1 /******************************************************************************
3 * Copyright 2014 Intel Corporation *
5 * Licensed under the Apache License, Version 2.0 (the "License"); *
6 * you may not use this file except in compliance with the License. *
7 * You may obtain a copy of the License at *
9 * http://www.apache.org/licenses/LICENSE-2.0 *
11 * Unless required by applicable law or agreed to in writing, software *
12 * distributed under the License is distributed on an "AS IS" BASIS, *
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. *
14 * See the License for the specific language governing permissions and *
15 * limitations under the License. *
17 ******************************************************************************
19 * Developers and authors: *
20 * Shay Gueron (1, 2), and Vlad Krasnov (1) *
21 * (1) Intel Corporation, Israel Development Center *
22 * (2) University of Haifa *
24 * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with *
27 ******************************************************************************/
31 #include "internal/cryptlib.h"
32 #include "internal/bn_int.h"
36 # define TOBN(hi,lo) lo,hi
38 # define TOBN(hi,lo) ((BN_ULONG)hi<<32|lo)
42 # define ALIGN32 __attribute((aligned(32)))
43 #elif defined(_MSC_VER)
44 # define ALIGN32 __declspec(align(32))
49 #define ALIGNPTR(p,N) ((unsigned char *)p+N-(size_t)p%N)
50 #define P256_LIMBS (256/BN_BITS2)
52 typedef unsigned short u16
;
55 BN_ULONG X
[P256_LIMBS
];
56 BN_ULONG Y
[P256_LIMBS
];
57 BN_ULONG Z
[P256_LIMBS
];
61 BN_ULONG X
[P256_LIMBS
];
62 BN_ULONG Y
[P256_LIMBS
];
65 typedef P256_POINT_AFFINE PRECOMP256_ROW
[64];
67 /* structure for precomputed multiples of the generator */
68 struct nistz256_pre_comp_st
{
69 const EC_GROUP
*group
; /* Parent EC_GROUP object */
70 size_t w
; /* Window size */
72 * Constant time access to the X and Y coordinates of the pre-computed,
73 * generator multiplies, in the Montgomery domain. Pre-calculated
74 * multiplies are stored in affine form.
76 PRECOMP256_ROW
*precomp
;
77 void *precomp_storage
;
81 /* Functions implemented in assembly */
82 /* Modular mul by 2: res = 2*a mod P */
83 void ecp_nistz256_mul_by_2(BN_ULONG res
[P256_LIMBS
],
84 const BN_ULONG a
[P256_LIMBS
]);
85 /* Modular div by 2: res = a/2 mod P */
86 void ecp_nistz256_div_by_2(BN_ULONG res
[P256_LIMBS
],
87 const BN_ULONG a
[P256_LIMBS
]);
88 /* Modular mul by 3: res = 3*a mod P */
89 void ecp_nistz256_mul_by_3(BN_ULONG res
[P256_LIMBS
],
90 const BN_ULONG a
[P256_LIMBS
]);
91 /* Modular add: res = a+b mod P */
92 void ecp_nistz256_add(BN_ULONG res
[P256_LIMBS
],
93 const BN_ULONG a
[P256_LIMBS
],
94 const BN_ULONG b
[P256_LIMBS
]);
95 /* Modular sub: res = a-b mod P */
96 void ecp_nistz256_sub(BN_ULONG res
[P256_LIMBS
],
97 const BN_ULONG a
[P256_LIMBS
],
98 const BN_ULONG b
[P256_LIMBS
]);
99 /* Modular neg: res = -a mod P */
100 void ecp_nistz256_neg(BN_ULONG res
[P256_LIMBS
], const BN_ULONG a
[P256_LIMBS
]);
101 /* Montgomery mul: res = a*b*2^-256 mod P */
102 void ecp_nistz256_mul_mont(BN_ULONG res
[P256_LIMBS
],
103 const BN_ULONG a
[P256_LIMBS
],
104 const BN_ULONG b
[P256_LIMBS
]);
105 /* Montgomery sqr: res = a*a*2^-256 mod P */
106 void ecp_nistz256_sqr_mont(BN_ULONG res
[P256_LIMBS
],
107 const BN_ULONG a
[P256_LIMBS
]);
108 /* Convert a number from Montgomery domain, by multiplying with 1 */
109 void ecp_nistz256_from_mont(BN_ULONG res
[P256_LIMBS
],
110 const BN_ULONG in
[P256_LIMBS
]);
111 /* Convert a number to Montgomery domain, by multiplying with 2^512 mod P*/
112 void ecp_nistz256_to_mont(BN_ULONG res
[P256_LIMBS
],
113 const BN_ULONG in
[P256_LIMBS
]);
114 /* Functions that perform constant time access to the precomputed tables */
115 void ecp_nistz256_scatter_w5(P256_POINT
*val
,
116 const P256_POINT
*in_t
, int idx
);
117 void ecp_nistz256_gather_w5(P256_POINT
*val
,
118 const P256_POINT
*in_t
, int idx
);
119 void ecp_nistz256_scatter_w7(P256_POINT_AFFINE
*val
,
120 const P256_POINT_AFFINE
*in_t
, int idx
);
121 void ecp_nistz256_gather_w7(P256_POINT_AFFINE
*val
,
122 const P256_POINT_AFFINE
*in_t
, int idx
);
124 /* One converted into the Montgomery domain */
125 static const BN_ULONG ONE
[P256_LIMBS
] = {
126 TOBN(0x00000000, 0x00000001), TOBN(0xffffffff, 0x00000000),
127 TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe)
130 static NISTZ256_PRE_COMP
*ecp_nistz256_pre_comp_new(const EC_GROUP
*group
);
132 /* Precomputed tables for the default generator */
133 extern const PRECOMP256_ROW ecp_nistz256_precomputed
[37];
135 /* Recode window to a signed digit, see ecp_nistputil.c for details */
136 static unsigned int _booth_recode_w5(unsigned int in
)
140 s
= ~((in
>> 5) - 1);
141 d
= (1 << 6) - in
- 1;
142 d
= (d
& s
) | (in
& ~s
);
143 d
= (d
>> 1) + (d
& 1);
145 return (d
<< 1) + (s
& 1);
148 static unsigned int _booth_recode_w7(unsigned int in
)
152 s
= ~((in
>> 7) - 1);
153 d
= (1 << 8) - in
- 1;
154 d
= (d
& s
) | (in
& ~s
);
155 d
= (d
>> 1) + (d
& 1);
157 return (d
<< 1) + (s
& 1);
160 static void copy_conditional(BN_ULONG dst
[P256_LIMBS
],
161 const BN_ULONG src
[P256_LIMBS
], BN_ULONG move
)
163 BN_ULONG mask1
= 0-move
;
164 BN_ULONG mask2
= ~mask1
;
166 dst
[0] = (src
[0] & mask1
) ^ (dst
[0] & mask2
);
167 dst
[1] = (src
[1] & mask1
) ^ (dst
[1] & mask2
);
168 dst
[2] = (src
[2] & mask1
) ^ (dst
[2] & mask2
);
169 dst
[3] = (src
[3] & mask1
) ^ (dst
[3] & mask2
);
170 if (P256_LIMBS
== 8) {
171 dst
[4] = (src
[4] & mask1
) ^ (dst
[4] & mask2
);
172 dst
[5] = (src
[5] & mask1
) ^ (dst
[5] & mask2
);
173 dst
[6] = (src
[6] & mask1
) ^ (dst
[6] & mask2
);
174 dst
[7] = (src
[7] & mask1
) ^ (dst
[7] & mask2
);
178 static BN_ULONG
is_zero(BN_ULONG in
)
187 static BN_ULONG
is_equal(const BN_ULONG a
[P256_LIMBS
],
188 const BN_ULONG b
[P256_LIMBS
])
196 if (P256_LIMBS
== 8) {
206 static BN_ULONG
is_one(const BN_ULONG a
[P256_LIMBS
])
211 res
|= a
[1] ^ ONE
[1];
212 res
|= a
[2] ^ ONE
[2];
213 res
|= a
[3] ^ ONE
[3];
214 if (P256_LIMBS
== 8) {
215 res
|= a
[4] ^ ONE
[4];
216 res
|= a
[5] ^ ONE
[5];
217 res
|= a
[6] ^ ONE
[6];
223 #ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
224 void ecp_nistz256_point_double(P256_POINT
*r
, const P256_POINT
*a
);
225 void ecp_nistz256_point_add(P256_POINT
*r
,
226 const P256_POINT
*a
, const P256_POINT
*b
);
227 void ecp_nistz256_point_add_affine(P256_POINT
*r
,
229 const P256_POINT_AFFINE
*b
);
231 /* Point double: r = 2*a */
232 static void ecp_nistz256_point_double(P256_POINT
*r
, const P256_POINT
*a
)
234 BN_ULONG S
[P256_LIMBS
];
235 BN_ULONG M
[P256_LIMBS
];
236 BN_ULONG Zsqr
[P256_LIMBS
];
237 BN_ULONG tmp0
[P256_LIMBS
];
239 const BN_ULONG
*in_x
= a
->X
;
240 const BN_ULONG
*in_y
= a
->Y
;
241 const BN_ULONG
*in_z
= a
->Z
;
243 BN_ULONG
*res_x
= r
->X
;
244 BN_ULONG
*res_y
= r
->Y
;
245 BN_ULONG
*res_z
= r
->Z
;
247 ecp_nistz256_mul_by_2(S
, in_y
);
249 ecp_nistz256_sqr_mont(Zsqr
, in_z
);
251 ecp_nistz256_sqr_mont(S
, S
);
253 ecp_nistz256_mul_mont(res_z
, in_z
, in_y
);
254 ecp_nistz256_mul_by_2(res_z
, res_z
);
256 ecp_nistz256_add(M
, in_x
, Zsqr
);
257 ecp_nistz256_sub(Zsqr
, in_x
, Zsqr
);
259 ecp_nistz256_sqr_mont(res_y
, S
);
260 ecp_nistz256_div_by_2(res_y
, res_y
);
262 ecp_nistz256_mul_mont(M
, M
, Zsqr
);
263 ecp_nistz256_mul_by_3(M
, M
);
265 ecp_nistz256_mul_mont(S
, S
, in_x
);
266 ecp_nistz256_mul_by_2(tmp0
, S
);
268 ecp_nistz256_sqr_mont(res_x
, M
);
270 ecp_nistz256_sub(res_x
, res_x
, tmp0
);
271 ecp_nistz256_sub(S
, S
, res_x
);
273 ecp_nistz256_mul_mont(S
, S
, M
);
274 ecp_nistz256_sub(res_y
, S
, res_y
);
277 /* Point addition: r = a+b */
278 static void ecp_nistz256_point_add(P256_POINT
*r
,
279 const P256_POINT
*a
, const P256_POINT
*b
)
281 BN_ULONG U2
[P256_LIMBS
], S2
[P256_LIMBS
];
282 BN_ULONG U1
[P256_LIMBS
], S1
[P256_LIMBS
];
283 BN_ULONG Z1sqr
[P256_LIMBS
];
284 BN_ULONG Z2sqr
[P256_LIMBS
];
285 BN_ULONG H
[P256_LIMBS
], R
[P256_LIMBS
];
286 BN_ULONG Hsqr
[P256_LIMBS
];
287 BN_ULONG Rsqr
[P256_LIMBS
];
288 BN_ULONG Hcub
[P256_LIMBS
];
290 BN_ULONG res_x
[P256_LIMBS
];
291 BN_ULONG res_y
[P256_LIMBS
];
292 BN_ULONG res_z
[P256_LIMBS
];
294 BN_ULONG in1infty
, in2infty
;
296 const BN_ULONG
*in1_x
= a
->X
;
297 const BN_ULONG
*in1_y
= a
->Y
;
298 const BN_ULONG
*in1_z
= a
->Z
;
300 const BN_ULONG
*in2_x
= b
->X
;
301 const BN_ULONG
*in2_y
= b
->Y
;
302 const BN_ULONG
*in2_z
= b
->Z
;
304 /* We encode infinity as (0,0), which is not on the curve,
306 in1infty
= (in1_x
[0] | in1_x
[1] | in1_x
[2] | in1_x
[3] |
307 in1_y
[0] | in1_y
[1] | in1_y
[2] | in1_y
[3]);
309 in1infty
|= (in1_x
[4] | in1_x
[5] | in1_x
[6] | in1_x
[7] |
310 in1_y
[4] | in1_y
[5] | in1_y
[6] | in1_y
[7]);
312 in2infty
= (in2_x
[0] | in2_x
[1] | in2_x
[2] | in2_x
[3] |
313 in2_y
[0] | in2_y
[1] | in2_y
[2] | in2_y
[3]);
315 in2infty
|= (in2_x
[4] | in2_x
[5] | in2_x
[6] | in2_x
[7] |
316 in2_y
[4] | in2_y
[5] | in2_y
[6] | in2_y
[7]);
318 in1infty
= is_zero(in1infty
);
319 in2infty
= is_zero(in2infty
);
321 ecp_nistz256_sqr_mont(Z2sqr
, in2_z
); /* Z2^2 */
322 ecp_nistz256_sqr_mont(Z1sqr
, in1_z
); /* Z1^2 */
324 ecp_nistz256_mul_mont(S1
, Z2sqr
, in2_z
); /* S1 = Z2^3 */
325 ecp_nistz256_mul_mont(S2
, Z1sqr
, in1_z
); /* S2 = Z1^3 */
327 ecp_nistz256_mul_mont(S1
, S1
, in1_y
); /* S1 = Y1*Z2^3 */
328 ecp_nistz256_mul_mont(S2
, S2
, in2_y
); /* S2 = Y2*Z1^3 */
329 ecp_nistz256_sub(R
, S2
, S1
); /* R = S2 - S1 */
331 ecp_nistz256_mul_mont(U1
, in1_x
, Z2sqr
); /* U1 = X1*Z2^2 */
332 ecp_nistz256_mul_mont(U2
, in2_x
, Z1sqr
); /* U2 = X2*Z1^2 */
333 ecp_nistz256_sub(H
, U2
, U1
); /* H = U2 - U1 */
336 * This should not happen during sign/ecdh, so no constant time violation
338 if (is_equal(U1
, U2
) && !in1infty
&& !in2infty
) {
339 if (is_equal(S1
, S2
)) {
340 ecp_nistz256_point_double(r
, a
);
343 memset(r
, 0, sizeof(*r
));
348 ecp_nistz256_sqr_mont(Rsqr
, R
); /* R^2 */
349 ecp_nistz256_mul_mont(res_z
, H
, in1_z
); /* Z3 = H*Z1*Z2 */
350 ecp_nistz256_sqr_mont(Hsqr
, H
); /* H^2 */
351 ecp_nistz256_mul_mont(res_z
, res_z
, in2_z
); /* Z3 = H*Z1*Z2 */
352 ecp_nistz256_mul_mont(Hcub
, Hsqr
, H
); /* H^3 */
354 ecp_nistz256_mul_mont(U2
, U1
, Hsqr
); /* U1*H^2 */
355 ecp_nistz256_mul_by_2(Hsqr
, U2
); /* 2*U1*H^2 */
357 ecp_nistz256_sub(res_x
, Rsqr
, Hsqr
);
358 ecp_nistz256_sub(res_x
, res_x
, Hcub
);
360 ecp_nistz256_sub(res_y
, U2
, res_x
);
362 ecp_nistz256_mul_mont(S2
, S1
, Hcub
);
363 ecp_nistz256_mul_mont(res_y
, R
, res_y
);
364 ecp_nistz256_sub(res_y
, res_y
, S2
);
366 copy_conditional(res_x
, in2_x
, in1infty
);
367 copy_conditional(res_y
, in2_y
, in1infty
);
368 copy_conditional(res_z
, in2_z
, in1infty
);
370 copy_conditional(res_x
, in1_x
, in2infty
);
371 copy_conditional(res_y
, in1_y
, in2infty
);
372 copy_conditional(res_z
, in1_z
, in2infty
);
374 memcpy(r
->X
, res_x
, sizeof(res_x
));
375 memcpy(r
->Y
, res_y
, sizeof(res_y
));
376 memcpy(r
->Z
, res_z
, sizeof(res_z
));
379 /* Point addition when b is known to be affine: r = a+b */
380 static void ecp_nistz256_point_add_affine(P256_POINT
*r
,
382 const P256_POINT_AFFINE
*b
)
384 BN_ULONG U2
[P256_LIMBS
], S2
[P256_LIMBS
];
385 BN_ULONG Z1sqr
[P256_LIMBS
];
386 BN_ULONG H
[P256_LIMBS
], R
[P256_LIMBS
];
387 BN_ULONG Hsqr
[P256_LIMBS
];
388 BN_ULONG Rsqr
[P256_LIMBS
];
389 BN_ULONG Hcub
[P256_LIMBS
];
391 BN_ULONG res_x
[P256_LIMBS
];
392 BN_ULONG res_y
[P256_LIMBS
];
393 BN_ULONG res_z
[P256_LIMBS
];
395 BN_ULONG in1infty
, in2infty
;
397 const BN_ULONG
*in1_x
= a
->X
;
398 const BN_ULONG
*in1_y
= a
->Y
;
399 const BN_ULONG
*in1_z
= a
->Z
;
401 const BN_ULONG
*in2_x
= b
->X
;
402 const BN_ULONG
*in2_y
= b
->Y
;
405 * In affine representation we encode infty as (0,0), which is not on the
408 in1infty
= (in1_x
[0] | in1_x
[1] | in1_x
[2] | in1_x
[3] |
409 in1_y
[0] | in1_y
[1] | in1_y
[2] | in1_y
[3]);
411 in1infty
|= (in1_x
[4] | in1_x
[5] | in1_x
[6] | in1_x
[7] |
412 in1_y
[4] | in1_y
[5] | in1_y
[6] | in1_y
[7]);
414 in2infty
= (in2_x
[0] | in2_x
[1] | in2_x
[2] | in2_x
[3] |
415 in2_y
[0] | in2_y
[1] | in2_y
[2] | in2_y
[3]);
417 in2infty
|= (in2_x
[4] | in2_x
[5] | in2_x
[6] | in2_x
[7] |
418 in2_y
[4] | in2_y
[5] | in2_y
[6] | in2_y
[7]);
420 in1infty
= is_zero(in1infty
);
421 in2infty
= is_zero(in2infty
);
423 ecp_nistz256_sqr_mont(Z1sqr
, in1_z
); /* Z1^2 */
425 ecp_nistz256_mul_mont(U2
, in2_x
, Z1sqr
); /* U2 = X2*Z1^2 */
426 ecp_nistz256_sub(H
, U2
, in1_x
); /* H = U2 - U1 */
428 ecp_nistz256_mul_mont(S2
, Z1sqr
, in1_z
); /* S2 = Z1^3 */
430 ecp_nistz256_mul_mont(res_z
, H
, in1_z
); /* Z3 = H*Z1*Z2 */
432 ecp_nistz256_mul_mont(S2
, S2
, in2_y
); /* S2 = Y2*Z1^3 */
433 ecp_nistz256_sub(R
, S2
, in1_y
); /* R = S2 - S1 */
435 ecp_nistz256_sqr_mont(Hsqr
, H
); /* H^2 */
436 ecp_nistz256_sqr_mont(Rsqr
, R
); /* R^2 */
437 ecp_nistz256_mul_mont(Hcub
, Hsqr
, H
); /* H^3 */
439 ecp_nistz256_mul_mont(U2
, in1_x
, Hsqr
); /* U1*H^2 */
440 ecp_nistz256_mul_by_2(Hsqr
, U2
); /* 2*U1*H^2 */
442 ecp_nistz256_sub(res_x
, Rsqr
, Hsqr
);
443 ecp_nistz256_sub(res_x
, res_x
, Hcub
);
444 ecp_nistz256_sub(H
, U2
, res_x
);
446 ecp_nistz256_mul_mont(S2
, in1_y
, Hcub
);
447 ecp_nistz256_mul_mont(H
, H
, R
);
448 ecp_nistz256_sub(res_y
, H
, S2
);
450 copy_conditional(res_x
, in2_x
, in1infty
);
451 copy_conditional(res_x
, in1_x
, in2infty
);
453 copy_conditional(res_y
, in2_y
, in1infty
);
454 copy_conditional(res_y
, in1_y
, in2infty
);
456 copy_conditional(res_z
, ONE
, in1infty
);
457 copy_conditional(res_z
, in1_z
, in2infty
);
459 memcpy(r
->X
, res_x
, sizeof(res_x
));
460 memcpy(r
->Y
, res_y
, sizeof(res_y
));
461 memcpy(r
->Z
, res_z
, sizeof(res_z
));
465 /* r = in^-1 mod p */
466 static void ecp_nistz256_mod_inverse(BN_ULONG r
[P256_LIMBS
],
467 const BN_ULONG in
[P256_LIMBS
])
470 * The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff
471 * ffffffff ffffffff We use FLT and used poly-2 as exponent
473 BN_ULONG p2
[P256_LIMBS
];
474 BN_ULONG p4
[P256_LIMBS
];
475 BN_ULONG p8
[P256_LIMBS
];
476 BN_ULONG p16
[P256_LIMBS
];
477 BN_ULONG p32
[P256_LIMBS
];
478 BN_ULONG res
[P256_LIMBS
];
481 ecp_nistz256_sqr_mont(res
, in
);
482 ecp_nistz256_mul_mont(p2
, res
, in
); /* 3*p */
484 ecp_nistz256_sqr_mont(res
, p2
);
485 ecp_nistz256_sqr_mont(res
, res
);
486 ecp_nistz256_mul_mont(p4
, res
, p2
); /* f*p */
488 ecp_nistz256_sqr_mont(res
, p4
);
489 ecp_nistz256_sqr_mont(res
, res
);
490 ecp_nistz256_sqr_mont(res
, res
);
491 ecp_nistz256_sqr_mont(res
, res
);
492 ecp_nistz256_mul_mont(p8
, res
, p4
); /* ff*p */
494 ecp_nistz256_sqr_mont(res
, p8
);
495 for (i
= 0; i
< 7; i
++)
496 ecp_nistz256_sqr_mont(res
, res
);
497 ecp_nistz256_mul_mont(p16
, res
, p8
); /* ffff*p */
499 ecp_nistz256_sqr_mont(res
, p16
);
500 for (i
= 0; i
< 15; i
++)
501 ecp_nistz256_sqr_mont(res
, res
);
502 ecp_nistz256_mul_mont(p32
, res
, p16
); /* ffffffff*p */
504 ecp_nistz256_sqr_mont(res
, p32
);
505 for (i
= 0; i
< 31; i
++)
506 ecp_nistz256_sqr_mont(res
, res
);
507 ecp_nistz256_mul_mont(res
, res
, in
);
509 for (i
= 0; i
< 32 * 4; i
++)
510 ecp_nistz256_sqr_mont(res
, res
);
511 ecp_nistz256_mul_mont(res
, res
, p32
);
513 for (i
= 0; i
< 32; i
++)
514 ecp_nistz256_sqr_mont(res
, res
);
515 ecp_nistz256_mul_mont(res
, res
, p32
);
517 for (i
= 0; i
< 16; i
++)
518 ecp_nistz256_sqr_mont(res
, res
);
519 ecp_nistz256_mul_mont(res
, res
, p16
);
521 for (i
= 0; i
< 8; i
++)
522 ecp_nistz256_sqr_mont(res
, res
);
523 ecp_nistz256_mul_mont(res
, res
, p8
);
525 ecp_nistz256_sqr_mont(res
, res
);
526 ecp_nistz256_sqr_mont(res
, res
);
527 ecp_nistz256_sqr_mont(res
, res
);
528 ecp_nistz256_sqr_mont(res
, res
);
529 ecp_nistz256_mul_mont(res
, res
, p4
);
531 ecp_nistz256_sqr_mont(res
, res
);
532 ecp_nistz256_sqr_mont(res
, res
);
533 ecp_nistz256_mul_mont(res
, res
, p2
);
535 ecp_nistz256_sqr_mont(res
, res
);
536 ecp_nistz256_sqr_mont(res
, res
);
537 ecp_nistz256_mul_mont(res
, res
, in
);
539 memcpy(r
, res
, sizeof(res
));
543 * ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
544 * returns one if it fits. Otherwise it returns zero.
546 __owur
static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out
[P256_LIMBS
],
549 return bn_copy_words(out
, in
, P256_LIMBS
);
552 /* r = sum(scalar[i]*point[i]) */
553 __owur
static int ecp_nistz256_windowed_mul(const EC_GROUP
*group
,
555 const BIGNUM
**scalar
,
556 const EC_POINT
**point
,
557 size_t num
, BN_CTX
*ctx
)
562 unsigned char (*p_str
)[33] = NULL
;
563 const unsigned int window_size
= 5;
564 const unsigned int mask
= (1 << (window_size
+ 1)) - 1;
566 P256_POINT
*temp
; /* place for 5 temporary points */
567 const BIGNUM
**scalars
= NULL
;
568 P256_POINT (*table
)[16] = NULL
;
569 void *table_storage
= NULL
;
571 if ((num
* 16 + 6) > OPENSSL_MALLOC_MAX_NELEMS(P256_POINT
)
573 OPENSSL_malloc((num
* 16 + 5) * sizeof(P256_POINT
) + 64)) == NULL
575 OPENSSL_malloc(num
* 33 * sizeof(unsigned char))) == NULL
576 || (scalars
= OPENSSL_malloc(num
* sizeof(BIGNUM
*))) == NULL
) {
577 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL
, ERR_R_MALLOC_FAILURE
);
581 table
= (void *)ALIGNPTR(table_storage
, 64);
582 temp
= (P256_POINT
*)(table
+ num
);
584 for (i
= 0; i
< num
; i
++) {
585 P256_POINT
*row
= table
[i
];
587 /* This is an unusual input, we don't guarantee constant-timeness. */
588 if ((BN_num_bits(scalar
[i
]) > 256) || BN_is_negative(scalar
[i
])) {
591 if ((mod
= BN_CTX_get(ctx
)) == NULL
)
593 if (!BN_nnmod(mod
, scalar
[i
], group
->order
, ctx
)) {
594 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL
, ERR_R_BN_LIB
);
599 scalars
[i
] = scalar
[i
];
601 for (j
= 0; j
< bn_get_top(scalars
[i
]) * BN_BYTES
; j
+= BN_BYTES
) {
602 BN_ULONG d
= bn_get_words(scalars
[i
])[j
/ BN_BYTES
];
604 p_str
[i
][j
+ 0] = (unsigned char)d
;
605 p_str
[i
][j
+ 1] = (unsigned char)(d
>> 8);
606 p_str
[i
][j
+ 2] = (unsigned char)(d
>> 16);
607 p_str
[i
][j
+ 3] = (unsigned char)(d
>>= 24);
610 p_str
[i
][j
+ 4] = (unsigned char)d
;
611 p_str
[i
][j
+ 5] = (unsigned char)(d
>> 8);
612 p_str
[i
][j
+ 6] = (unsigned char)(d
>> 16);
613 p_str
[i
][j
+ 7] = (unsigned char)(d
>> 24);
619 if (!ecp_nistz256_bignum_to_field_elem(temp
[0].X
, point
[i
]->X
)
620 || !ecp_nistz256_bignum_to_field_elem(temp
[0].Y
, point
[i
]->Y
)
621 || !ecp_nistz256_bignum_to_field_elem(temp
[0].Z
, point
[i
]->Z
)) {
622 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL
,
623 EC_R_COORDINATES_OUT_OF_RANGE
);
628 * row[0] is implicitly (0,0,0) (the point at infinity), therefore it
629 * is not stored. All other values are actually stored with an offset
633 ecp_nistz256_scatter_w5 (row
, &temp
[0], 1);
634 ecp_nistz256_point_double(&temp
[1], &temp
[0]); /*1+1=2 */
635 ecp_nistz256_scatter_w5 (row
, &temp
[1], 2);
636 ecp_nistz256_point_add (&temp
[2], &temp
[1], &temp
[0]); /*2+1=3 */
637 ecp_nistz256_scatter_w5 (row
, &temp
[2], 3);
638 ecp_nistz256_point_double(&temp
[1], &temp
[1]); /*2*2=4 */
639 ecp_nistz256_scatter_w5 (row
, &temp
[1], 4);
640 ecp_nistz256_point_double(&temp
[2], &temp
[2]); /*2*3=6 */
641 ecp_nistz256_scatter_w5 (row
, &temp
[2], 6);
642 ecp_nistz256_point_add (&temp
[3], &temp
[1], &temp
[0]); /*4+1=5 */
643 ecp_nistz256_scatter_w5 (row
, &temp
[3], 5);
644 ecp_nistz256_point_add (&temp
[4], &temp
[2], &temp
[0]); /*6+1=7 */
645 ecp_nistz256_scatter_w5 (row
, &temp
[4], 7);
646 ecp_nistz256_point_double(&temp
[1], &temp
[1]); /*2*4=8 */
647 ecp_nistz256_scatter_w5 (row
, &temp
[1], 8);
648 ecp_nistz256_point_double(&temp
[2], &temp
[2]); /*2*6=12 */
649 ecp_nistz256_scatter_w5 (row
, &temp
[2], 12);
650 ecp_nistz256_point_double(&temp
[3], &temp
[3]); /*2*5=10 */
651 ecp_nistz256_scatter_w5 (row
, &temp
[3], 10);
652 ecp_nistz256_point_double(&temp
[4], &temp
[4]); /*2*7=14 */
653 ecp_nistz256_scatter_w5 (row
, &temp
[4], 14);
654 ecp_nistz256_point_add (&temp
[2], &temp
[2], &temp
[0]); /*12+1=13*/
655 ecp_nistz256_scatter_w5 (row
, &temp
[2], 13);
656 ecp_nistz256_point_add (&temp
[3], &temp
[3], &temp
[0]); /*10+1=11*/
657 ecp_nistz256_scatter_w5 (row
, &temp
[3], 11);
658 ecp_nistz256_point_add (&temp
[4], &temp
[4], &temp
[0]); /*14+1=15*/
659 ecp_nistz256_scatter_w5 (row
, &temp
[4], 15);
660 ecp_nistz256_point_add (&temp
[2], &temp
[1], &temp
[0]); /*8+1=9 */
661 ecp_nistz256_scatter_w5 (row
, &temp
[2], 9);
662 ecp_nistz256_point_double(&temp
[1], &temp
[1]); /*2*8=16 */
663 ecp_nistz256_scatter_w5 (row
, &temp
[1], 16);
668 wvalue
= p_str
[0][(idx
- 1) / 8];
669 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
672 * We gather to temp[0], because we know it's position relative
675 ecp_nistz256_gather_w5(&temp
[0], table
[0], _booth_recode_w5(wvalue
) >> 1);
676 memcpy(r
, &temp
[0], sizeof(temp
[0]));
679 for (i
= (idx
== 255 ? 1 : 0); i
< num
; i
++) {
680 unsigned int off
= (idx
- 1) / 8;
682 wvalue
= p_str
[i
][off
] | p_str
[i
][off
+ 1] << 8;
683 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
685 wvalue
= _booth_recode_w5(wvalue
);
687 ecp_nistz256_gather_w5(&temp
[0], table
[i
], wvalue
>> 1);
689 ecp_nistz256_neg(temp
[1].Y
, temp
[0].Y
);
690 copy_conditional(temp
[0].Y
, temp
[1].Y
, (wvalue
& 1));
692 ecp_nistz256_point_add(r
, r
, &temp
[0]);
697 ecp_nistz256_point_double(r
, r
);
698 ecp_nistz256_point_double(r
, r
);
699 ecp_nistz256_point_double(r
, r
);
700 ecp_nistz256_point_double(r
, r
);
701 ecp_nistz256_point_double(r
, r
);
705 for (i
= 0; i
< num
; i
++) {
706 wvalue
= p_str
[i
][0];
707 wvalue
= (wvalue
<< 1) & mask
;
709 wvalue
= _booth_recode_w5(wvalue
);
711 ecp_nistz256_gather_w5(&temp
[0], table
[i
], wvalue
>> 1);
713 ecp_nistz256_neg(temp
[1].Y
, temp
[0].Y
);
714 copy_conditional(temp
[0].Y
, temp
[1].Y
, wvalue
& 1);
716 ecp_nistz256_point_add(r
, r
, &temp
[0]);
721 OPENSSL_free(table_storage
);
723 OPENSSL_free(scalars
);
727 /* Coordinates of G, for which we have precomputed tables */
728 const static BN_ULONG def_xG
[P256_LIMBS
] = {
729 TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
730 TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
733 const static BN_ULONG def_yG
[P256_LIMBS
] = {
734 TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
735 TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
739 * ecp_nistz256_is_affine_G returns one if |generator| is the standard, P-256
742 static int ecp_nistz256_is_affine_G(const EC_POINT
*generator
)
744 return (bn_get_top(generator
->X
) == P256_LIMBS
) &&
745 (bn_get_top(generator
->Y
) == P256_LIMBS
) &&
746 (bn_get_top(generator
->Z
) == (P256_LIMBS
- P256_LIMBS
/ 8)) &&
747 is_equal(bn_get_words(generator
->X
), def_xG
) &&
748 is_equal(bn_get_words(generator
->Y
), def_yG
) &&
749 is_one(bn_get_words(generator
->Z
));
752 __owur
static int ecp_nistz256_mult_precompute(EC_GROUP
*group
, BN_CTX
*ctx
)
755 * We precompute a table for a Booth encoded exponent (wNAF) based
756 * computation. Each table holds 64 values for safe access, with an
757 * implicit value of infinity at index zero. We use window of size 7, and
758 * therefore require ceil(256/7) = 37 tables.
761 EC_POINT
*P
= NULL
, *T
= NULL
;
762 const EC_POINT
*generator
;
763 NISTZ256_PRE_COMP
*pre_comp
;
764 BN_CTX
*new_ctx
= NULL
;
765 int i
, j
, k
, ret
= 0;
768 PRECOMP256_ROW
*preComputedTable
= NULL
;
769 unsigned char *precomp_storage
= NULL
;
771 /* if there is an old NISTZ256_PRE_COMP object, throw it away */
772 EC_pre_comp_free(group
);
773 generator
= EC_GROUP_get0_generator(group
);
774 if (generator
== NULL
) {
775 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE
, EC_R_UNDEFINED_GENERATOR
);
779 if (ecp_nistz256_is_affine_G(generator
)) {
781 * No need to calculate tables for the standard generator because we
782 * have them statically.
787 if ((pre_comp
= ecp_nistz256_pre_comp_new(group
)) == NULL
)
791 ctx
= new_ctx
= BN_CTX_new();
797 order
= BN_CTX_get(ctx
);
802 if (!EC_GROUP_get_order(group
, order
, ctx
))
805 if (BN_is_zero(order
)) {
806 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE
, EC_R_UNKNOWN_ORDER
);
812 if ((precomp_storage
=
813 OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE
) + 64)) == NULL
) {
814 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE
, ERR_R_MALLOC_FAILURE
);
818 preComputedTable
= (void *)ALIGNPTR(precomp_storage
, 64);
820 P
= EC_POINT_new(group
);
821 T
= EC_POINT_new(group
);
822 if (P
== NULL
|| T
== NULL
)
826 * The zero entry is implicitly infinity, and we skip it, storing other
827 * values with -1 offset.
829 if (!EC_POINT_copy(T
, generator
))
832 for (k
= 0; k
< 64; k
++) {
833 if (!EC_POINT_copy(P
, T
))
835 for (j
= 0; j
< 37; j
++) {
836 P256_POINT_AFFINE temp
;
838 * It would be faster to use EC_POINTs_make_affine and
839 * make multiple points affine at the same time.
841 if (!EC_POINT_make_affine(group
, P
, ctx
))
843 if (!ecp_nistz256_bignum_to_field_elem(temp
.X
, P
->X
) ||
844 !ecp_nistz256_bignum_to_field_elem(temp
.Y
, P
->Y
)) {
845 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE
,
846 EC_R_COORDINATES_OUT_OF_RANGE
);
849 ecp_nistz256_scatter_w7(preComputedTable
[j
], &temp
, k
);
850 for (i
= 0; i
< 7; i
++) {
851 if (!EC_POINT_dbl(group
, P
, P
, ctx
))
855 if (!EC_POINT_add(group
, T
, T
, generator
, ctx
))
859 pre_comp
->group
= group
;
861 pre_comp
->precomp
= preComputedTable
;
862 pre_comp
->precomp_storage
= precomp_storage
;
863 precomp_storage
= NULL
;
864 SETPRECOMP(group
, nistz256
, pre_comp
);
871 BN_CTX_free(new_ctx
);
873 EC_nistz256_pre_comp_free(pre_comp
);
874 OPENSSL_free(precomp_storage
);
881 * Note that by default ECP_NISTZ256_AVX2 is undefined. While it's great
882 * code processing 4 points in parallel, corresponding serial operation
883 * is several times slower, because it uses 29x29=58-bit multiplication
884 * as opposite to 64x64=128-bit in integer-only scalar case. As result
885 * it doesn't provide *significant* performance improvement. Note that
886 * just defining ECP_NISTZ256_AVX2 is not sufficient to make it work,
887 * you'd need to compile even asm/ecp_nistz256-avx.pl module.
889 #if defined(ECP_NISTZ256_AVX2)
890 # if !(defined(__x86_64) || defined(__x86_64__) || \
891 defined(_M_AMD64) || defined(_MX64)) || \
892 !(defined(__GNUC__) || defined(_MSC_VER)) /* this is for ALIGN32 */
893 # undef ECP_NISTZ256_AVX2
895 /* Constant time access, loading four values, from four consecutive tables */
896 void ecp_nistz256_avx2_multi_gather_w7(void *result
, const void *in
,
897 int index0
, int index1
, int index2
,
899 void ecp_nistz256_avx2_transpose_convert(void *RESULTx4
, const void *in
);
900 void ecp_nistz256_avx2_convert_transpose_back(void *result
, const void *Ax4
);
901 void ecp_nistz256_avx2_point_add_affine_x4(void *RESULTx4
, const void *Ax4
,
903 void ecp_nistz256_avx2_point_add_affines_x4(void *RESULTx4
, const void *Ax4
,
905 void ecp_nistz256_avx2_to_mont(void *RESULTx4
, const void *Ax4
);
906 void ecp_nistz256_avx2_from_mont(void *RESULTx4
, const void *Ax4
);
907 void ecp_nistz256_avx2_set1(void *RESULTx4
);
908 int ecp_nistz_avx2_eligible(void);
910 static void booth_recode_w7(unsigned char *sign
,
911 unsigned char *digit
, unsigned char in
)
915 s
= ~((in
>> 7) - 1);
916 d
= (1 << 8) - in
- 1;
917 d
= (d
& s
) | (in
& ~s
);
918 d
= (d
>> 1) + (d
& 1);
925 * ecp_nistz256_avx2_mul_g performs multiplication by G, using only the
926 * precomputed table. It does 4 affine point additions in parallel,
927 * significantly speeding up point multiplication for a fixed value.
929 static void ecp_nistz256_avx2_mul_g(P256_POINT
*r
,
930 unsigned char p_str
[33],
931 const P256_POINT_AFFINE(*preComputedTable
)[64])
933 const unsigned int window_size
= 7;
934 const unsigned int mask
= (1 << (window_size
+ 1)) - 1;
936 /* Using 4 windows at a time */
937 unsigned char sign0
, digit0
;
938 unsigned char sign1
, digit1
;
939 unsigned char sign2
, digit2
;
940 unsigned char sign3
, digit3
;
941 unsigned int idx
= 0;
942 BN_ULONG tmp
[P256_LIMBS
];
945 ALIGN32 BN_ULONG aX4
[4 * 9 * 3] = { 0 };
946 ALIGN32 BN_ULONG bX4
[4 * 9 * 2] = { 0 };
947 ALIGN32 P256_POINT_AFFINE point_arr
[4];
948 ALIGN32 P256_POINT res_point_arr
[4];
950 /* Initial four windows */
951 wvalue
= *((u16
*) & p_str
[0]);
952 wvalue
= (wvalue
<< 1) & mask
;
954 booth_recode_w7(&sign0
, &digit0
, wvalue
);
955 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
956 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
958 booth_recode_w7(&sign1
, &digit1
, wvalue
);
959 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
960 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
962 booth_recode_w7(&sign2
, &digit2
, wvalue
);
963 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
964 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
966 booth_recode_w7(&sign3
, &digit3
, wvalue
);
968 ecp_nistz256_avx2_multi_gather_w7(point_arr
, preComputedTable
[0],
969 digit0
, digit1
, digit2
, digit3
);
971 ecp_nistz256_neg(tmp
, point_arr
[0].Y
);
972 copy_conditional(point_arr
[0].Y
, tmp
, sign0
);
973 ecp_nistz256_neg(tmp
, point_arr
[1].Y
);
974 copy_conditional(point_arr
[1].Y
, tmp
, sign1
);
975 ecp_nistz256_neg(tmp
, point_arr
[2].Y
);
976 copy_conditional(point_arr
[2].Y
, tmp
, sign2
);
977 ecp_nistz256_neg(tmp
, point_arr
[3].Y
);
978 copy_conditional(point_arr
[3].Y
, tmp
, sign3
);
980 ecp_nistz256_avx2_transpose_convert(aX4
, point_arr
);
981 ecp_nistz256_avx2_to_mont(aX4
, aX4
);
982 ecp_nistz256_avx2_to_mont(&aX4
[4 * 9], &aX4
[4 * 9]);
983 ecp_nistz256_avx2_set1(&aX4
[4 * 9 * 2]);
985 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
986 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
988 booth_recode_w7(&sign0
, &digit0
, wvalue
);
989 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
990 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
992 booth_recode_w7(&sign1
, &digit1
, wvalue
);
993 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
994 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
996 booth_recode_w7(&sign2
, &digit2
, wvalue
);
997 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
998 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1000 booth_recode_w7(&sign3
, &digit3
, wvalue
);
1002 ecp_nistz256_avx2_multi_gather_w7(point_arr
, preComputedTable
[4 * 1],
1003 digit0
, digit1
, digit2
, digit3
);
1005 ecp_nistz256_neg(tmp
, point_arr
[0].Y
);
1006 copy_conditional(point_arr
[0].Y
, tmp
, sign0
);
1007 ecp_nistz256_neg(tmp
, point_arr
[1].Y
);
1008 copy_conditional(point_arr
[1].Y
, tmp
, sign1
);
1009 ecp_nistz256_neg(tmp
, point_arr
[2].Y
);
1010 copy_conditional(point_arr
[2].Y
, tmp
, sign2
);
1011 ecp_nistz256_neg(tmp
, point_arr
[3].Y
);
1012 copy_conditional(point_arr
[3].Y
, tmp
, sign3
);
1014 ecp_nistz256_avx2_transpose_convert(bX4
, point_arr
);
1015 ecp_nistz256_avx2_to_mont(bX4
, bX4
);
1016 ecp_nistz256_avx2_to_mont(&bX4
[4 * 9], &bX4
[4 * 9]);
1017 /* Optimized when both inputs are affine */
1018 ecp_nistz256_avx2_point_add_affines_x4(aX4
, aX4
, bX4
);
1020 for (i
= 2; i
< 9; i
++) {
1021 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1022 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1024 booth_recode_w7(&sign0
, &digit0
, wvalue
);
1025 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1026 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1028 booth_recode_w7(&sign1
, &digit1
, wvalue
);
1029 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1030 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1032 booth_recode_w7(&sign2
, &digit2
, wvalue
);
1033 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1034 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1036 booth_recode_w7(&sign3
, &digit3
, wvalue
);
1038 ecp_nistz256_avx2_multi_gather_w7(point_arr
,
1039 preComputedTable
[4 * i
],
1040 digit0
, digit1
, digit2
, digit3
);
1042 ecp_nistz256_neg(tmp
, point_arr
[0].Y
);
1043 copy_conditional(point_arr
[0].Y
, tmp
, sign0
);
1044 ecp_nistz256_neg(tmp
, point_arr
[1].Y
);
1045 copy_conditional(point_arr
[1].Y
, tmp
, sign1
);
1046 ecp_nistz256_neg(tmp
, point_arr
[2].Y
);
1047 copy_conditional(point_arr
[2].Y
, tmp
, sign2
);
1048 ecp_nistz256_neg(tmp
, point_arr
[3].Y
);
1049 copy_conditional(point_arr
[3].Y
, tmp
, sign3
);
1051 ecp_nistz256_avx2_transpose_convert(bX4
, point_arr
);
1052 ecp_nistz256_avx2_to_mont(bX4
, bX4
);
1053 ecp_nistz256_avx2_to_mont(&bX4
[4 * 9], &bX4
[4 * 9]);
1055 ecp_nistz256_avx2_point_add_affine_x4(aX4
, aX4
, bX4
);
1058 ecp_nistz256_avx2_from_mont(&aX4
[4 * 9 * 0], &aX4
[4 * 9 * 0]);
1059 ecp_nistz256_avx2_from_mont(&aX4
[4 * 9 * 1], &aX4
[4 * 9 * 1]);
1060 ecp_nistz256_avx2_from_mont(&aX4
[4 * 9 * 2], &aX4
[4 * 9 * 2]);
1062 ecp_nistz256_avx2_convert_transpose_back(res_point_arr
, aX4
);
1063 /* Last window is performed serially */
1064 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1065 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1066 booth_recode_w7(&sign0
, &digit0
, wvalue
);
1067 ecp_nistz256_gather_w7((P256_POINT_AFFINE
*)r
,
1068 preComputedTable
[36], digit0
);
1069 ecp_nistz256_neg(tmp
, r
->Y
);
1070 copy_conditional(r
->Y
, tmp
, sign0
);
1071 memcpy(r
->Z
, ONE
, sizeof(ONE
));
1072 /* Sum the four windows */
1073 ecp_nistz256_point_add(r
, r
, &res_point_arr
[0]);
1074 ecp_nistz256_point_add(r
, r
, &res_point_arr
[1]);
1075 ecp_nistz256_point_add(r
, r
, &res_point_arr
[2]);
1076 ecp_nistz256_point_add(r
, r
, &res_point_arr
[3]);
1081 __owur
static int ecp_nistz256_set_from_affine(EC_POINT
*out
, const EC_GROUP
*group
,
1082 const P256_POINT_AFFINE
*in
,
1086 BN_ULONG d_x
[P256_LIMBS
], d_y
[P256_LIMBS
];
1097 memcpy(d_x
, in
->X
, sizeof(d_x
));
1098 bn_set_static_words(x
, d_x
, P256_LIMBS
);
1100 memcpy(d_y
, in
->Y
, sizeof(d_y
));
1101 bn_set_static_words(y
, d_y
, P256_LIMBS
);
1103 ret
= EC_POINT_set_affine_coordinates_GFp(group
, out
, x
, y
, ctx
);
1111 /* r = scalar*G + sum(scalars[i]*points[i]) */
1112 __owur
static int ecp_nistz256_points_mul(const EC_GROUP
*group
,
1114 const BIGNUM
*scalar
,
1116 const EC_POINT
*points
[],
1117 const BIGNUM
*scalars
[], BN_CTX
*ctx
)
1119 int i
= 0, ret
= 0, no_precomp_for_generator
= 0, p_is_infinity
= 0;
1121 unsigned char p_str
[33] = { 0 };
1122 const PRECOMP256_ROW
*preComputedTable
= NULL
;
1123 const NISTZ256_PRE_COMP
*pre_comp
= NULL
;
1124 const EC_POINT
*generator
= NULL
;
1125 BN_CTX
*new_ctx
= NULL
;
1126 const BIGNUM
**new_scalars
= NULL
;
1127 const EC_POINT
**new_points
= NULL
;
1128 unsigned int idx
= 0;
1129 const unsigned int window_size
= 7;
1130 const unsigned int mask
= (1 << (window_size
+ 1)) - 1;
1131 unsigned int wvalue
;
1134 P256_POINT_AFFINE a
;
1138 if ((num
+ 1) == 0 || (num
+ 1) > OPENSSL_MALLOC_MAX_NELEMS(void *)) {
1139 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, ERR_R_MALLOC_FAILURE
);
1143 if (group
->meth
!= r
->meth
) {
1144 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, EC_R_INCOMPATIBLE_OBJECTS
);
1148 if ((scalar
== NULL
) && (num
== 0))
1149 return EC_POINT_set_to_infinity(group
, r
);
1151 for (j
= 0; j
< num
; j
++) {
1152 if (group
->meth
!= points
[j
]->meth
) {
1153 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, EC_R_INCOMPATIBLE_OBJECTS
);
1159 ctx
= new_ctx
= BN_CTX_new();
1167 generator
= EC_GROUP_get0_generator(group
);
1168 if (generator
== NULL
) {
1169 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, EC_R_UNDEFINED_GENERATOR
);
1173 /* look if we can use precomputed multiples of generator */
1174 pre_comp
= group
->pre_comp
.nistz256
;
1178 * If there is a precomputed table for the generator, check that
1179 * it was generated with the same generator.
1181 EC_POINT
*pre_comp_generator
= EC_POINT_new(group
);
1182 if (pre_comp_generator
== NULL
)
1185 if (!ecp_nistz256_set_from_affine(pre_comp_generator
,
1186 group
, pre_comp
->precomp
[0],
1188 EC_POINT_free(pre_comp_generator
);
1192 if (0 == EC_POINT_cmp(group
, generator
, pre_comp_generator
, ctx
))
1193 preComputedTable
= (const PRECOMP256_ROW
*)pre_comp
->precomp
;
1195 EC_POINT_free(pre_comp_generator
);
1198 if (preComputedTable
== NULL
&& ecp_nistz256_is_affine_G(generator
)) {
1200 * If there is no precomputed data, but the generator is the
1201 * default, a hardcoded table of precomputed data is used. This
1202 * is because applications, such as Apache, do not use
1203 * EC_KEY_precompute_mult.
1205 preComputedTable
= ecp_nistz256_precomputed
;
1208 if (preComputedTable
) {
1209 if ((BN_num_bits(scalar
) > 256)
1210 || BN_is_negative(scalar
)) {
1211 if ((tmp_scalar
= BN_CTX_get(ctx
)) == NULL
)
1214 if (!BN_nnmod(tmp_scalar
, scalar
, group
->order
, ctx
)) {
1215 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, ERR_R_BN_LIB
);
1218 scalar
= tmp_scalar
;
1221 for (i
= 0; i
< bn_get_top(scalar
) * BN_BYTES
; i
+= BN_BYTES
) {
1222 BN_ULONG d
= bn_get_words(scalar
)[i
/ BN_BYTES
];
1224 p_str
[i
+ 0] = (unsigned char)d
;
1225 p_str
[i
+ 1] = (unsigned char)(d
>> 8);
1226 p_str
[i
+ 2] = (unsigned char)(d
>> 16);
1227 p_str
[i
+ 3] = (unsigned char)(d
>>= 24);
1228 if (BN_BYTES
== 8) {
1230 p_str
[i
+ 4] = (unsigned char)d
;
1231 p_str
[i
+ 5] = (unsigned char)(d
>> 8);
1232 p_str
[i
+ 6] = (unsigned char)(d
>> 16);
1233 p_str
[i
+ 7] = (unsigned char)(d
>> 24);
1240 #if defined(ECP_NISTZ256_AVX2)
1241 if (ecp_nistz_avx2_eligible()) {
1242 ecp_nistz256_avx2_mul_g(&p
.p
, p_str
, preComputedTable
);
1247 wvalue
= (p_str
[0] << 1) & mask
;
1250 wvalue
= _booth_recode_w7(wvalue
);
1252 ecp_nistz256_gather_w7(&p
.a
, preComputedTable
[0],
1255 ecp_nistz256_neg(p
.p
.Z
, p
.p
.Y
);
1256 copy_conditional(p
.p
.Y
, p
.p
.Z
, wvalue
& 1);
1258 memcpy(p
.p
.Z
, ONE
, sizeof(ONE
));
1260 for (i
= 1; i
< 37; i
++) {
1261 unsigned int off
= (idx
- 1) / 8;
1262 wvalue
= p_str
[off
] | p_str
[off
+ 1] << 8;
1263 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1266 wvalue
= _booth_recode_w7(wvalue
);
1268 ecp_nistz256_gather_w7(&t
.a
,
1269 preComputedTable
[i
], wvalue
>> 1);
1271 ecp_nistz256_neg(t
.p
.Z
, t
.a
.Y
);
1272 copy_conditional(t
.a
.Y
, t
.p
.Z
, wvalue
& 1);
1274 ecp_nistz256_point_add_affine(&p
.p
, &p
.p
, &t
.a
);
1279 no_precomp_for_generator
= 1;
1284 if (no_precomp_for_generator
) {
1286 * Without a precomputed table for the generator, it has to be
1287 * handled like a normal point.
1289 new_scalars
= OPENSSL_malloc((num
+ 1) * sizeof(BIGNUM
*));
1290 if (new_scalars
== NULL
) {
1291 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, ERR_R_MALLOC_FAILURE
);
1295 new_points
= OPENSSL_malloc((num
+ 1) * sizeof(EC_POINT
*));
1296 if (new_points
== NULL
) {
1297 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, ERR_R_MALLOC_FAILURE
);
1301 memcpy(new_scalars
, scalars
, num
* sizeof(BIGNUM
*));
1302 new_scalars
[num
] = scalar
;
1303 memcpy(new_points
, points
, num
* sizeof(EC_POINT
*));
1304 new_points
[num
] = generator
;
1306 scalars
= new_scalars
;
1307 points
= new_points
;
1312 P256_POINT
*out
= &t
.p
;
1316 if (!ecp_nistz256_windowed_mul(group
, out
, scalars
, points
, num
, ctx
))
1320 ecp_nistz256_point_add(&p
.p
, &p
.p
, out
);
1323 /* Not constant-time, but we're only operating on the public output. */
1324 if (!bn_set_words(r
->X
, p
.p
.X
, P256_LIMBS
) ||
1325 !bn_set_words(r
->Y
, p
.p
.Y
, P256_LIMBS
) ||
1326 !bn_set_words(r
->Z
, p
.p
.Z
, P256_LIMBS
)) {
1329 r
->Z_is_one
= is_one(p
.p
.Z
) & 1;
1336 BN_CTX_free(new_ctx
);
1337 OPENSSL_free(new_points
);
1338 OPENSSL_free(new_scalars
);
1342 __owur
static int ecp_nistz256_get_affine(const EC_GROUP
*group
,
1343 const EC_POINT
*point
,
1344 BIGNUM
*x
, BIGNUM
*y
, BN_CTX
*ctx
)
1346 BN_ULONG z_inv2
[P256_LIMBS
];
1347 BN_ULONG z_inv3
[P256_LIMBS
];
1348 BN_ULONG x_aff
[P256_LIMBS
];
1349 BN_ULONG y_aff
[P256_LIMBS
];
1350 BN_ULONG point_x
[P256_LIMBS
], point_y
[P256_LIMBS
], point_z
[P256_LIMBS
];
1351 BN_ULONG x_ret
[P256_LIMBS
], y_ret
[P256_LIMBS
];
1353 if (EC_POINT_is_at_infinity(group
, point
)) {
1354 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE
, EC_R_POINT_AT_INFINITY
);
1358 if (!ecp_nistz256_bignum_to_field_elem(point_x
, point
->X
) ||
1359 !ecp_nistz256_bignum_to_field_elem(point_y
, point
->Y
) ||
1360 !ecp_nistz256_bignum_to_field_elem(point_z
, point
->Z
)) {
1361 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE
, EC_R_COORDINATES_OUT_OF_RANGE
);
1365 ecp_nistz256_mod_inverse(z_inv3
, point_z
);
1366 ecp_nistz256_sqr_mont(z_inv2
, z_inv3
);
1367 ecp_nistz256_mul_mont(x_aff
, z_inv2
, point_x
);
1370 ecp_nistz256_from_mont(x_ret
, x_aff
);
1371 if (!bn_set_words(x
, x_ret
, P256_LIMBS
))
1376 ecp_nistz256_mul_mont(z_inv3
, z_inv3
, z_inv2
);
1377 ecp_nistz256_mul_mont(y_aff
, z_inv3
, point_y
);
1378 ecp_nistz256_from_mont(y_ret
, y_aff
);
1379 if (!bn_set_words(y
, y_ret
, P256_LIMBS
))
1386 static NISTZ256_PRE_COMP
*ecp_nistz256_pre_comp_new(const EC_GROUP
*group
)
1388 NISTZ256_PRE_COMP
*ret
= NULL
;
1393 ret
= OPENSSL_zalloc(sizeof(*ret
));
1396 ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW
, ERR_R_MALLOC_FAILURE
);
1401 ret
->w
= 6; /* default */
1402 ret
->precomp
= NULL
;
1403 ret
->precomp_storage
= NULL
;
1404 ret
->references
= 1;
1408 NISTZ256_PRE_COMP
*EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP
*p
)
1411 CRYPTO_add(&p
->references
, 1, CRYPTO_LOCK_EC_PRE_COMP
);
1415 void EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP
*pre
)
1418 || CRYPTO_add(&pre
->references
, -1, CRYPTO_LOCK_EC_PRE_COMP
) > 0)
1420 OPENSSL_free(pre
->precomp_storage
);
1425 static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP
*group
)
1427 /* There is a hard-coded table for the default generator. */
1428 const EC_POINT
*generator
= EC_GROUP_get0_generator(group
);
1430 if (generator
!= NULL
&& ecp_nistz256_is_affine_G(generator
)) {
1431 /* There is a hard-coded table for the default generator. */
1435 return HAVEPRECOMP(group
, nistz256
);
1438 const EC_METHOD
*EC_GFp_nistz256_method(void)
1440 static const EC_METHOD ret
= {
1441 EC_FLAGS_DEFAULT_OCT
,
1442 NID_X9_62_prime_field
,
1443 ec_GFp_mont_group_init
,
1444 ec_GFp_mont_group_finish
,
1445 ec_GFp_mont_group_clear_finish
,
1446 ec_GFp_mont_group_copy
,
1447 ec_GFp_mont_group_set_curve
,
1448 ec_GFp_simple_group_get_curve
,
1449 ec_GFp_simple_group_get_degree
,
1450 ec_GFp_simple_group_check_discriminant
,
1451 ec_GFp_simple_point_init
,
1452 ec_GFp_simple_point_finish
,
1453 ec_GFp_simple_point_clear_finish
,
1454 ec_GFp_simple_point_copy
,
1455 ec_GFp_simple_point_set_to_infinity
,
1456 ec_GFp_simple_set_Jprojective_coordinates_GFp
,
1457 ec_GFp_simple_get_Jprojective_coordinates_GFp
,
1458 ec_GFp_simple_point_set_affine_coordinates
,
1459 ecp_nistz256_get_affine
,
1463 ec_GFp_simple_invert
,
1464 ec_GFp_simple_is_at_infinity
,
1465 ec_GFp_simple_is_on_curve
,
1467 ec_GFp_simple_make_affine
,
1468 ec_GFp_simple_points_make_affine
,
1469 ecp_nistz256_points_mul
, /* mul */
1470 ecp_nistz256_mult_precompute
, /* precompute_mult */
1471 ecp_nistz256_window_have_precompute_mult
, /* have_precompute_mult */
1472 ec_GFp_mont_field_mul
,
1473 ec_GFp_mont_field_sqr
,
1475 ec_GFp_mont_field_encode
,
1476 ec_GFp_mont_field_decode
,
1477 ec_GFp_mont_field_set_to_one