2 * Copyright 2014-2016 The OpenSSL Project Authors. All Rights Reserved.
4 * Licensed under the OpenSSL license (the "License"). You may not use
5 * this file except in compliance with the License. You can obtain a copy
6 * in the file LICENSE in the source distribution or at
7 * https://www.openssl.org/source/license.html
10 /******************************************************************************
12 * Copyright 2014 Intel Corporation *
14 * Licensed under the Apache License, Version 2.0 (the "License"); *
15 * you may not use this file except in compliance with the License. *
16 * You may obtain a copy of the License at *
18 * http://www.apache.org/licenses/LICENSE-2.0 *
20 * Unless required by applicable law or agreed to in writing, software *
21 * distributed under the License is distributed on an "AS IS" BASIS, *
22 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. *
23 * See the License for the specific language governing permissions and *
24 * limitations under the License. *
26 ******************************************************************************
28 * Developers and authors: *
29 * Shay Gueron (1, 2), and Vlad Krasnov (1) *
30 * (1) Intel Corporation, Israel Development Center *
31 * (2) University of Haifa *
33 * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with *
36 ******************************************************************************/
40 #include "internal/cryptlib.h"
41 #include "internal/bn_int.h"
45 # define TOBN(hi,lo) lo,hi
47 # define TOBN(hi,lo) ((BN_ULONG)hi<<32|lo)
51 # define ALIGN32 __attribute((aligned(32)))
52 #elif defined(_MSC_VER)
53 # define ALIGN32 __declspec(align(32))
58 #define ALIGNPTR(p,N) ((unsigned char *)p+N-(size_t)p%N)
59 #define P256_LIMBS (256/BN_BITS2)
61 typedef unsigned short u16
;
64 BN_ULONG X
[P256_LIMBS
];
65 BN_ULONG Y
[P256_LIMBS
];
66 BN_ULONG Z
[P256_LIMBS
];
70 BN_ULONG X
[P256_LIMBS
];
71 BN_ULONG Y
[P256_LIMBS
];
74 typedef P256_POINT_AFFINE PRECOMP256_ROW
[64];
76 /* structure for precomputed multiples of the generator */
77 struct nistz256_pre_comp_st
{
78 const EC_GROUP
*group
; /* Parent EC_GROUP object */
79 size_t w
; /* Window size */
81 * Constant time access to the X and Y coordinates of the pre-computed,
82 * generator multiplies, in the Montgomery domain. Pre-calculated
83 * multiplies are stored in affine form.
85 PRECOMP256_ROW
*precomp
;
86 void *precomp_storage
;
91 /* Functions implemented in assembly */
92 /* Modular mul by 2: res = 2*a mod P */
93 void ecp_nistz256_mul_by_2(BN_ULONG res
[P256_LIMBS
],
94 const BN_ULONG a
[P256_LIMBS
]);
95 /* Modular div by 2: res = a/2 mod P */
96 void ecp_nistz256_div_by_2(BN_ULONG res
[P256_LIMBS
],
97 const BN_ULONG a
[P256_LIMBS
]);
98 /* Modular mul by 3: res = 3*a mod P */
99 void ecp_nistz256_mul_by_3(BN_ULONG res
[P256_LIMBS
],
100 const BN_ULONG a
[P256_LIMBS
]);
101 /* Modular add: res = a+b mod P */
102 void ecp_nistz256_add(BN_ULONG res
[P256_LIMBS
],
103 const BN_ULONG a
[P256_LIMBS
],
104 const BN_ULONG b
[P256_LIMBS
]);
105 /* Modular sub: res = a-b mod P */
106 void ecp_nistz256_sub(BN_ULONG res
[P256_LIMBS
],
107 const BN_ULONG a
[P256_LIMBS
],
108 const BN_ULONG b
[P256_LIMBS
]);
109 /* Modular neg: res = -a mod P */
110 void ecp_nistz256_neg(BN_ULONG res
[P256_LIMBS
], const BN_ULONG a
[P256_LIMBS
]);
111 /* Montgomery mul: res = a*b*2^-256 mod P */
112 void ecp_nistz256_mul_mont(BN_ULONG res
[P256_LIMBS
],
113 const BN_ULONG a
[P256_LIMBS
],
114 const BN_ULONG b
[P256_LIMBS
]);
115 /* Montgomery sqr: res = a*a*2^-256 mod P */
116 void ecp_nistz256_sqr_mont(BN_ULONG res
[P256_LIMBS
],
117 const BN_ULONG a
[P256_LIMBS
]);
118 /* Convert a number from Montgomery domain, by multiplying with 1 */
119 void ecp_nistz256_from_mont(BN_ULONG res
[P256_LIMBS
],
120 const BN_ULONG in
[P256_LIMBS
]);
121 /* Convert a number to Montgomery domain, by multiplying with 2^512 mod P*/
122 void ecp_nistz256_to_mont(BN_ULONG res
[P256_LIMBS
],
123 const BN_ULONG in
[P256_LIMBS
]);
124 /* Functions that perform constant time access to the precomputed tables */
125 void ecp_nistz256_scatter_w5(P256_POINT
*val
,
126 const P256_POINT
*in_t
, int idx
);
127 void ecp_nistz256_gather_w5(P256_POINT
*val
,
128 const P256_POINT
*in_t
, int idx
);
129 void ecp_nistz256_scatter_w7(P256_POINT_AFFINE
*val
,
130 const P256_POINT_AFFINE
*in_t
, int idx
);
131 void ecp_nistz256_gather_w7(P256_POINT_AFFINE
*val
,
132 const P256_POINT_AFFINE
*in_t
, int idx
);
134 /* One converted into the Montgomery domain */
135 static const BN_ULONG ONE
[P256_LIMBS
] = {
136 TOBN(0x00000000, 0x00000001), TOBN(0xffffffff, 0x00000000),
137 TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe)
140 static NISTZ256_PRE_COMP
*ecp_nistz256_pre_comp_new(const EC_GROUP
*group
);
142 /* Precomputed tables for the default generator */
143 extern const PRECOMP256_ROW ecp_nistz256_precomputed
[37];
145 /* Recode window to a signed digit, see ecp_nistputil.c for details */
146 static unsigned int _booth_recode_w5(unsigned int in
)
150 s
= ~((in
>> 5) - 1);
151 d
= (1 << 6) - in
- 1;
152 d
= (d
& s
) | (in
& ~s
);
153 d
= (d
>> 1) + (d
& 1);
155 return (d
<< 1) + (s
& 1);
158 static unsigned int _booth_recode_w7(unsigned int in
)
162 s
= ~((in
>> 7) - 1);
163 d
= (1 << 8) - in
- 1;
164 d
= (d
& s
) | (in
& ~s
);
165 d
= (d
>> 1) + (d
& 1);
167 return (d
<< 1) + (s
& 1);
170 static void copy_conditional(BN_ULONG dst
[P256_LIMBS
],
171 const BN_ULONG src
[P256_LIMBS
], BN_ULONG move
)
173 BN_ULONG mask1
= 0-move
;
174 BN_ULONG mask2
= ~mask1
;
176 dst
[0] = (src
[0] & mask1
) ^ (dst
[0] & mask2
);
177 dst
[1] = (src
[1] & mask1
) ^ (dst
[1] & mask2
);
178 dst
[2] = (src
[2] & mask1
) ^ (dst
[2] & mask2
);
179 dst
[3] = (src
[3] & mask1
) ^ (dst
[3] & mask2
);
180 if (P256_LIMBS
== 8) {
181 dst
[4] = (src
[4] & mask1
) ^ (dst
[4] & mask2
);
182 dst
[5] = (src
[5] & mask1
) ^ (dst
[5] & mask2
);
183 dst
[6] = (src
[6] & mask1
) ^ (dst
[6] & mask2
);
184 dst
[7] = (src
[7] & mask1
) ^ (dst
[7] & mask2
);
188 static BN_ULONG
is_zero(BN_ULONG in
)
196 static BN_ULONG
is_equal(const BN_ULONG a
[P256_LIMBS
],
197 const BN_ULONG b
[P256_LIMBS
])
205 if (P256_LIMBS
== 8) {
215 static BN_ULONG
is_one(const BN_ULONG a
[P256_LIMBS
])
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];
232 #ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
233 void ecp_nistz256_point_double(P256_POINT
*r
, const P256_POINT
*a
);
234 void ecp_nistz256_point_add(P256_POINT
*r
,
235 const P256_POINT
*a
, const P256_POINT
*b
);
236 void ecp_nistz256_point_add_affine(P256_POINT
*r
,
238 const P256_POINT_AFFINE
*b
);
240 /* Point double: r = 2*a */
241 static void ecp_nistz256_point_double(P256_POINT
*r
, const P256_POINT
*a
)
243 BN_ULONG S
[P256_LIMBS
];
244 BN_ULONG M
[P256_LIMBS
];
245 BN_ULONG Zsqr
[P256_LIMBS
];
246 BN_ULONG tmp0
[P256_LIMBS
];
248 const BN_ULONG
*in_x
= a
->X
;
249 const BN_ULONG
*in_y
= a
->Y
;
250 const BN_ULONG
*in_z
= a
->Z
;
252 BN_ULONG
*res_x
= r
->X
;
253 BN_ULONG
*res_y
= r
->Y
;
254 BN_ULONG
*res_z
= r
->Z
;
256 ecp_nistz256_mul_by_2(S
, in_y
);
258 ecp_nistz256_sqr_mont(Zsqr
, in_z
);
260 ecp_nistz256_sqr_mont(S
, S
);
262 ecp_nistz256_mul_mont(res_z
, in_z
, in_y
);
263 ecp_nistz256_mul_by_2(res_z
, res_z
);
265 ecp_nistz256_add(M
, in_x
, Zsqr
);
266 ecp_nistz256_sub(Zsqr
, in_x
, Zsqr
);
268 ecp_nistz256_sqr_mont(res_y
, S
);
269 ecp_nistz256_div_by_2(res_y
, res_y
);
271 ecp_nistz256_mul_mont(M
, M
, Zsqr
);
272 ecp_nistz256_mul_by_3(M
, M
);
274 ecp_nistz256_mul_mont(S
, S
, in_x
);
275 ecp_nistz256_mul_by_2(tmp0
, S
);
277 ecp_nistz256_sqr_mont(res_x
, M
);
279 ecp_nistz256_sub(res_x
, res_x
, tmp0
);
280 ecp_nistz256_sub(S
, S
, res_x
);
282 ecp_nistz256_mul_mont(S
, S
, M
);
283 ecp_nistz256_sub(res_y
, S
, res_y
);
286 /* Point addition: r = a+b */
287 static void ecp_nistz256_point_add(P256_POINT
*r
,
288 const P256_POINT
*a
, const P256_POINT
*b
)
290 BN_ULONG U2
[P256_LIMBS
], S2
[P256_LIMBS
];
291 BN_ULONG U1
[P256_LIMBS
], S1
[P256_LIMBS
];
292 BN_ULONG Z1sqr
[P256_LIMBS
];
293 BN_ULONG Z2sqr
[P256_LIMBS
];
294 BN_ULONG H
[P256_LIMBS
], R
[P256_LIMBS
];
295 BN_ULONG Hsqr
[P256_LIMBS
];
296 BN_ULONG Rsqr
[P256_LIMBS
];
297 BN_ULONG Hcub
[P256_LIMBS
];
299 BN_ULONG res_x
[P256_LIMBS
];
300 BN_ULONG res_y
[P256_LIMBS
];
301 BN_ULONG res_z
[P256_LIMBS
];
303 BN_ULONG in1infty
, in2infty
;
305 const BN_ULONG
*in1_x
= a
->X
;
306 const BN_ULONG
*in1_y
= a
->Y
;
307 const BN_ULONG
*in1_z
= a
->Z
;
309 const BN_ULONG
*in2_x
= b
->X
;
310 const BN_ULONG
*in2_y
= b
->Y
;
311 const BN_ULONG
*in2_z
= b
->Z
;
313 /* We encode infinity as (0,0), which is not on the curve,
315 in1infty
= (in1_x
[0] | in1_x
[1] | in1_x
[2] | in1_x
[3] |
316 in1_y
[0] | in1_y
[1] | in1_y
[2] | in1_y
[3]);
318 in1infty
|= (in1_x
[4] | in1_x
[5] | in1_x
[6] | in1_x
[7] |
319 in1_y
[4] | in1_y
[5] | in1_y
[6] | in1_y
[7]);
321 in2infty
= (in2_x
[0] | in2_x
[1] | in2_x
[2] | in2_x
[3] |
322 in2_y
[0] | in2_y
[1] | in2_y
[2] | in2_y
[3]);
324 in2infty
|= (in2_x
[4] | in2_x
[5] | in2_x
[6] | in2_x
[7] |
325 in2_y
[4] | in2_y
[5] | in2_y
[6] | in2_y
[7]);
327 in1infty
= is_zero(in1infty
);
328 in2infty
= is_zero(in2infty
);
330 ecp_nistz256_sqr_mont(Z2sqr
, in2_z
); /* Z2^2 */
331 ecp_nistz256_sqr_mont(Z1sqr
, in1_z
); /* Z1^2 */
333 ecp_nistz256_mul_mont(S1
, Z2sqr
, in2_z
); /* S1 = Z2^3 */
334 ecp_nistz256_mul_mont(S2
, Z1sqr
, in1_z
); /* S2 = Z1^3 */
336 ecp_nistz256_mul_mont(S1
, S1
, in1_y
); /* S1 = Y1*Z2^3 */
337 ecp_nistz256_mul_mont(S2
, S2
, in2_y
); /* S2 = Y2*Z1^3 */
338 ecp_nistz256_sub(R
, S2
, S1
); /* R = S2 - S1 */
340 ecp_nistz256_mul_mont(U1
, in1_x
, Z2sqr
); /* U1 = X1*Z2^2 */
341 ecp_nistz256_mul_mont(U2
, in2_x
, Z1sqr
); /* U2 = X2*Z1^2 */
342 ecp_nistz256_sub(H
, U2
, U1
); /* H = U2 - U1 */
345 * This should not happen during sign/ecdh, so no constant time violation
347 if (is_equal(U1
, U2
) && !in1infty
&& !in2infty
) {
348 if (is_equal(S1
, S2
)) {
349 ecp_nistz256_point_double(r
, a
);
352 memset(r
, 0, sizeof(*r
));
357 ecp_nistz256_sqr_mont(Rsqr
, R
); /* R^2 */
358 ecp_nistz256_mul_mont(res_z
, H
, in1_z
); /* Z3 = H*Z1*Z2 */
359 ecp_nistz256_sqr_mont(Hsqr
, H
); /* H^2 */
360 ecp_nistz256_mul_mont(res_z
, res_z
, in2_z
); /* Z3 = H*Z1*Z2 */
361 ecp_nistz256_mul_mont(Hcub
, Hsqr
, H
); /* H^3 */
363 ecp_nistz256_mul_mont(U2
, U1
, Hsqr
); /* U1*H^2 */
364 ecp_nistz256_mul_by_2(Hsqr
, U2
); /* 2*U1*H^2 */
366 ecp_nistz256_sub(res_x
, Rsqr
, Hsqr
);
367 ecp_nistz256_sub(res_x
, res_x
, Hcub
);
369 ecp_nistz256_sub(res_y
, U2
, res_x
);
371 ecp_nistz256_mul_mont(S2
, S1
, Hcub
);
372 ecp_nistz256_mul_mont(res_y
, R
, res_y
);
373 ecp_nistz256_sub(res_y
, res_y
, S2
);
375 copy_conditional(res_x
, in2_x
, in1infty
);
376 copy_conditional(res_y
, in2_y
, in1infty
);
377 copy_conditional(res_z
, in2_z
, in1infty
);
379 copy_conditional(res_x
, in1_x
, in2infty
);
380 copy_conditional(res_y
, in1_y
, in2infty
);
381 copy_conditional(res_z
, in1_z
, in2infty
);
383 memcpy(r
->X
, res_x
, sizeof(res_x
));
384 memcpy(r
->Y
, res_y
, sizeof(res_y
));
385 memcpy(r
->Z
, res_z
, sizeof(res_z
));
388 /* Point addition when b is known to be affine: r = a+b */
389 static void ecp_nistz256_point_add_affine(P256_POINT
*r
,
391 const P256_POINT_AFFINE
*b
)
393 BN_ULONG U2
[P256_LIMBS
], S2
[P256_LIMBS
];
394 BN_ULONG Z1sqr
[P256_LIMBS
];
395 BN_ULONG H
[P256_LIMBS
], R
[P256_LIMBS
];
396 BN_ULONG Hsqr
[P256_LIMBS
];
397 BN_ULONG Rsqr
[P256_LIMBS
];
398 BN_ULONG Hcub
[P256_LIMBS
];
400 BN_ULONG res_x
[P256_LIMBS
];
401 BN_ULONG res_y
[P256_LIMBS
];
402 BN_ULONG res_z
[P256_LIMBS
];
404 BN_ULONG in1infty
, in2infty
;
406 const BN_ULONG
*in1_x
= a
->X
;
407 const BN_ULONG
*in1_y
= a
->Y
;
408 const BN_ULONG
*in1_z
= a
->Z
;
410 const BN_ULONG
*in2_x
= b
->X
;
411 const BN_ULONG
*in2_y
= b
->Y
;
414 * In affine representation we encode infty as (0,0), which is not on the
417 in1infty
= (in1_x
[0] | in1_x
[1] | in1_x
[2] | in1_x
[3] |
418 in1_y
[0] | in1_y
[1] | in1_y
[2] | in1_y
[3]);
420 in1infty
|= (in1_x
[4] | in1_x
[5] | in1_x
[6] | in1_x
[7] |
421 in1_y
[4] | in1_y
[5] | in1_y
[6] | in1_y
[7]);
423 in2infty
= (in2_x
[0] | in2_x
[1] | in2_x
[2] | in2_x
[3] |
424 in2_y
[0] | in2_y
[1] | in2_y
[2] | in2_y
[3]);
426 in2infty
|= (in2_x
[4] | in2_x
[5] | in2_x
[6] | in2_x
[7] |
427 in2_y
[4] | in2_y
[5] | in2_y
[6] | in2_y
[7]);
429 in1infty
= is_zero(in1infty
);
430 in2infty
= is_zero(in2infty
);
432 ecp_nistz256_sqr_mont(Z1sqr
, in1_z
); /* Z1^2 */
434 ecp_nistz256_mul_mont(U2
, in2_x
, Z1sqr
); /* U2 = X2*Z1^2 */
435 ecp_nistz256_sub(H
, U2
, in1_x
); /* H = U2 - U1 */
437 ecp_nistz256_mul_mont(S2
, Z1sqr
, in1_z
); /* S2 = Z1^3 */
439 ecp_nistz256_mul_mont(res_z
, H
, in1_z
); /* Z3 = H*Z1*Z2 */
441 ecp_nistz256_mul_mont(S2
, S2
, in2_y
); /* S2 = Y2*Z1^3 */
442 ecp_nistz256_sub(R
, S2
, in1_y
); /* R = S2 - S1 */
444 ecp_nistz256_sqr_mont(Hsqr
, H
); /* H^2 */
445 ecp_nistz256_sqr_mont(Rsqr
, R
); /* R^2 */
446 ecp_nistz256_mul_mont(Hcub
, Hsqr
, H
); /* H^3 */
448 ecp_nistz256_mul_mont(U2
, in1_x
, Hsqr
); /* U1*H^2 */
449 ecp_nistz256_mul_by_2(Hsqr
, U2
); /* 2*U1*H^2 */
451 ecp_nistz256_sub(res_x
, Rsqr
, Hsqr
);
452 ecp_nistz256_sub(res_x
, res_x
, Hcub
);
453 ecp_nistz256_sub(H
, U2
, res_x
);
455 ecp_nistz256_mul_mont(S2
, in1_y
, Hcub
);
456 ecp_nistz256_mul_mont(H
, H
, R
);
457 ecp_nistz256_sub(res_y
, H
, S2
);
459 copy_conditional(res_x
, in2_x
, in1infty
);
460 copy_conditional(res_x
, in1_x
, in2infty
);
462 copy_conditional(res_y
, in2_y
, in1infty
);
463 copy_conditional(res_y
, in1_y
, in2infty
);
465 copy_conditional(res_z
, ONE
, in1infty
);
466 copy_conditional(res_z
, in1_z
, in2infty
);
468 memcpy(r
->X
, res_x
, sizeof(res_x
));
469 memcpy(r
->Y
, res_y
, sizeof(res_y
));
470 memcpy(r
->Z
, res_z
, sizeof(res_z
));
474 /* r = in^-1 mod p */
475 static void ecp_nistz256_mod_inverse(BN_ULONG r
[P256_LIMBS
],
476 const BN_ULONG in
[P256_LIMBS
])
479 * The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff
480 * ffffffff ffffffff We use FLT and used poly-2 as exponent
482 BN_ULONG p2
[P256_LIMBS
];
483 BN_ULONG p4
[P256_LIMBS
];
484 BN_ULONG p8
[P256_LIMBS
];
485 BN_ULONG p16
[P256_LIMBS
];
486 BN_ULONG p32
[P256_LIMBS
];
487 BN_ULONG res
[P256_LIMBS
];
490 ecp_nistz256_sqr_mont(res
, in
);
491 ecp_nistz256_mul_mont(p2
, res
, in
); /* 3*p */
493 ecp_nistz256_sqr_mont(res
, p2
);
494 ecp_nistz256_sqr_mont(res
, res
);
495 ecp_nistz256_mul_mont(p4
, res
, p2
); /* f*p */
497 ecp_nistz256_sqr_mont(res
, p4
);
498 ecp_nistz256_sqr_mont(res
, res
);
499 ecp_nistz256_sqr_mont(res
, res
);
500 ecp_nistz256_sqr_mont(res
, res
);
501 ecp_nistz256_mul_mont(p8
, res
, p4
); /* ff*p */
503 ecp_nistz256_sqr_mont(res
, p8
);
504 for (i
= 0; i
< 7; i
++)
505 ecp_nistz256_sqr_mont(res
, res
);
506 ecp_nistz256_mul_mont(p16
, res
, p8
); /* ffff*p */
508 ecp_nistz256_sqr_mont(res
, p16
);
509 for (i
= 0; i
< 15; i
++)
510 ecp_nistz256_sqr_mont(res
, res
);
511 ecp_nistz256_mul_mont(p32
, res
, p16
); /* ffffffff*p */
513 ecp_nistz256_sqr_mont(res
, p32
);
514 for (i
= 0; i
< 31; i
++)
515 ecp_nistz256_sqr_mont(res
, res
);
516 ecp_nistz256_mul_mont(res
, res
, in
);
518 for (i
= 0; i
< 32 * 4; i
++)
519 ecp_nistz256_sqr_mont(res
, res
);
520 ecp_nistz256_mul_mont(res
, res
, p32
);
522 for (i
= 0; i
< 32; i
++)
523 ecp_nistz256_sqr_mont(res
, res
);
524 ecp_nistz256_mul_mont(res
, res
, p32
);
526 for (i
= 0; i
< 16; i
++)
527 ecp_nistz256_sqr_mont(res
, res
);
528 ecp_nistz256_mul_mont(res
, res
, p16
);
530 for (i
= 0; i
< 8; i
++)
531 ecp_nistz256_sqr_mont(res
, res
);
532 ecp_nistz256_mul_mont(res
, res
, p8
);
534 ecp_nistz256_sqr_mont(res
, res
);
535 ecp_nistz256_sqr_mont(res
, res
);
536 ecp_nistz256_sqr_mont(res
, res
);
537 ecp_nistz256_sqr_mont(res
, res
);
538 ecp_nistz256_mul_mont(res
, res
, p4
);
540 ecp_nistz256_sqr_mont(res
, res
);
541 ecp_nistz256_sqr_mont(res
, res
);
542 ecp_nistz256_mul_mont(res
, res
, p2
);
544 ecp_nistz256_sqr_mont(res
, res
);
545 ecp_nistz256_sqr_mont(res
, res
);
546 ecp_nistz256_mul_mont(res
, res
, in
);
548 memcpy(r
, res
, sizeof(res
));
552 * ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
553 * returns one if it fits. Otherwise it returns zero.
555 __owur
static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out
[P256_LIMBS
],
558 return bn_copy_words(out
, in
, P256_LIMBS
);
561 /* r = sum(scalar[i]*point[i]) */
562 __owur
static int ecp_nistz256_windowed_mul(const EC_GROUP
*group
,
564 const BIGNUM
**scalar
,
565 const EC_POINT
**point
,
566 size_t num
, BN_CTX
*ctx
)
571 unsigned char (*p_str
)[33] = NULL
;
572 const unsigned int window_size
= 5;
573 const unsigned int mask
= (1 << (window_size
+ 1)) - 1;
575 P256_POINT
*temp
; /* place for 5 temporary points */
576 const BIGNUM
**scalars
= NULL
;
577 P256_POINT (*table
)[16] = NULL
;
578 void *table_storage
= NULL
;
580 if ((num
* 16 + 6) > OPENSSL_MALLOC_MAX_NELEMS(P256_POINT
)
582 OPENSSL_malloc((num
* 16 + 5) * sizeof(P256_POINT
) + 64)) == NULL
584 OPENSSL_malloc(num
* 33 * sizeof(unsigned char))) == NULL
585 || (scalars
= OPENSSL_malloc(num
* sizeof(BIGNUM
*))) == NULL
) {
586 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL
, ERR_R_MALLOC_FAILURE
);
590 table
= (void *)ALIGNPTR(table_storage
, 64);
591 temp
= (P256_POINT
*)(table
+ num
);
593 for (i
= 0; i
< num
; i
++) {
594 P256_POINT
*row
= table
[i
];
596 /* This is an unusual input, we don't guarantee constant-timeness. */
597 if ((BN_num_bits(scalar
[i
]) > 256) || BN_is_negative(scalar
[i
])) {
600 if ((mod
= BN_CTX_get(ctx
)) == NULL
)
602 if (!BN_nnmod(mod
, scalar
[i
], group
->order
, ctx
)) {
603 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL
, ERR_R_BN_LIB
);
608 scalars
[i
] = scalar
[i
];
610 for (j
= 0; j
< bn_get_top(scalars
[i
]) * BN_BYTES
; j
+= BN_BYTES
) {
611 BN_ULONG d
= bn_get_words(scalars
[i
])[j
/ BN_BYTES
];
613 p_str
[i
][j
+ 0] = (unsigned char)d
;
614 p_str
[i
][j
+ 1] = (unsigned char)(d
>> 8);
615 p_str
[i
][j
+ 2] = (unsigned char)(d
>> 16);
616 p_str
[i
][j
+ 3] = (unsigned char)(d
>>= 24);
619 p_str
[i
][j
+ 4] = (unsigned char)d
;
620 p_str
[i
][j
+ 5] = (unsigned char)(d
>> 8);
621 p_str
[i
][j
+ 6] = (unsigned char)(d
>> 16);
622 p_str
[i
][j
+ 7] = (unsigned char)(d
>> 24);
628 if (!ecp_nistz256_bignum_to_field_elem(temp
[0].X
, point
[i
]->X
)
629 || !ecp_nistz256_bignum_to_field_elem(temp
[0].Y
, point
[i
]->Y
)
630 || !ecp_nistz256_bignum_to_field_elem(temp
[0].Z
, point
[i
]->Z
)) {
631 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL
,
632 EC_R_COORDINATES_OUT_OF_RANGE
);
637 * row[0] is implicitly (0,0,0) (the point at infinity), therefore it
638 * is not stored. All other values are actually stored with an offset
642 ecp_nistz256_scatter_w5 (row
, &temp
[0], 1);
643 ecp_nistz256_point_double(&temp
[1], &temp
[0]); /*1+1=2 */
644 ecp_nistz256_scatter_w5 (row
, &temp
[1], 2);
645 ecp_nistz256_point_add (&temp
[2], &temp
[1], &temp
[0]); /*2+1=3 */
646 ecp_nistz256_scatter_w5 (row
, &temp
[2], 3);
647 ecp_nistz256_point_double(&temp
[1], &temp
[1]); /*2*2=4 */
648 ecp_nistz256_scatter_w5 (row
, &temp
[1], 4);
649 ecp_nistz256_point_double(&temp
[2], &temp
[2]); /*2*3=6 */
650 ecp_nistz256_scatter_w5 (row
, &temp
[2], 6);
651 ecp_nistz256_point_add (&temp
[3], &temp
[1], &temp
[0]); /*4+1=5 */
652 ecp_nistz256_scatter_w5 (row
, &temp
[3], 5);
653 ecp_nistz256_point_add (&temp
[4], &temp
[2], &temp
[0]); /*6+1=7 */
654 ecp_nistz256_scatter_w5 (row
, &temp
[4], 7);
655 ecp_nistz256_point_double(&temp
[1], &temp
[1]); /*2*4=8 */
656 ecp_nistz256_scatter_w5 (row
, &temp
[1], 8);
657 ecp_nistz256_point_double(&temp
[2], &temp
[2]); /*2*6=12 */
658 ecp_nistz256_scatter_w5 (row
, &temp
[2], 12);
659 ecp_nistz256_point_double(&temp
[3], &temp
[3]); /*2*5=10 */
660 ecp_nistz256_scatter_w5 (row
, &temp
[3], 10);
661 ecp_nistz256_point_double(&temp
[4], &temp
[4]); /*2*7=14 */
662 ecp_nistz256_scatter_w5 (row
, &temp
[4], 14);
663 ecp_nistz256_point_add (&temp
[2], &temp
[2], &temp
[0]); /*12+1=13*/
664 ecp_nistz256_scatter_w5 (row
, &temp
[2], 13);
665 ecp_nistz256_point_add (&temp
[3], &temp
[3], &temp
[0]); /*10+1=11*/
666 ecp_nistz256_scatter_w5 (row
, &temp
[3], 11);
667 ecp_nistz256_point_add (&temp
[4], &temp
[4], &temp
[0]); /*14+1=15*/
668 ecp_nistz256_scatter_w5 (row
, &temp
[4], 15);
669 ecp_nistz256_point_add (&temp
[2], &temp
[1], &temp
[0]); /*8+1=9 */
670 ecp_nistz256_scatter_w5 (row
, &temp
[2], 9);
671 ecp_nistz256_point_double(&temp
[1], &temp
[1]); /*2*8=16 */
672 ecp_nistz256_scatter_w5 (row
, &temp
[1], 16);
677 wvalue
= p_str
[0][(idx
- 1) / 8];
678 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
681 * We gather to temp[0], because we know it's position relative
684 ecp_nistz256_gather_w5(&temp
[0], table
[0], _booth_recode_w5(wvalue
) >> 1);
685 memcpy(r
, &temp
[0], sizeof(temp
[0]));
688 for (i
= (idx
== 255 ? 1 : 0); i
< num
; i
++) {
689 unsigned int off
= (idx
- 1) / 8;
691 wvalue
= p_str
[i
][off
] | p_str
[i
][off
+ 1] << 8;
692 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
694 wvalue
= _booth_recode_w5(wvalue
);
696 ecp_nistz256_gather_w5(&temp
[0], table
[i
], wvalue
>> 1);
698 ecp_nistz256_neg(temp
[1].Y
, temp
[0].Y
);
699 copy_conditional(temp
[0].Y
, temp
[1].Y
, (wvalue
& 1));
701 ecp_nistz256_point_add(r
, r
, &temp
[0]);
706 ecp_nistz256_point_double(r
, r
);
707 ecp_nistz256_point_double(r
, r
);
708 ecp_nistz256_point_double(r
, r
);
709 ecp_nistz256_point_double(r
, r
);
710 ecp_nistz256_point_double(r
, r
);
714 for (i
= 0; i
< num
; i
++) {
715 wvalue
= p_str
[i
][0];
716 wvalue
= (wvalue
<< 1) & mask
;
718 wvalue
= _booth_recode_w5(wvalue
);
720 ecp_nistz256_gather_w5(&temp
[0], table
[i
], wvalue
>> 1);
722 ecp_nistz256_neg(temp
[1].Y
, temp
[0].Y
);
723 copy_conditional(temp
[0].Y
, temp
[1].Y
, wvalue
& 1);
725 ecp_nistz256_point_add(r
, r
, &temp
[0]);
730 OPENSSL_free(table_storage
);
732 OPENSSL_free(scalars
);
736 /* Coordinates of G, for which we have precomputed tables */
737 const static BN_ULONG def_xG
[P256_LIMBS
] = {
738 TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
739 TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
742 const static BN_ULONG def_yG
[P256_LIMBS
] = {
743 TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
744 TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
748 * ecp_nistz256_is_affine_G returns one if |generator| is the standard, P-256
751 static int ecp_nistz256_is_affine_G(const EC_POINT
*generator
)
753 return (bn_get_top(generator
->X
) == P256_LIMBS
) &&
754 (bn_get_top(generator
->Y
) == P256_LIMBS
) &&
755 (bn_get_top(generator
->Z
) == (P256_LIMBS
- P256_LIMBS
/ 8)) &&
756 is_equal(bn_get_words(generator
->X
), def_xG
) &&
757 is_equal(bn_get_words(generator
->Y
), def_yG
) &&
758 is_one(bn_get_words(generator
->Z
));
761 __owur
static int ecp_nistz256_mult_precompute(EC_GROUP
*group
, BN_CTX
*ctx
)
764 * We precompute a table for a Booth encoded exponent (wNAF) based
765 * computation. Each table holds 64 values for safe access, with an
766 * implicit value of infinity at index zero. We use window of size 7, and
767 * therefore require ceil(256/7) = 37 tables.
770 EC_POINT
*P
= NULL
, *T
= NULL
;
771 const EC_POINT
*generator
;
772 NISTZ256_PRE_COMP
*pre_comp
;
773 BN_CTX
*new_ctx
= NULL
;
774 int i
, j
, k
, ret
= 0;
777 PRECOMP256_ROW
*preComputedTable
= NULL
;
778 unsigned char *precomp_storage
= NULL
;
780 /* if there is an old NISTZ256_PRE_COMP object, throw it away */
781 EC_pre_comp_free(group
);
782 generator
= EC_GROUP_get0_generator(group
);
783 if (generator
== NULL
) {
784 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE
, EC_R_UNDEFINED_GENERATOR
);
788 if (ecp_nistz256_is_affine_G(generator
)) {
790 * No need to calculate tables for the standard generator because we
791 * have them statically.
796 if ((pre_comp
= ecp_nistz256_pre_comp_new(group
)) == NULL
)
800 ctx
= new_ctx
= BN_CTX_new();
807 order
= EC_GROUP_get0_order(group
);
811 if (BN_is_zero(order
)) {
812 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE
, EC_R_UNKNOWN_ORDER
);
818 if ((precomp_storage
=
819 OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE
) + 64)) == NULL
) {
820 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE
, ERR_R_MALLOC_FAILURE
);
824 preComputedTable
= (void *)ALIGNPTR(precomp_storage
, 64);
826 P
= EC_POINT_new(group
);
827 T
= EC_POINT_new(group
);
828 if (P
== NULL
|| T
== NULL
)
832 * The zero entry is implicitly infinity, and we skip it, storing other
833 * values with -1 offset.
835 if (!EC_POINT_copy(T
, generator
))
838 for (k
= 0; k
< 64; k
++) {
839 if (!EC_POINT_copy(P
, T
))
841 for (j
= 0; j
< 37; j
++) {
842 P256_POINT_AFFINE temp
;
844 * It would be faster to use EC_POINTs_make_affine and
845 * make multiple points affine at the same time.
847 if (!EC_POINT_make_affine(group
, P
, ctx
))
849 if (!ecp_nistz256_bignum_to_field_elem(temp
.X
, P
->X
) ||
850 !ecp_nistz256_bignum_to_field_elem(temp
.Y
, P
->Y
)) {
851 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE
,
852 EC_R_COORDINATES_OUT_OF_RANGE
);
855 ecp_nistz256_scatter_w7(preComputedTable
[j
], &temp
, k
);
856 for (i
= 0; i
< 7; i
++) {
857 if (!EC_POINT_dbl(group
, P
, P
, ctx
))
861 if (!EC_POINT_add(group
, T
, T
, generator
, ctx
))
865 pre_comp
->group
= group
;
867 pre_comp
->precomp
= preComputedTable
;
868 pre_comp
->precomp_storage
= precomp_storage
;
869 precomp_storage
= NULL
;
870 SETPRECOMP(group
, nistz256
, pre_comp
);
877 BN_CTX_free(new_ctx
);
879 EC_nistz256_pre_comp_free(pre_comp
);
880 OPENSSL_free(precomp_storage
);
887 * Note that by default ECP_NISTZ256_AVX2 is undefined. While it's great
888 * code processing 4 points in parallel, corresponding serial operation
889 * is several times slower, because it uses 29x29=58-bit multiplication
890 * as opposite to 64x64=128-bit in integer-only scalar case. As result
891 * it doesn't provide *significant* performance improvement. Note that
892 * just defining ECP_NISTZ256_AVX2 is not sufficient to make it work,
893 * you'd need to compile even asm/ecp_nistz256-avx.pl module.
895 #if defined(ECP_NISTZ256_AVX2)
896 # if !(defined(__x86_64) || defined(__x86_64__) || \
897 defined(_M_AMD64) || defined(_MX64)) || \
898 !(defined(__GNUC__) || defined(_MSC_VER)) /* this is for ALIGN32 */
899 # undef ECP_NISTZ256_AVX2
901 /* Constant time access, loading four values, from four consecutive tables */
902 void ecp_nistz256_avx2_multi_gather_w7(void *result
, const void *in
,
903 int index0
, int index1
, int index2
,
905 void ecp_nistz256_avx2_transpose_convert(void *RESULTx4
, const void *in
);
906 void ecp_nistz256_avx2_convert_transpose_back(void *result
, const void *Ax4
);
907 void ecp_nistz256_avx2_point_add_affine_x4(void *RESULTx4
, const void *Ax4
,
909 void ecp_nistz256_avx2_point_add_affines_x4(void *RESULTx4
, const void *Ax4
,
911 void ecp_nistz256_avx2_to_mont(void *RESULTx4
, const void *Ax4
);
912 void ecp_nistz256_avx2_from_mont(void *RESULTx4
, const void *Ax4
);
913 void ecp_nistz256_avx2_set1(void *RESULTx4
);
914 int ecp_nistz_avx2_eligible(void);
916 static void booth_recode_w7(unsigned char *sign
,
917 unsigned char *digit
, unsigned char in
)
921 s
= ~((in
>> 7) - 1);
922 d
= (1 << 8) - in
- 1;
923 d
= (d
& s
) | (in
& ~s
);
924 d
= (d
>> 1) + (d
& 1);
931 * ecp_nistz256_avx2_mul_g performs multiplication by G, using only the
932 * precomputed table. It does 4 affine point additions in parallel,
933 * significantly speeding up point multiplication for a fixed value.
935 static void ecp_nistz256_avx2_mul_g(P256_POINT
*r
,
936 unsigned char p_str
[33],
937 const P256_POINT_AFFINE(*preComputedTable
)[64])
939 const unsigned int window_size
= 7;
940 const unsigned int mask
= (1 << (window_size
+ 1)) - 1;
942 /* Using 4 windows at a time */
943 unsigned char sign0
, digit0
;
944 unsigned char sign1
, digit1
;
945 unsigned char sign2
, digit2
;
946 unsigned char sign3
, digit3
;
947 unsigned int idx
= 0;
948 BN_ULONG tmp
[P256_LIMBS
];
951 ALIGN32 BN_ULONG aX4
[4 * 9 * 3] = { 0 };
952 ALIGN32 BN_ULONG bX4
[4 * 9 * 2] = { 0 };
953 ALIGN32 P256_POINT_AFFINE point_arr
[4];
954 ALIGN32 P256_POINT res_point_arr
[4];
956 /* Initial four windows */
957 wvalue
= *((u16
*) & p_str
[0]);
958 wvalue
= (wvalue
<< 1) & mask
;
960 booth_recode_w7(&sign0
, &digit0
, wvalue
);
961 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
962 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
964 booth_recode_w7(&sign1
, &digit1
, wvalue
);
965 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
966 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
968 booth_recode_w7(&sign2
, &digit2
, wvalue
);
969 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
970 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
972 booth_recode_w7(&sign3
, &digit3
, wvalue
);
974 ecp_nistz256_avx2_multi_gather_w7(point_arr
, preComputedTable
[0],
975 digit0
, digit1
, digit2
, digit3
);
977 ecp_nistz256_neg(tmp
, point_arr
[0].Y
);
978 copy_conditional(point_arr
[0].Y
, tmp
, sign0
);
979 ecp_nistz256_neg(tmp
, point_arr
[1].Y
);
980 copy_conditional(point_arr
[1].Y
, tmp
, sign1
);
981 ecp_nistz256_neg(tmp
, point_arr
[2].Y
);
982 copy_conditional(point_arr
[2].Y
, tmp
, sign2
);
983 ecp_nistz256_neg(tmp
, point_arr
[3].Y
);
984 copy_conditional(point_arr
[3].Y
, tmp
, sign3
);
986 ecp_nistz256_avx2_transpose_convert(aX4
, point_arr
);
987 ecp_nistz256_avx2_to_mont(aX4
, aX4
);
988 ecp_nistz256_avx2_to_mont(&aX4
[4 * 9], &aX4
[4 * 9]);
989 ecp_nistz256_avx2_set1(&aX4
[4 * 9 * 2]);
991 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
992 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
994 booth_recode_w7(&sign0
, &digit0
, wvalue
);
995 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
996 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
998 booth_recode_w7(&sign1
, &digit1
, wvalue
);
999 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1000 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1002 booth_recode_w7(&sign2
, &digit2
, wvalue
);
1003 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1004 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1006 booth_recode_w7(&sign3
, &digit3
, wvalue
);
1008 ecp_nistz256_avx2_multi_gather_w7(point_arr
, preComputedTable
[4 * 1],
1009 digit0
, digit1
, digit2
, digit3
);
1011 ecp_nistz256_neg(tmp
, point_arr
[0].Y
);
1012 copy_conditional(point_arr
[0].Y
, tmp
, sign0
);
1013 ecp_nistz256_neg(tmp
, point_arr
[1].Y
);
1014 copy_conditional(point_arr
[1].Y
, tmp
, sign1
);
1015 ecp_nistz256_neg(tmp
, point_arr
[2].Y
);
1016 copy_conditional(point_arr
[2].Y
, tmp
, sign2
);
1017 ecp_nistz256_neg(tmp
, point_arr
[3].Y
);
1018 copy_conditional(point_arr
[3].Y
, tmp
, sign3
);
1020 ecp_nistz256_avx2_transpose_convert(bX4
, point_arr
);
1021 ecp_nistz256_avx2_to_mont(bX4
, bX4
);
1022 ecp_nistz256_avx2_to_mont(&bX4
[4 * 9], &bX4
[4 * 9]);
1023 /* Optimized when both inputs are affine */
1024 ecp_nistz256_avx2_point_add_affines_x4(aX4
, aX4
, bX4
);
1026 for (i
= 2; i
< 9; i
++) {
1027 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1028 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1030 booth_recode_w7(&sign0
, &digit0
, wvalue
);
1031 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1032 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1034 booth_recode_w7(&sign1
, &digit1
, wvalue
);
1035 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1036 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1038 booth_recode_w7(&sign2
, &digit2
, wvalue
);
1039 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1040 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1042 booth_recode_w7(&sign3
, &digit3
, wvalue
);
1044 ecp_nistz256_avx2_multi_gather_w7(point_arr
,
1045 preComputedTable
[4 * i
],
1046 digit0
, digit1
, digit2
, digit3
);
1048 ecp_nistz256_neg(tmp
, point_arr
[0].Y
);
1049 copy_conditional(point_arr
[0].Y
, tmp
, sign0
);
1050 ecp_nistz256_neg(tmp
, point_arr
[1].Y
);
1051 copy_conditional(point_arr
[1].Y
, tmp
, sign1
);
1052 ecp_nistz256_neg(tmp
, point_arr
[2].Y
);
1053 copy_conditional(point_arr
[2].Y
, tmp
, sign2
);
1054 ecp_nistz256_neg(tmp
, point_arr
[3].Y
);
1055 copy_conditional(point_arr
[3].Y
, tmp
, sign3
);
1057 ecp_nistz256_avx2_transpose_convert(bX4
, point_arr
);
1058 ecp_nistz256_avx2_to_mont(bX4
, bX4
);
1059 ecp_nistz256_avx2_to_mont(&bX4
[4 * 9], &bX4
[4 * 9]);
1061 ecp_nistz256_avx2_point_add_affine_x4(aX4
, aX4
, bX4
);
1064 ecp_nistz256_avx2_from_mont(&aX4
[4 * 9 * 0], &aX4
[4 * 9 * 0]);
1065 ecp_nistz256_avx2_from_mont(&aX4
[4 * 9 * 1], &aX4
[4 * 9 * 1]);
1066 ecp_nistz256_avx2_from_mont(&aX4
[4 * 9 * 2], &aX4
[4 * 9 * 2]);
1068 ecp_nistz256_avx2_convert_transpose_back(res_point_arr
, aX4
);
1069 /* Last window is performed serially */
1070 wvalue
= *((u16
*) & p_str
[(idx
- 1) / 8]);
1071 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1072 booth_recode_w7(&sign0
, &digit0
, wvalue
);
1073 ecp_nistz256_gather_w7((P256_POINT_AFFINE
*)r
,
1074 preComputedTable
[36], digit0
);
1075 ecp_nistz256_neg(tmp
, r
->Y
);
1076 copy_conditional(r
->Y
, tmp
, sign0
);
1077 memcpy(r
->Z
, ONE
, sizeof(ONE
));
1078 /* Sum the four windows */
1079 ecp_nistz256_point_add(r
, r
, &res_point_arr
[0]);
1080 ecp_nistz256_point_add(r
, r
, &res_point_arr
[1]);
1081 ecp_nistz256_point_add(r
, r
, &res_point_arr
[2]);
1082 ecp_nistz256_point_add(r
, r
, &res_point_arr
[3]);
1087 __owur
static int ecp_nistz256_set_from_affine(EC_POINT
*out
, const EC_GROUP
*group
,
1088 const P256_POINT_AFFINE
*in
,
1092 BN_ULONG d_x
[P256_LIMBS
], d_y
[P256_LIMBS
];
1103 memcpy(d_x
, in
->X
, sizeof(d_x
));
1104 bn_set_static_words(x
, d_x
, P256_LIMBS
);
1106 memcpy(d_y
, in
->Y
, sizeof(d_y
));
1107 bn_set_static_words(y
, d_y
, P256_LIMBS
);
1109 ret
= EC_POINT_set_affine_coordinates_GFp(group
, out
, x
, y
, ctx
);
1117 /* r = scalar*G + sum(scalars[i]*points[i]) */
1118 __owur
static int ecp_nistz256_points_mul(const EC_GROUP
*group
,
1120 const BIGNUM
*scalar
,
1122 const EC_POINT
*points
[],
1123 const BIGNUM
*scalars
[], BN_CTX
*ctx
)
1125 int i
= 0, ret
= 0, no_precomp_for_generator
= 0, p_is_infinity
= 0;
1127 unsigned char p_str
[33] = { 0 };
1128 const PRECOMP256_ROW
*preComputedTable
= NULL
;
1129 const NISTZ256_PRE_COMP
*pre_comp
= NULL
;
1130 const EC_POINT
*generator
= NULL
;
1131 BN_CTX
*new_ctx
= NULL
;
1132 const BIGNUM
**new_scalars
= NULL
;
1133 const EC_POINT
**new_points
= NULL
;
1134 unsigned int idx
= 0;
1135 const unsigned int window_size
= 7;
1136 const unsigned int mask
= (1 << (window_size
+ 1)) - 1;
1137 unsigned int wvalue
;
1140 P256_POINT_AFFINE a
;
1144 if ((num
+ 1) == 0 || (num
+ 1) > OPENSSL_MALLOC_MAX_NELEMS(void *)) {
1145 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, ERR_R_MALLOC_FAILURE
);
1149 if (group
->meth
!= r
->meth
) {
1150 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, EC_R_INCOMPATIBLE_OBJECTS
);
1154 if ((scalar
== NULL
) && (num
== 0))
1155 return EC_POINT_set_to_infinity(group
, r
);
1157 for (j
= 0; j
< num
; j
++) {
1158 if (group
->meth
!= points
[j
]->meth
) {
1159 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, EC_R_INCOMPATIBLE_OBJECTS
);
1165 ctx
= new_ctx
= BN_CTX_new();
1173 generator
= EC_GROUP_get0_generator(group
);
1174 if (generator
== NULL
) {
1175 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, EC_R_UNDEFINED_GENERATOR
);
1179 /* look if we can use precomputed multiples of generator */
1180 pre_comp
= group
->pre_comp
.nistz256
;
1184 * If there is a precomputed table for the generator, check that
1185 * it was generated with the same generator.
1187 EC_POINT
*pre_comp_generator
= EC_POINT_new(group
);
1188 if (pre_comp_generator
== NULL
)
1191 if (!ecp_nistz256_set_from_affine(pre_comp_generator
,
1192 group
, pre_comp
->precomp
[0],
1194 EC_POINT_free(pre_comp_generator
);
1198 if (0 == EC_POINT_cmp(group
, generator
, pre_comp_generator
, ctx
))
1199 preComputedTable
= (const PRECOMP256_ROW
*)pre_comp
->precomp
;
1201 EC_POINT_free(pre_comp_generator
);
1204 if (preComputedTable
== NULL
&& ecp_nistz256_is_affine_G(generator
)) {
1206 * If there is no precomputed data, but the generator is the
1207 * default, a hardcoded table of precomputed data is used. This
1208 * is because applications, such as Apache, do not use
1209 * EC_KEY_precompute_mult.
1211 preComputedTable
= ecp_nistz256_precomputed
;
1214 if (preComputedTable
) {
1215 if ((BN_num_bits(scalar
) > 256)
1216 || BN_is_negative(scalar
)) {
1217 if ((tmp_scalar
= BN_CTX_get(ctx
)) == NULL
)
1220 if (!BN_nnmod(tmp_scalar
, scalar
, group
->order
, ctx
)) {
1221 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, ERR_R_BN_LIB
);
1224 scalar
= tmp_scalar
;
1227 for (i
= 0; i
< bn_get_top(scalar
) * BN_BYTES
; i
+= BN_BYTES
) {
1228 BN_ULONG d
= bn_get_words(scalar
)[i
/ BN_BYTES
];
1230 p_str
[i
+ 0] = (unsigned char)d
;
1231 p_str
[i
+ 1] = (unsigned char)(d
>> 8);
1232 p_str
[i
+ 2] = (unsigned char)(d
>> 16);
1233 p_str
[i
+ 3] = (unsigned char)(d
>>= 24);
1234 if (BN_BYTES
== 8) {
1236 p_str
[i
+ 4] = (unsigned char)d
;
1237 p_str
[i
+ 5] = (unsigned char)(d
>> 8);
1238 p_str
[i
+ 6] = (unsigned char)(d
>> 16);
1239 p_str
[i
+ 7] = (unsigned char)(d
>> 24);
1246 #if defined(ECP_NISTZ256_AVX2)
1247 if (ecp_nistz_avx2_eligible()) {
1248 ecp_nistz256_avx2_mul_g(&p
.p
, p_str
, preComputedTable
);
1253 wvalue
= (p_str
[0] << 1) & mask
;
1256 wvalue
= _booth_recode_w7(wvalue
);
1258 ecp_nistz256_gather_w7(&p
.a
, preComputedTable
[0],
1261 ecp_nistz256_neg(p
.p
.Z
, p
.p
.Y
);
1262 copy_conditional(p
.p
.Y
, p
.p
.Z
, wvalue
& 1);
1264 memcpy(p
.p
.Z
, ONE
, sizeof(ONE
));
1266 for (i
= 1; i
< 37; i
++) {
1267 unsigned int off
= (idx
- 1) / 8;
1268 wvalue
= p_str
[off
] | p_str
[off
+ 1] << 8;
1269 wvalue
= (wvalue
>> ((idx
- 1) % 8)) & mask
;
1272 wvalue
= _booth_recode_w7(wvalue
);
1274 ecp_nistz256_gather_w7(&t
.a
,
1275 preComputedTable
[i
], wvalue
>> 1);
1277 ecp_nistz256_neg(t
.p
.Z
, t
.a
.Y
);
1278 copy_conditional(t
.a
.Y
, t
.p
.Z
, wvalue
& 1);
1280 ecp_nistz256_point_add_affine(&p
.p
, &p
.p
, &t
.a
);
1285 no_precomp_for_generator
= 1;
1290 if (no_precomp_for_generator
) {
1292 * Without a precomputed table for the generator, it has to be
1293 * handled like a normal point.
1295 new_scalars
= OPENSSL_malloc((num
+ 1) * sizeof(BIGNUM
*));
1296 if (new_scalars
== NULL
) {
1297 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, ERR_R_MALLOC_FAILURE
);
1301 new_points
= OPENSSL_malloc((num
+ 1) * sizeof(EC_POINT
*));
1302 if (new_points
== NULL
) {
1303 ECerr(EC_F_ECP_NISTZ256_POINTS_MUL
, ERR_R_MALLOC_FAILURE
);
1307 memcpy(new_scalars
, scalars
, num
* sizeof(BIGNUM
*));
1308 new_scalars
[num
] = scalar
;
1309 memcpy(new_points
, points
, num
* sizeof(EC_POINT
*));
1310 new_points
[num
] = generator
;
1312 scalars
= new_scalars
;
1313 points
= new_points
;
1318 P256_POINT
*out
= &t
.p
;
1322 if (!ecp_nistz256_windowed_mul(group
, out
, scalars
, points
, num
, ctx
))
1326 ecp_nistz256_point_add(&p
.p
, &p
.p
, out
);
1329 /* Not constant-time, but we're only operating on the public output. */
1330 if (!bn_set_words(r
->X
, p
.p
.X
, P256_LIMBS
) ||
1331 !bn_set_words(r
->Y
, p
.p
.Y
, P256_LIMBS
) ||
1332 !bn_set_words(r
->Z
, p
.p
.Z
, P256_LIMBS
)) {
1335 r
->Z_is_one
= is_one(p
.p
.Z
) & 1;
1342 BN_CTX_free(new_ctx
);
1343 OPENSSL_free(new_points
);
1344 OPENSSL_free(new_scalars
);
1348 __owur
static int ecp_nistz256_get_affine(const EC_GROUP
*group
,
1349 const EC_POINT
*point
,
1350 BIGNUM
*x
, BIGNUM
*y
, BN_CTX
*ctx
)
1352 BN_ULONG z_inv2
[P256_LIMBS
];
1353 BN_ULONG z_inv3
[P256_LIMBS
];
1354 BN_ULONG x_aff
[P256_LIMBS
];
1355 BN_ULONG y_aff
[P256_LIMBS
];
1356 BN_ULONG point_x
[P256_LIMBS
], point_y
[P256_LIMBS
], point_z
[P256_LIMBS
];
1357 BN_ULONG x_ret
[P256_LIMBS
], y_ret
[P256_LIMBS
];
1359 if (EC_POINT_is_at_infinity(group
, point
)) {
1360 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE
, EC_R_POINT_AT_INFINITY
);
1364 if (!ecp_nistz256_bignum_to_field_elem(point_x
, point
->X
) ||
1365 !ecp_nistz256_bignum_to_field_elem(point_y
, point
->Y
) ||
1366 !ecp_nistz256_bignum_to_field_elem(point_z
, point
->Z
)) {
1367 ECerr(EC_F_ECP_NISTZ256_GET_AFFINE
, EC_R_COORDINATES_OUT_OF_RANGE
);
1371 ecp_nistz256_mod_inverse(z_inv3
, point_z
);
1372 ecp_nistz256_sqr_mont(z_inv2
, z_inv3
);
1373 ecp_nistz256_mul_mont(x_aff
, z_inv2
, point_x
);
1376 ecp_nistz256_from_mont(x_ret
, x_aff
);
1377 if (!bn_set_words(x
, x_ret
, P256_LIMBS
))
1382 ecp_nistz256_mul_mont(z_inv3
, z_inv3
, z_inv2
);
1383 ecp_nistz256_mul_mont(y_aff
, z_inv3
, point_y
);
1384 ecp_nistz256_from_mont(y_ret
, y_aff
);
1385 if (!bn_set_words(y
, y_ret
, P256_LIMBS
))
1392 static NISTZ256_PRE_COMP
*ecp_nistz256_pre_comp_new(const EC_GROUP
*group
)
1394 NISTZ256_PRE_COMP
*ret
= NULL
;
1399 ret
= OPENSSL_zalloc(sizeof(*ret
));
1402 ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW
, ERR_R_MALLOC_FAILURE
);
1407 ret
->w
= 6; /* default */
1408 ret
->references
= 1;
1410 ret
->lock
= CRYPTO_THREAD_lock_new();
1411 if (ret
->lock
== NULL
) {
1412 ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW
, ERR_R_MALLOC_FAILURE
);
1419 NISTZ256_PRE_COMP
*EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP
*p
)
1423 CRYPTO_atomic_add(&p
->references
, 1, &i
, p
->lock
);
1427 void EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP
*pre
)
1434 CRYPTO_atomic_add(&pre
->references
, -1, &i
, pre
->lock
);
1435 REF_PRINT_COUNT("EC_nistz256", x
);
1438 REF_ASSERT_ISNT(i
< 0);
1440 OPENSSL_free(pre
->precomp_storage
);
1441 CRYPTO_THREAD_lock_free(pre
->lock
);
1446 static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP
*group
)
1448 /* There is a hard-coded table for the default generator. */
1449 const EC_POINT
*generator
= EC_GROUP_get0_generator(group
);
1451 if (generator
!= NULL
&& ecp_nistz256_is_affine_G(generator
)) {
1452 /* There is a hard-coded table for the default generator. */
1456 return HAVEPRECOMP(group
, nistz256
);
1459 const EC_METHOD
*EC_GFp_nistz256_method(void)
1461 static const EC_METHOD ret
= {
1462 EC_FLAGS_DEFAULT_OCT
,
1463 NID_X9_62_prime_field
,
1464 ec_GFp_mont_group_init
,
1465 ec_GFp_mont_group_finish
,
1466 ec_GFp_mont_group_clear_finish
,
1467 ec_GFp_mont_group_copy
,
1468 ec_GFp_mont_group_set_curve
,
1469 ec_GFp_simple_group_get_curve
,
1470 ec_GFp_simple_group_get_degree
,
1471 ec_group_simple_order_bits
,
1472 ec_GFp_simple_group_check_discriminant
,
1473 ec_GFp_simple_point_init
,
1474 ec_GFp_simple_point_finish
,
1475 ec_GFp_simple_point_clear_finish
,
1476 ec_GFp_simple_point_copy
,
1477 ec_GFp_simple_point_set_to_infinity
,
1478 ec_GFp_simple_set_Jprojective_coordinates_GFp
,
1479 ec_GFp_simple_get_Jprojective_coordinates_GFp
,
1480 ec_GFp_simple_point_set_affine_coordinates
,
1481 ecp_nistz256_get_affine
,
1485 ec_GFp_simple_invert
,
1486 ec_GFp_simple_is_at_infinity
,
1487 ec_GFp_simple_is_on_curve
,
1489 ec_GFp_simple_make_affine
,
1490 ec_GFp_simple_points_make_affine
,
1491 ecp_nistz256_points_mul
, /* mul */
1492 ecp_nistz256_mult_precompute
, /* precompute_mult */
1493 ecp_nistz256_window_have_precompute_mult
, /* have_precompute_mult */
1494 ec_GFp_mont_field_mul
,
1495 ec_GFp_mont_field_sqr
,
1497 ec_GFp_mont_field_encode
,
1498 ec_GFp_mont_field_decode
,
1499 ec_GFp_mont_field_set_to_one
,
1500 ec_key_simple_priv2oct
,
1501 ec_key_simple_oct2priv
,
1502 0, /* set private */
1503 ec_key_simple_generate_key
,
1504 ec_key_simple_check_key
,
1505 ec_key_simple_generate_public_key
,
1508 ecdh_simple_compute_key