 5;
616     const unsigned int mask = (1 << (window_size + 1)) - 1;
617     unsigned int wvalue;
618     P256_POINT *temp; /* place for 5 temporary points */
619     const BIGNUM **scalars = NULL;
620     P256_POINT(*table)
621     [16] = NULL;
622     void *table_storage = NULL;
623 
624     if ((num * 16 + 6) > OPENSSL_MALLOC_MAX_NELEMS(P256_POINT)
625         || (table_storage = OPENSSL_malloc((num * 16 + 5) * sizeof(P256_POINT) + 64)) == NULL
626         || (p_str = OPENSSL_malloc(num * 33 * sizeof(unsigned char))) == NULL
627         || (scalars = OPENSSL_malloc(num * sizeof(BIGNUM *))) == NULL)
628         goto err;
629 
630     table = (void *)ALIGNPTR(table_storage, 64);
631     temp = (P256_POINT *)(table + num);
632 
633     for (i = 0; i < num; i++) {
634         P256_POINT *row = table[i];
635 
636         /* This is an unusual input, we don't guarantee constant-timeness. */
637         if ((BN_num_bits(scalar[i]) > 256) || BN_is_negative(scalar[i])) {
638             BIGNUM *mod;
639 
640             if ((mod = BN_CTX_get(ctx)) == NULL)
641                 goto err;
642             if (!BN_nnmod(mod, scalar[i], group->order, ctx)) {
643                 ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
644                 goto err;
645             }
646             scalars[i] = mod;
647         } else
648             scalars[i] = scalar[i];
649 
650         for (j = 0; j < bn_get_top(scalars[i]) * BN_BYTES; j += BN_BYTES) {
651             BN_ULONG d = bn_get_words(scalars[i])[j / BN_BYTES];
652 
653             p_str[i][j + 0] = (unsigned char)d;
654             p_str[i][j + 1] = (unsigned char)(d >> 8);
655             p_str[i][j + 2] = (unsigned char)(d >> 16);
656             p_str[i][j + 3] = (unsigned char)(d >>= 24);
657             if (BN_BYTES == 8) {
658                 d >>= 8;
659                 p_str[i][j + 4] = (unsigned char)d;
660                 p_str[i][j + 5] = (unsigned char)(d >> 8);
661                 p_str[i][j + 6] = (unsigned char)(d >> 16);
662                 p_str[i][j + 7] = (unsigned char)(d >> 24);
663             }
664         }
665         for (; j < 33; j++)
666             p_str[i][j] = 0;
667 
668         if (!ecp_nistz256_bignum_to_field_elem(temp[0].X, point[i]->X)
669             || !ecp_nistz256_bignum_to_field_elem(temp[0].Y, point[i]->Y)
670             || !ecp_nistz256_bignum_to_field_elem(temp[0].Z, point[i]->Z)) {
671             ERR_raise(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
672             goto err;
673         }
674 
675         /*
676          * row[0] is implicitly (0,0,0) (the point at infinity), therefore it
677          * is not stored. All other values are actually stored with an offset
678          * of -1 in table.
679          */
680 
681         ecp_nistz256_scatter_w5(row, &temp[0], 1);
682         ecp_nistz256_point_double(&temp[1], &temp[0]); /*1+1=2  */
683         ecp_nistz256_scatter_w5(row, &temp[1], 2);
684         ecp_nistz256_point_add(&temp[2], &temp[1], &temp[0]); /*2+1=3  */
685         ecp_nistz256_scatter_w5(row, &temp[2], 3);
686         ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*2=4  */
687         ecp_nistz256_scatter_w5(row, &temp[1], 4);
688         ecp_nistz256_point_double(&temp[2], &temp[2]); /*2*3=6  */
689         ecp_nistz256_scatter_w5(row, &temp[2], 6);
690         ecp_nistz256_point_add(&temp[3], &temp[1], &temp[0]); /*4+1=5  */
691         ecp_nistz256_scatter_w5(row, &temp[3], 5);
692         ecp_nistz256_point_add(&temp[4], &temp[2], &temp[0]); /*6+1=7  */
693         ecp_nistz256_scatter_w5(row, &temp[4], 7);
694         ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*4=8  */
695         ecp_nistz256_scatter_w5(row, &temp[1], 8);
696         ecp_nistz256_point_double(&temp[2], &temp[2]); /*2*6=12 */
697         ecp_nistz256_scatter_w5(row, &temp[2], 12);
698         ecp_nistz256_point_double(&temp[3], &temp[3]); /*2*5=10 */
699         ecp_nistz256_scatter_w5(row, &temp[3], 10);
700         ecp_nistz256_point_double(&temp[4], &temp[4]); /*2*7=14 */
701         ecp_nistz256_scatter_w5(row, &temp[4], 14);
702         ecp_nistz256_point_add(&temp[2], &temp[2], &temp[0]); /*12+1=13*/
703         ecp_nistz256_scatter_w5(row, &temp[2], 13);
704         ecp_nistz256_point_add(&temp[3], &temp[3], &temp[0]); /*10+1=11*/
705         ecp_nistz256_scatter_w5(row, &temp[3], 11);
706         ecp_nistz256_point_add(&temp[4], &temp[4], &temp[0]); /*14+1=15*/
707         ecp_nistz256_scatter_w5(row, &temp[4], 15);
708         ecp_nistz256_point_add(&temp[2], &temp[1], &temp[0]); /*8+1=9  */
709         ecp_nistz256_scatter_w5(row, &temp[2], 9);
710         ecp_nistz256_point_double(&temp[1], &temp[1]); /*2*8=16 */
711         ecp_nistz256_scatter_w5(row, &temp[1], 16);
712     }
713 
714     idx = 255;
715 
716     wvalue = p_str[0][(idx - 1) / 8];
717     wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
718 
719     /*
720      * We gather to temp[0], because we know it's position relative
721      * to table
722      */
723     ecp_nistz256_gather_w5(&temp[0], table[0], _booth_recode_w5(wvalue) >> 1);
724     memcpy(r, &temp[0], sizeof(temp[0]));
725 
726     while (idx >= 5) {
727         for (i = (idx == 255 ? 1 : 0); i < num; i++) {
728             unsigned int off = (idx - 1) / 8;
729 
730             wvalue = p_str[i][off] | p_str[i][off + 1] << 8;
731             wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
732 
733             wvalue = _booth_recode_w5(wvalue);
734 
735             ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1);
736 
737             ecp_nistz256_neg(temp[1].Y, temp[0].Y);
738             copy_conditional(temp[0].Y, temp[1].Y, (wvalue & 1));
739 
740             ecp_nistz256_point_add(r, r, &temp[0]);
741         }
742 
743         idx -= window_size;
744 
745         ecp_nistz256_point_double(r, r);
746         ecp_nistz256_point_double(r, r);
747         ecp_nistz256_point_double(r, r);
748         ecp_nistz256_point_double(r, r);
749         ecp_nistz256_point_double(r, r);
750     }
751 
752     /* Final window */
753     for (i = 0; i < num; i++) {
754         wvalue = p_str[i][0];
755         wvalue = (wvalue << 1) & mask;
756 
757         wvalue = _booth_recode_w5(wvalue);
758 
759         ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1);
760 
761         ecp_nistz256_neg(temp[1].Y, temp[0].Y);
762         copy_conditional(temp[0].Y, temp[1].Y, wvalue & 1);
763 
764         ecp_nistz256_point_add(r, r, &temp[0]);
765     }
766 
767     ret = 1;
768 err:
769     OPENSSL_free(table_storage);
770     OPENSSL_free(p_str);
771     OPENSSL_free(scalars);
772     return ret;
773 }
774 
775 /* Coordinates of G, for which we have precomputed tables */
776 static const BN_ULONG def_xG[P256_LIMBS] = {
777     TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
778     TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
779 };
780 
781 static const BN_ULONG def_yG[P256_LIMBS] = {
782     TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
783     TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
784 };
785 
786 /*
787  * ecp_nistz256_is_affine_G returns one if |generator| is the standard, P-256
788  * generator.
789  */
ecp_nistz256_is_affine_G(const EC_POINT * generator)790 static int ecp_nistz256_is_affine_G(const EC_POINT *generator)
791 {
792     return (bn_get_top(generator->X) == P256_LIMBS) && (bn_get_top(generator->Y) == P256_LIMBS) && is_equal(bn_get_words(generator->X), def_xG) && is_equal(bn_get_words(generator->Y), def_yG) && is_one(generator->Z);
793 }
794 
ecp_nistz256_mult_precompute(EC_GROUP * group,BN_CTX * ctx)795 __owur static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx)
796 {
797     /*
798      * We precompute a table for a Booth encoded exponent (wNAF) based
799      * computation. Each table holds 64 values for safe access, with an
800      * implicit value of infinity at index zero. We use window of size 7, and
801      * therefore require ceil(256/7) = 37 tables.
802      */
803     const BIGNUM *order;
804     EC_POINT *P = NULL, *T = NULL;
805     const EC_POINT *generator;
806     NISTZ256_PRE_COMP *pre_comp;
807     BN_CTX *new_ctx = NULL;
808     int i, j, k, ret = 0;
809     size_t w;
810 
811     PRECOMP256_ROW *preComputedTable = NULL;
812     unsigned char *precomp_storage = NULL;
813 
814     /* if there is an old NISTZ256_PRE_COMP object, throw it away */
815     EC_pre_comp_free(group);
816     generator = EC_GROUP_get0_generator(group);
817     if (generator == NULL) {
818         ERR_raise(ERR_LIB_EC, EC_R_UNDEFINED_GENERATOR);
819         return 0;
820     }
821 
822     if (ecp_nistz256_is_affine_G(generator)) {
823         /*
824          * No need to calculate tables for the standard generator because we
825          * have them statically.
826          */
827         return 1;
828     }
829 
830     if ((pre_comp = ecp_nistz256_pre_comp_new(group)) == NULL)
831         return 0;
832 
833     if (ctx == NULL) {
834         ctx = new_ctx = BN_CTX_new_ex(group->libctx);
835         if (ctx == NULL)
836             goto err;
837     }
838 
839     BN_CTX_start(ctx);
840 
841     order = EC_GROUP_get0_order(group);
842     if (order == NULL)
843         goto err;
844 
845     if (BN_is_zero(order)) {
846         ERR_raise(ERR_LIB_EC, EC_R_UNKNOWN_ORDER);
847         goto err;
848     }
849 
850     w = 7;
851 
852     if ((precomp_storage = OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE) + 64)) == NULL)
853         goto err;
854 
855     preComputedTable = (void *)ALIGNPTR(precomp_storage, 64);
856 
857     P = EC_POINT_new(group);
858     T = EC_POINT_new(group);
859     if (P == NULL || T == NULL)
860         goto err;
861 
862     /*
863      * The zero entry is implicitly infinity, and we skip it, storing other
864      * values with -1 offset.
865      */
866     if (!EC_POINT_copy(T, generator))
867         goto err;
868 
869     for (k = 0; k < 64; k++) {
870         if (!EC_POINT_copy(P, T))
871             goto err;
872         for (j = 0; j < 37; j++) {
873             P256_POINT_AFFINE temp;
874             /*
875              * It would be faster to use EC_POINTs_make_affine and
876              * make multiple points affine at the same time.
877              */
878             if (group->meth->make_affine == NULL
879                 || !group->meth->make_affine(group, P, ctx))
880                 goto err;
881             if (!ecp_nistz256_bignum_to_field_elem(temp.X, P->X) || !ecp_nistz256_bignum_to_field_elem(temp.Y, P->Y)) {
882                 ERR_raise(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
883                 goto err;
884             }
885             ecp_nistz256_scatter_w7(preComputedTable[j], &temp, k);
886             for (i = 0; i < 7; i++) {
887                 if (!EC_POINT_dbl(group, P, P, ctx))
888                     goto err;
889             }
890         }
891         if (!EC_POINT_add(group, T, T, generator, ctx))
892             goto err;
893     }
894 
895     pre_comp->group = group;
896     pre_comp->w = w;
897     pre_comp->precomp = preComputedTable;
898     pre_comp->precomp_storage = precomp_storage;
899     precomp_storage = NULL;
900     SETPRECOMP(group, nistz256, pre_comp);
901     pre_comp = NULL;
902     ret = 1;
903 
904 err:
905     BN_CTX_end(ctx);
906     BN_CTX_free(new_ctx);
907 
908     EC_nistz256_pre_comp_free(pre_comp);
909     OPENSSL_free(precomp_storage);
910     EC_POINT_free(P);
911     EC_POINT_free(T);
912     return ret;
913 }
914 
ecp_nistz256_set_from_affine(EC_POINT * out,const EC_GROUP * group,const P256_POINT_AFFINE * in,BN_CTX * ctx)915 __owur static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group,
916     const P256_POINT_AFFINE *in,
917     BN_CTX *ctx)
918 {
919     int ret = 0;
920 
921     if ((ret = bn_set_words(out->X, in->X, P256_LIMBS))
922         && (ret = bn_set_words(out->Y, in->Y, P256_LIMBS))
923         && (ret = bn_set_words(out->Z, ONE, P256_LIMBS)))
924         out->Z_is_one = 1;
925 
926     return ret;
927 }
928 
929 /* r = scalar*G + sum(scalars[i]*points[i]) */
ecp_nistz256_points_mul(const EC_GROUP * group,EC_POINT * r,const BIGNUM * scalar,size_t num,const EC_POINT * points[],const BIGNUM * scalars[],BN_CTX * ctx)930 __owur static int ecp_nistz256_points_mul(const EC_GROUP *group,
931     EC_POINT *r,
932     const BIGNUM *scalar,
933     size_t num,
934     const EC_POINT *points[],
935     const BIGNUM *scalars[], BN_CTX *ctx)
936 {
937     int i = 0, ret = 0, no_precomp_for_generator = 0, p_is_infinity = 0;
938     unsigned char p_str[33] = { 0 };
939     const PRECOMP256_ROW *preComputedTable = NULL;
940     const NISTZ256_PRE_COMP *pre_comp = NULL;
941     const EC_POINT *generator = NULL;
942     const BIGNUM **new_scalars = NULL;
943     const EC_POINT **new_points = NULL;
944     unsigned int idx = 0;
945     const unsigned int window_size = 7;
946     const unsigned int mask = (1 << (window_size + 1)) - 1;
947     unsigned int wvalue;
948     ALIGN32 union {
949         P256_POINT p;
950         P256_POINT_AFFINE a;
951     } t, p;
952     BIGNUM *tmp_scalar;
953 
954     if ((num + 1) == 0 || (num + 1) > OPENSSL_MALLOC_MAX_NELEMS(void *)) {
955         ERR_raise(ERR_LIB_EC, ERR_R_PASSED_INVALID_ARGUMENT);
956         return 0;
957     }
958 
959     memset(&p, 0, sizeof(p));
960     BN_CTX_start(ctx);
961 
962     if (scalar) {
963         generator = EC_GROUP_get0_generator(group);
964         if (generator == NULL) {
965             ERR_raise(ERR_LIB_EC, EC_R_UNDEFINED_GENERATOR);
966             goto err;
967         }
968 
969         /* look if we can use precomputed multiples of generator */
970         pre_comp = group->pre_comp.nistz256;
971 
972         if (pre_comp) {
973             /*
974              * If there is a precomputed table for the generator, check that
975              * it was generated with the same generator.
976              */
977             EC_POINT *pre_comp_generator = EC_POINT_new(group);
978             if (pre_comp_generator == NULL)
979                 goto err;
980 
981             ecp_nistz256_gather_w7(&p.a, pre_comp->precomp[0], 1);
982             if (!ecp_nistz256_set_from_affine(pre_comp_generator,
983                     group, &p.a, ctx)) {
984                 EC_POINT_free(pre_comp_generator);
985                 goto err;
986             }
987 
988             if (0 == EC_POINT_cmp(group, generator, pre_comp_generator, ctx))
989                 preComputedTable = (const PRECOMP256_ROW *)pre_comp->precomp;
990 
991             EC_POINT_free(pre_comp_generator);
992         }
993 
994         if (preComputedTable == NULL && ecp_nistz256_is_affine_G(generator)) {
995             /*
996              * If there is no precomputed data, but the generator is the
997              * default, a hardcoded table of precomputed data is used. This
998              * is because applications, such as Apache, do not use
999              * EC_KEY_precompute_mult.
1000              */
1001             preComputedTable = ecp_nistz256_precomputed;
1002         }
1003 
1004         if (preComputedTable) {
1005             BN_ULONG infty;
1006 
1007             if ((BN_num_bits(scalar) > 256)
1008                 || BN_is_negative(scalar)) {
1009                 if ((tmp_scalar = BN_CTX_get(ctx)) == NULL)
1010                     goto err;
1011 
1012                 if (!BN_nnmod(tmp_scalar, scalar, group->order, ctx)) {
1013                     ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1014                     goto err;
1015                 }
1016                 scalar = tmp_scalar;
1017             }
1018 
1019             for (i = 0; i < bn_get_top(scalar) * BN_BYTES; i += BN_BYTES) {
1020                 BN_ULONG d = bn_get_words(scalar)[i / BN_BYTES];
1021 
1022                 p_str[i + 0] = (unsigned char)d;
1023                 p_str[i + 1] = (unsigned char)(d >> 8);
1024                 p_str[i + 2] = (unsigned char)(d >> 16);
1025                 p_str[i + 3] = (unsigned char)(d >>= 24);
1026                 if (BN_BYTES == 8) {
1027                     d >>= 8;
1028                     p_str[i + 4] = (unsigned char)d;
1029                     p_str[i + 5] = (unsigned char)(d >> 8);
1030                     p_str[i + 6] = (unsigned char)(d >> 16);
1031                     p_str[i + 7] = (unsigned char)(d >> 24);
1032                 }
1033             }
1034 
1035             for (; i < 33; i++)
1036                 p_str[i] = 0;
1037 
1038             /* First window */
1039             wvalue = (p_str[0] << 1) & mask;
1040             idx += window_size;
1041 
1042             wvalue = _booth_recode_w7(wvalue);
1043 
1044             ecp_nistz256_gather_w7(&p.a, preComputedTable[0],
1045                 wvalue >> 1);
1046 
1047             ecp_nistz256_neg(p.p.Z, p.p.Y);
1048             copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
1049 
1050             /*
1051              * Since affine infinity is encoded as (0,0) and
1052              * Jacobian is (,,0), we need to harmonize them
1053              * by assigning "one" or zero to Z.
1054              */
1055             infty = (p.p.X[0] | p.p.X[1] | p.p.X[2] | p.p.X[3] | p.p.Y[0] | p.p.Y[1] | p.p.Y[2] | p.p.Y[3]);
1056             if (P256_LIMBS == 8)
1057                 infty |= (p.p.X[4] | p.p.X[5] | p.p.X[6] | p.p.X[7] | p.p.Y[4] | p.p.Y[5] | p.p.Y[6] | p.p.Y[7]);
1058 
1059             infty = 0 - is_zero(infty);
1060             infty = ~infty;
1061 
1062             p.p.Z[0] = ONE[0] & infty;
1063             p.p.Z[1] = ONE[1] & infty;
1064             p.p.Z[2] = ONE[2] & infty;
1065             p.p.Z[3] = ONE[3] & infty;
1066             if (P256_LIMBS == 8) {
1067                 p.p.Z[4] = ONE[4] & infty;
1068                 p.p.Z[5] = ONE[5] & infty;
1069                 p.p.Z[6] = ONE[6] & infty;
1070                 p.p.Z[7] = ONE[7] & infty;
1071             }
1072 
1073             for (i = 1; i < 37; i++) {
1074                 unsigned int off = (idx - 1) / 8;
1075                 wvalue = p_str[off] | p_str[off + 1] << 8;
1076                 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1077                 idx += window_size;
1078 
1079                 wvalue = _booth_recode_w7(wvalue);
1080 
1081                 ecp_nistz256_gather_w7(&t.a,
1082                     preComputedTable[i], wvalue >> 1);
1083 
1084                 ecp_nistz256_neg(t.p.Z, t.a.Y);
1085                 copy_conditional(t.a.Y, t.p.Z, wvalue & 1);
1086 
1087                 ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a);
1088             }
1089         } else {
1090             p_is_infinity = 1;
1091             no_precomp_for_generator = 1;
1092         }
1093     } else
1094         p_is_infinity = 1;
1095 
1096     if (no_precomp_for_generator) {
1097         /*
1098          * Without a precomputed table for the generator, it has to be
1099          * handled like a normal point.
1100          */
1101         new_scalars = OPENSSL_malloc((num + 1) * sizeof(BIGNUM *));
1102         if (new_scalars == NULL)
1103             goto err;
1104 
1105         new_points = OPENSSL_malloc((num + 1) * sizeof(EC_POINT *));
1106         if (new_points == NULL)
1107             goto err;
1108 
1109         memcpy(new_scalars, scalars, num * sizeof(BIGNUM *));
1110         new_scalars[num] = scalar;
1111         memcpy(new_points, points, num * sizeof(EC_POINT *));
1112         new_points[num] = generator;
1113 
1114         scalars = new_scalars;
1115         points = new_points;
1116         num++;
1117     }
1118 
1119     if (num) {
1120         P256_POINT *out = &t.p;
1121         if (p_is_infinity)
1122             out = &p.p;
1123 
1124         if (!ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx))
1125             goto err;
1126 
1127         if (!p_is_infinity)
1128             ecp_nistz256_point_add(&p.p, &p.p, out);
1129     }
1130 
1131     /* Not constant-time, but we're only operating on the public output. */
1132     if (!bn_set_words(r->X, p.p.X, P256_LIMBS) || !bn_set_words(r->Y, p.p.Y, P256_LIMBS) || !bn_set_words(r->Z, p.p.Z, P256_LIMBS)) {
1133         goto err;
1134     }
1135     r->Z_is_one = is_one(r->Z) & 1;
1136 
1137     ret = 1;
1138 
1139 err:
1140     BN_CTX_end(ctx);
1141     OPENSSL_free(new_points);
1142     OPENSSL_free(new_scalars);
1143     return ret;
1144 }
1145 
ecp_nistz256_get_affine(const EC_GROUP * group,const EC_POINT * point,BIGNUM * x,BIGNUM * y,BN_CTX * ctx)1146 __owur static int ecp_nistz256_get_affine(const EC_GROUP *group,
1147     const EC_POINT *point,
1148     BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
1149 {
1150     BN_ULONG z_inv2[P256_LIMBS];
1151     BN_ULONG z_inv3[P256_LIMBS];
1152     BN_ULONG x_aff[P256_LIMBS];
1153     BN_ULONG y_aff[P256_LIMBS];
1154     BN_ULONG point_x[P256_LIMBS], point_y[P256_LIMBS], point_z[P256_LIMBS];
1155     BN_ULONG x_ret[P256_LIMBS], y_ret[P256_LIMBS];
1156 
1157     if (EC_POINT_is_at_infinity(group, point)) {
1158         ERR_raise(ERR_LIB_EC, EC_R_POINT_AT_INFINITY);
1159         return 0;
1160     }
1161 
1162     if (!ecp_nistz256_bignum_to_field_elem(point_x, point->X) || !ecp_nistz256_bignum_to_field_elem(point_y, point->Y) || !ecp_nistz256_bignum_to_field_elem(point_z, point->Z)) {
1163         ERR_raise(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
1164         return 0;
1165     }
1166 
1167     ecp_nistz256_mod_inverse(z_inv3, point_z);
1168     ecp_nistz256_sqr_mont(z_inv2, z_inv3);
1169     ecp_nistz256_mul_mont(x_aff, z_inv2, point_x);
1170 
1171     if (x != NULL) {
1172         ecp_nistz256_from_mont(x_ret, x_aff);
1173         if (!bn_set_words(x, x_ret, P256_LIMBS))
1174             return 0;
1175     }
1176 
1177     if (y != NULL) {
1178         ecp_nistz256_mul_mont(z_inv3, z_inv3, z_inv2);
1179         ecp_nistz256_mul_mont(y_aff, z_inv3, point_y);
1180         ecp_nistz256_from_mont(y_ret, y_aff);
1181         if (!bn_set_words(y, y_ret, P256_LIMBS))
1182             return 0;
1183     }
1184 
1185     return 1;
1186 }
1187 
ecp_nistz256_pre_comp_new(const EC_GROUP * group)1188 static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group)
1189 {
1190     NISTZ256_PRE_COMP *ret = NULL;
1191 
1192     if (!group)
1193         return NULL;
1194 
1195     ret = OPENSSL_zalloc(sizeof(*ret));
1196 
1197     if (ret == NULL)
1198         return ret;
1199 
1200     ret->group = group;
1201     ret->w = 6; /* default */
1202 
1203     if (!CRYPTO_NEW_REF(&ret->references, 1)) {
1204         OPENSSL_free(ret);
1205         return NULL;
1206     }
1207     return ret;
1208 }
1209 
EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP * p)1210 NISTZ256_PRE_COMP *EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP *p)
1211 {
1212     int i;
1213     if (p != NULL)
1214         CRYPTO_UP_REF(&p->references, &i);
1215     return p;
1216 }
1217 
EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP * pre)1218 void EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP *pre)
1219 {
1220     int i;
1221 
1222     if (pre == NULL)
1223         return;
1224 
1225     CRYPTO_DOWN_REF(&pre->references, &i);
1226     REF_PRINT_COUNT("EC_nistz256", i, pre);
1227     if (i > 0)
1228         return;
1229     REF_ASSERT_ISNT(i < 0);
1230 
1231     OPENSSL_free(pre->precomp_storage);
1232     CRYPTO_FREE_REF(&pre->references);
1233     OPENSSL_free(pre);
1234 }
1235 
ecp_nistz256_window_have_precompute_mult(const EC_GROUP * group)1236 static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP *group)
1237 {
1238     /* There is a hard-coded table for the default generator. */
1239     const EC_POINT *generator = EC_GROUP_get0_generator(group);
1240 
1241     if (generator != NULL && ecp_nistz256_is_affine_G(generator)) {
1242         /* There is a hard-coded table for the default generator. */
1243         return 1;
1244     }
1245 
1246     return HAVEPRECOMP(group, nistz256);
1247 }
1248 
1249 #if defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || defined(_M_X64) || defined(__powerpc64__) || defined(_ARCH_PP64) || defined(__aarch64__)
1250 /*
1251  * Montgomery mul modulo Order(P): res = a*b*2^-256 mod Order(P)
1252  */
1253 void ecp_nistz256_ord_mul_mont(BN_ULONG res[P256_LIMBS],
1254     const BN_ULONG a[P256_LIMBS],
1255     const BN_ULONG b[P256_LIMBS]);
1256 void ecp_nistz256_ord_sqr_mont(BN_ULONG res[P256_LIMBS],
1257     const BN_ULONG a[P256_LIMBS],
1258     BN_ULONG rep);
1259 
ecp_nistz256_inv_mod_ord(const EC_GROUP * group,BIGNUM * r,const BIGNUM * x,BN_CTX * ctx)1260 static int ecp_nistz256_inv_mod_ord(const EC_GROUP *group, BIGNUM *r,
1261     const BIGNUM *x, BN_CTX *ctx)
1262 {
1263     /* RR = 2^512 mod ord(p256) */
1264     static const BN_ULONG RR[P256_LIMBS] = {
1265         TOBN(0x83244c95, 0xbe79eea2), TOBN(0x4699799c, 0x49bd6fa6),
1266         TOBN(0x2845b239, 0x2b6bec59), TOBN(0x66e12d94, 0xf3d95620)
1267     };
1268     /* The constant 1 (unlike ONE that is one in Montgomery representation) */
1269     static const BN_ULONG one[P256_LIMBS] = {
1270         TOBN(0, 1), TOBN(0, 0), TOBN(0, 0), TOBN(0, 0)
1271     };
1272     /*
1273      * We don't use entry 0 in the table, so we omit it and address
1274      * with -1 offset.
1275      */
1276     BN_ULONG table[15][P256_LIMBS];
1277     BN_ULONG out[P256_LIMBS], t[P256_LIMBS];
1278     int i, ret = 0;
1279     enum {
1280         i_1 = 0,
1281         i_10,
1282         i_11,
1283         i_101,
1284         i_111,
1285         i_1010,
1286         i_1111,
1287         i_10101,
1288         i_101010,
1289         i_101111,
1290         i_x6,
1291         i_x8,
1292         i_x16,
1293         i_x32
1294     };
1295 
1296     /*
1297      * Catch allocation failure early.
1298      */
1299     if (bn_wexpand(r, P256_LIMBS) == NULL) {
1300         ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1301         goto err;
1302     }
1303 
1304     if ((BN_num_bits(x) > 256) || BN_is_negative(x)) {
1305         BIGNUM *tmp;
1306 
1307         if ((tmp = BN_CTX_get(ctx)) == NULL
1308             || !BN_nnmod(tmp, x, group->order, ctx)) {
1309             ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1310             goto err;
1311         }
1312         x = tmp;
1313     }
1314 
1315     if (!ecp_nistz256_bignum_to_field_elem(t, x)) {
1316         ERR_raise(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
1317         goto err;
1318     }
1319 
1320     ecp_nistz256_ord_mul_mont(table[0], t, RR);
1321 #if 0
1322     /*
1323      * Original sparse-then-fixed-window algorithm, retained for reference.
1324      */
1325     for (i = 2; i < 16; i += 2) {
1326         ecp_nistz256_ord_sqr_mont(table[i-1], table[i/2-1], 1);
1327         ecp_nistz256_ord_mul_mont(table[i], table[i-1], table[0]);
1328     }
1329 
1330     /*
1331      * The top 128bit of the exponent are highly redudndant, so we
1332      * perform an optimized flow
1333      */
1334     ecp_nistz256_ord_sqr_mont(t, table[15-1], 4);   /* f0 */
1335     ecp_nistz256_ord_mul_mont(t, t, table[15-1]);   /* ff */
1336 
1337     ecp_nistz256_ord_sqr_mont(out, t, 8);           /* ff00 */
1338     ecp_nistz256_ord_mul_mont(out, out, t);         /* ffff */
1339 
1340     ecp_nistz256_ord_sqr_mont(t, out, 16);          /* ffff0000 */
1341     ecp_nistz256_ord_mul_mont(t, t, out);           /* ffffffff */
1342 
1343     ecp_nistz256_ord_sqr_mont(out, t, 64);          /* ffffffff0000000000000000 */
1344     ecp_nistz256_ord_mul_mont(out, out, t);         /* ffffffff00000000ffffffff */
1345 
1346     ecp_nistz256_ord_sqr_mont(out, out, 32);        /* ffffffff00000000ffffffff00000000 */
1347     ecp_nistz256_ord_mul_mont(out, out, t);         /* ffffffff00000000ffffffffffffffff */
1348 
1349     /*
1350      * The bottom 128 bit of the exponent are processed with fixed 4-bit window
1351      */
1352     for (i = 0; i < 32; i++) {
1353         /* expLo - the low 128 bits of the exponent we use (ord(p256) - 2),
1354          * split into nibbles */
1355         static const unsigned char expLo[32]  = {
1356             0xb,0xc,0xe,0x6,0xf,0xa,0xa,0xd,0xa,0x7,0x1,0x7,0x9,0xe,0x8,0x4,
1357             0xf,0x3,0xb,0x9,0xc,0xa,0xc,0x2,0xf,0xc,0x6,0x3,0x2,0x5,0x4,0xf
1358         };
1359 
1360         ecp_nistz256_ord_sqr_mont(out, out, 4);
1361         /* The exponent is public, no need in constant-time access */
1362         ecp_nistz256_ord_mul_mont(out, out, table[expLo[i]-1]);
1363     }
1364 #else
1365     /*
1366      * https://briansmith.org/ecc-inversion-addition-chains-01#p256_scalar_inversion
1367      *
1368      * Even though this code path spares 12 squarings, 4.5%, and 13
1369      * multiplications, 25%, on grand scale sign operation is not that
1370      * much faster, not more that 2%...
1371      */
1372 
1373     /* pre-calculate powers */
1374     ecp_nistz256_ord_sqr_mont(table[i_10], table[i_1], 1);
1375 
1376     ecp_nistz256_ord_mul_mont(table[i_11], table[i_1], table[i_10]);
1377 
1378     ecp_nistz256_ord_mul_mont(table[i_101], table[i_11], table[i_10]);
1379 
1380     ecp_nistz256_ord_mul_mont(table[i_111], table[i_101], table[i_10]);
1381 
1382     ecp_nistz256_ord_sqr_mont(table[i_1010], table[i_101], 1);
1383 
1384     ecp_nistz256_ord_mul_mont(table[i_1111], table[i_1010], table[i_101]);
1385 
1386     ecp_nistz256_ord_sqr_mont(table[i_10101], table[i_1010], 1);
1387     ecp_nistz256_ord_mul_mont(table[i_10101], table[i_10101], table[i_1]);
1388 
1389     ecp_nistz256_ord_sqr_mont(table[i_101010], table[i_10101], 1);
1390 
1391     ecp_nistz256_ord_mul_mont(table[i_101111], table[i_101010], table[i_101]);
1392 
1393     ecp_nistz256_ord_mul_mont(table[i_x6], table[i_101010], table[i_10101]);
1394 
1395     ecp_nistz256_ord_sqr_mont(table[i_x8], table[i_x6], 2);
1396     ecp_nistz256_ord_mul_mont(table[i_x8], table[i_x8], table[i_11]);
1397 
1398     ecp_nistz256_ord_sqr_mont(table[i_x16], table[i_x8], 8);
1399     ecp_nistz256_ord_mul_mont(table[i_x16], table[i_x16], table[i_x8]);
1400 
1401     ecp_nistz256_ord_sqr_mont(table[i_x32], table[i_x16], 16);
1402     ecp_nistz256_ord_mul_mont(table[i_x32], table[i_x32], table[i_x16]);
1403 
1404     /* calculations */
1405     ecp_nistz256_ord_sqr_mont(out, table[i_x32], 64);
1406     ecp_nistz256_ord_mul_mont(out, out, table[i_x32]);
1407 
1408     for (i = 0; i < 27; i++) {
1409         static const struct {
1410             unsigned char p, i;
1411         } chain[27] = {
1412             { 32, i_x32 }, { 6, i_101111 }, { 5, i_111 },
1413             { 4, i_11 }, { 5, i_1111 }, { 5, i_10101 },
1414             { 4, i_101 }, { 3, i_101 }, { 3, i_101 },
1415             { 5, i_111 }, { 9, i_101111 }, { 6, i_1111 },
1416             { 2, i_1 }, { 5, i_1 }, { 6, i_1111 },
1417             { 5, i_111 }, { 4, i_111 }, { 5, i_111 },
1418             { 5, i_101 }, { 3, i_11 }, { 10, i_101111 },
1419             { 2, i_11 }, { 5, i_11 }, { 5, i_11 },
1420             { 3, i_1 }, { 7, i_10101 }, { 6, i_1111 }
1421         };
1422 
1423         ecp_nistz256_ord_sqr_mont(out, out, chain[i].p);
1424         ecp_nistz256_ord_mul_mont(out, out, table[chain[i].i]);
1425     }
1426 #endif
1427     ecp_nistz256_ord_mul_mont(out, out, one);
1428 
1429     /*
1430      * Can't fail, but check return code to be consistent anyway.
1431      */
1432     if (!bn_set_words(r, out, P256_LIMBS))
1433         goto err;
1434 
1435     ret = 1;
1436 err:
1437     return ret;
1438 }
1439 #else
1440 #define ecp_nistz256_inv_mod_ord NULL
1441 #endif
1442 
ecp_nistz256group_full_init(EC_GROUP * group,const unsigned char * params)1443 static int ecp_nistz256group_full_init(EC_GROUP *group,
1444     const unsigned char *params)
1445 {
1446     BN_CTX *ctx = NULL;
1447     BN_MONT_CTX *mont = NULL, *ordmont = NULL;
1448     const int param_len = 32;
1449     const int seed_len = 20;
1450     int ok = 0;
1451     uint32_t hi_order_n = 0xccd1c8aa;
1452     uint32_t lo_order_n = 0xee00bc4f;
1453     BIGNUM *p = NULL, *a = NULL, *b = NULL, *x = NULL, *y = NULL, *one = NULL,
1454            *order = NULL;
1455     EC_POINT *P = NULL;
1456 
1457     if ((ctx = BN_CTX_new_ex(group->libctx)) == NULL) {
1458         ERR_raise(ERR_LIB_EC, ERR_R_MALLOC_FAILURE);
1459         return 0;
1460     }
1461 
1462     if (!EC_GROUP_set_seed(group, params, seed_len)) {
1463         ERR_raise(ERR_LIB_EC, ERR_R_EC_LIB);
1464         goto err;
1465     }
1466     params += seed_len;
1467 
1468     if ((p = BN_bin2bn(params + 0 * param_len, param_len, NULL)) == NULL
1469         || (a = BN_bin2bn(params + 1 * param_len, param_len, NULL)) == NULL
1470         || (b = BN_bin2bn(params + 2 * param_len, param_len, NULL)) == NULL) {
1471         ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1472         goto err;
1473     }
1474 
1475     /*
1476      * Set up curve params and montgomery for field
1477      * Start by setting up montgomery and one
1478      */
1479     mont = BN_MONT_CTX_new();
1480     if (mont == NULL)
1481         goto err;
1482 
1483     if (!ossl_bn_mont_ctx_set(mont, p, 256, params + 6 * param_len, param_len,
1484             1, 0))
1485         goto err;
1486 
1487     one = BN_new();
1488     if (one == NULL) {
1489         ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1490         goto err;
1491     }
1492     if (!BN_to_montgomery(one, BN_value_one(), mont, ctx)) {
1493         ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1494         goto err;
1495     }
1496     group->field_data1 = mont;
1497     mont = NULL;
1498     group->field_data2 = one;
1499     one = NULL;
1500 
1501     if (!ossl_ec_GFp_simple_group_set_curve(group, p, a, b, ctx)) {
1502         ERR_raise(ERR_LIB_EC, ERR_R_EC_LIB);
1503         goto err;
1504     }
1505 
1506     if ((P = EC_POINT_new(group)) == NULL) {
1507         ERR_raise(ERR_LIB_EC, ERR_R_EC_LIB);
1508         goto err;
1509     }
1510 
1511     if ((x = BN_bin2bn(params + 3 * param_len, param_len, NULL)) == NULL
1512         || (y = BN_bin2bn(params + 4 * param_len, param_len, NULL)) == NULL) {
1513         ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1514         goto err;
1515     }
1516     if (!EC_POINT_set_affine_coordinates(group, P, x, y, ctx)) {
1517         ERR_raise(ERR_LIB_EC, ERR_R_EC_LIB);
1518         goto err;
1519     }
1520     if ((order = BN_bin2bn(params + 5 * param_len, param_len, NULL)) == NULL
1521         || !BN_set_word(x, (BN_ULONG)1)) { /* cofactor is 1 */
1522         ERR_raise(ERR_LIB_EC, ERR_R_BN_LIB);
1523         goto err;
1524     }
1525 
1526     /*
1527      * Set up generator and order and montgomery data
1528      */
1529     group->generator = EC_POINT_new(group);
1530     if (group->generator == NULL) {
1531         ERR_raise(ERR_LIB_EC, ERR_R_EC_LIB);
1532         goto err;
1533     }
1534     if (!EC_POINT_copy(group->generator, P))
1535         goto err;
1536     if (!BN_copy(group->order, order))
1537         goto err;
1538     if (!BN_set_word(group->cofactor, 1))
1539         goto err;
1540 
1541     ordmont = BN_MONT_CTX_new();
1542     if (ordmont == NULL)
1543         goto err;
1544     if (!ossl_bn_mont_ctx_set(ordmont, order, 256, params + 7 * param_len,
1545             param_len, lo_order_n, hi_order_n))
1546         goto err;
1547 
1548     group->mont_data = ordmont;
1549     ordmont = NULL;
1550 
1551     ok = 1;
1552 
1553 err:
1554     EC_POINT_free(P);
1555     BN_CTX_free(ctx);
1556     BN_MONT_CTX_free(mont);
1557     BN_MONT_CTX_free(ordmont);
1558     BN_free(p);
1559     BN_free(one);
1560     BN_free(a);
1561     BN_free(b);
1562     BN_free(order);
1563     BN_free(x);
1564     BN_free(y);
1565 
1566     return ok;
1567 }
1568 
EC_GFp_nistz256_method(void)1569 const EC_METHOD *EC_GFp_nistz256_method(void)
1570 {
1571     static const EC_METHOD ret = {
1572         EC_FLAGS_DEFAULT_OCT,
1573         NID_X9_62_prime_field,
1574         ossl_ec_GFp_mont_group_init,
1575         ossl_ec_GFp_mont_group_finish,
1576         ossl_ec_GFp_mont_group_clear_finish,
1577         ossl_ec_GFp_mont_group_copy,
1578         ossl_ec_GFp_mont_group_set_curve,
1579         ossl_ec_GFp_simple_group_get_curve,
1580         ossl_ec_GFp_simple_group_get_degree,
1581         ossl_ec_group_simple_order_bits,
1582         ossl_ec_GFp_simple_group_check_discriminant,
1583         ossl_ec_GFp_simple_point_init,
1584         ossl_ec_GFp_simple_point_finish,
1585         ossl_ec_GFp_simple_point_clear_finish,
1586         ossl_ec_GFp_simple_point_copy,
1587         ossl_ec_GFp_simple_point_set_to_infinity,
1588         ossl_ec_GFp_simple_point_set_affine_coordinates,
1589         ecp_nistz256_get_affine,
1590         0, 0, 0,
1591         ossl_ec_GFp_simple_add,
1592         ossl_ec_GFp_simple_dbl,
1593         ossl_ec_GFp_simple_invert,
1594         ossl_ec_GFp_simple_is_at_infinity,
1595         ossl_ec_GFp_simple_is_on_curve,
1596         ossl_ec_GFp_simple_cmp,
1597         ossl_ec_GFp_simple_make_affine,
1598         ossl_ec_GFp_simple_points_make_affine,
1599         ecp_nistz256_points_mul, /* mul */
1600         ecp_nistz256_mult_precompute, /* precompute_mult */
1601         ecp_nistz256_window_have_precompute_mult, /* have_precompute_mult */
1602         ossl_ec_GFp_mont_field_mul,
1603         ossl_ec_GFp_mont_field_sqr,
1604         0, /* field_div */
1605         ossl_ec_GFp_mont_field_inv,
1606         ossl_ec_GFp_mont_field_encode,
1607         ossl_ec_GFp_mont_field_decode,
1608         ossl_ec_GFp_mont_field_set_to_one,
1609         ossl_ec_key_simple_priv2oct,
1610         ossl_ec_key_simple_oct2priv,
1611         0, /* set private */
1612         ossl_ec_key_simple_generate_key,
1613         ossl_ec_key_simple_check_key,
1614         ossl_ec_key_simple_generate_public_key,
1615         0, /* keycopy */
1616         0, /* keyfinish */
1617         ossl_ecdh_simple_compute_key,
1618         ossl_ecdsa_simple_sign_setup,
1619         ossl_ecdsa_simple_sign_sig,
1620         ossl_ecdsa_simple_verify_sig,
1621         ecp_nistz256_inv_mod_ord, /* can be #define-d NULL */
1622         0, /* blind_coordinates */
1623         0, /* ladder_pre */
1624         0, /* ladder_step */
1625         0, /* ladder_post */
1626         ecp_nistz256group_full_init
1627     };
1628 
1629     return &ret;
1630 }
1631