• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1/* Copyright 2015 The BoringSSL Authors
2 *
3 * Permission to use, copy, modify, and/or distribute this software for any
4 * purpose with or without fee is hereby granted, provided that the above
5 * copyright notice and this permission notice appear in all copies.
6 *
7 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
8 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
9 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
10 * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
11 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
12 * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
13 * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
14
15// A 64-bit implementation of the NIST P-224 elliptic curve point multiplication
16//
17// Inspired by Daniel J. Bernstein's public domain nistp224 implementation
18// and Adam Langley's public domain 64-bit C implementation of curve25519.
19
20#include <openssl/base.h>
21
22#include <openssl/bn.h>
23#include <openssl/ec.h>
24#include <openssl/err.h>
25#include <openssl/mem.h>
26
27#include <assert.h>
28#include <string.h>
29
30#include "internal.h"
31#include "../delocate.h"
32#include "../../internal.h"
33
34
35#if defined(BORINGSSL_HAS_UINT128) && !defined(OPENSSL_SMALL)
36
37// Field elements are represented as a_0 + 2^56*a_1 + 2^112*a_2 + 2^168*a_3
38// using 64-bit coefficients called 'limbs', and sometimes (for multiplication
39// results) as b_0 + 2^56*b_1 + 2^112*b_2 + 2^168*b_3 + 2^224*b_4 + 2^280*b_5 +
40// 2^336*b_6 using 128-bit coefficients called 'widelimbs'. A 4-p224_limb
41// representation is an 'p224_felem'; a 7-p224_widelimb representation is a
42// 'p224_widefelem'. Even within felems, bits of adjacent limbs overlap, and we
43// don't always reduce the representations: we ensure that inputs to each
44// p224_felem multiplication satisfy a_i < 2^60, so outputs satisfy b_i <
45// 4*2^60*2^60, and fit into a 128-bit word without overflow. The coefficients
46// are then again partially reduced to obtain an p224_felem satisfying a_i <
47// 2^57. We only reduce to the unique minimal representation at the end of the
48// computation.
49
50typedef uint64_t p224_limb;
51typedef uint128_t p224_widelimb;
52
53typedef p224_limb p224_felem[4];
54typedef p224_widelimb p224_widefelem[7];
55
56// Precomputed multiples of the standard generator
57// Points are given in coordinates (X, Y, Z) where Z normally is 1
58// (0 for the point at infinity).
59// For each field element, slice a_0 is word 0, etc.
60//
61// The table has 2 * 16 elements, starting with the following:
62// index | bits    | point
63// ------+---------+------------------------------
64//     0 | 0 0 0 0 | 0G
65//     1 | 0 0 0 1 | 1G
66//     2 | 0 0 1 0 | 2^56G
67//     3 | 0 0 1 1 | (2^56 + 1)G
68//     4 | 0 1 0 0 | 2^112G
69//     5 | 0 1 0 1 | (2^112 + 1)G
70//     6 | 0 1 1 0 | (2^112 + 2^56)G
71//     7 | 0 1 1 1 | (2^112 + 2^56 + 1)G
72//     8 | 1 0 0 0 | 2^168G
73//     9 | 1 0 0 1 | (2^168 + 1)G
74//    10 | 1 0 1 0 | (2^168 + 2^56)G
75//    11 | 1 0 1 1 | (2^168 + 2^56 + 1)G
76//    12 | 1 1 0 0 | (2^168 + 2^112)G
77//    13 | 1 1 0 1 | (2^168 + 2^112 + 1)G
78//    14 | 1 1 1 0 | (2^168 + 2^112 + 2^56)G
79//    15 | 1 1 1 1 | (2^168 + 2^112 + 2^56 + 1)G
80// followed by a copy of this with each element multiplied by 2^28.
81//
82// The reason for this is so that we can clock bits into four different
83// locations when doing simple scalar multiplies against the base point,
84// and then another four locations using the second 16 elements.
85static const p224_felem g_p224_pre_comp[2][16][3] = {
86    {{{0, 0, 0, 0}, {0, 0, 0, 0}, {0, 0, 0, 0}},
87     {{0x3280d6115c1d21, 0xc1d356c2112234, 0x7f321390b94a03, 0xb70e0cbd6bb4bf},
88      {0xd5819985007e34, 0x75a05a07476444, 0xfb4c22dfe6cd43, 0xbd376388b5f723},
89      {1, 0, 0, 0}},
90     {{0xfd9675666ebbe9, 0xbca7664d40ce5e, 0x2242df8d8a2a43, 0x1f49bbb0f99bc5},
91      {0x29e0b892dc9c43, 0xece8608436e662, 0xdc858f185310d0, 0x9812dd4eb8d321},
92      {1, 0, 0, 0}},
93     {{0x6d3e678d5d8eb8, 0x559eed1cb362f1, 0x16e9a3bbce8a3f, 0xeedcccd8c2a748},
94      {0xf19f90ed50266d, 0xabf2b4bf65f9df, 0x313865468fafec, 0x5cb379ba910a17},
95      {1, 0, 0, 0}},
96     {{0x0641966cab26e3, 0x91fb2991fab0a0, 0xefec27a4e13a0b, 0x0499aa8a5f8ebe},
97      {0x7510407766af5d, 0x84d929610d5450, 0x81d77aae82f706, 0x6916f6d4338c5b},
98      {1, 0, 0, 0}},
99     {{0xea95ac3b1f15c6, 0x086000905e82d4, 0xdd323ae4d1c8b1, 0x932b56be7685a3},
100      {0x9ef93dea25dbbf, 0x41665960f390f0, 0xfdec76dbe2a8a7, 0x523e80f019062a},
101      {1, 0, 0, 0}},
102     {{0x822fdd26732c73, 0xa01c83531b5d0f, 0x363f37347c1ba4, 0xc391b45c84725c},
103      {0xbbd5e1b2d6ad24, 0xddfbcde19dfaec, 0xc393da7e222a7f, 0x1efb7890ede244},
104      {1, 0, 0, 0}},
105     {{0x4c9e90ca217da1, 0xd11beca79159bb, 0xff8d33c2c98b7c, 0x2610b39409f849},
106      {0x44d1352ac64da0, 0xcdbb7b2c46b4fb, 0x966c079b753c89, 0xfe67e4e820b112},
107      {1, 0, 0, 0}},
108     {{0xe28cae2df5312d, 0xc71b61d16f5c6e, 0x79b7619a3e7c4c, 0x05c73240899b47},
109      {0x9f7f6382c73e3a, 0x18615165c56bda, 0x641fab2116fd56, 0x72855882b08394},
110      {1, 0, 0, 0}},
111     {{0x0469182f161c09, 0x74a98ca8d00fb5, 0xb89da93489a3e0, 0x41c98768fb0c1d},
112      {0xe5ea05fb32da81, 0x3dce9ffbca6855, 0x1cfe2d3fbf59e6, 0x0e5e03408738a7},
113      {1, 0, 0, 0}},
114     {{0xdab22b2333e87f, 0x4430137a5dd2f6, 0xe03ab9f738beb8, 0xcb0c5d0dc34f24},
115      {0x764a7df0c8fda5, 0x185ba5c3fa2044, 0x9281d688bcbe50, 0xc40331df893881},
116      {1, 0, 0, 0}},
117     {{0xb89530796f0f60, 0xade92bd26909a3, 0x1a0c83fb4884da, 0x1765bf22a5a984},
118      {0x772a9ee75db09e, 0x23bc6c67cec16f, 0x4c1edba8b14e2f, 0xe2a215d9611369},
119      {1, 0, 0, 0}},
120     {{0x571e509fb5efb3, 0xade88696410552, 0xc8ae85fada74fe, 0x6c7e4be83bbde3},
121      {0xff9f51160f4652, 0xb47ce2495a6539, 0xa2946c53b582f4, 0x286d2db3ee9a60},
122      {1, 0, 0, 0}},
123     {{0x40bbd5081a44af, 0x0995183b13926c, 0xbcefba6f47f6d0, 0x215619e9cc0057},
124      {0x8bc94d3b0df45e, 0xf11c54a3694f6f, 0x8631b93cdfe8b5, 0xe7e3f4b0982db9},
125      {1, 0, 0, 0}},
126     {{0xb17048ab3e1c7b, 0xac38f36ff8a1d8, 0x1c29819435d2c6, 0xc813132f4c07e9},
127      {0x2891425503b11f, 0x08781030579fea, 0xf5426ba5cc9674, 0x1e28ebf18562bc},
128      {1, 0, 0, 0}},
129     {{0x9f31997cc864eb, 0x06cd91d28b5e4c, 0xff17036691a973, 0xf1aef351497c58},
130      {0xdd1f2d600564ff, 0xdead073b1402db, 0x74a684435bd693, 0xeea7471f962558},
131      {1, 0, 0, 0}}},
132    {{{0, 0, 0, 0}, {0, 0, 0, 0}, {0, 0, 0, 0}},
133     {{0x9665266dddf554, 0x9613d78b60ef2d, 0xce27a34cdba417, 0xd35ab74d6afc31},
134      {0x85ccdd22deb15e, 0x2137e5783a6aab, 0xa141cffd8c93c6, 0x355a1830e90f2d},
135      {1, 0, 0, 0}},
136     {{0x1a494eadaade65, 0xd6da4da77fe53c, 0xe7992996abec86, 0x65c3553c6090e3},
137      {0xfa610b1fb09346, 0xf1c6540b8a4aaf, 0xc51a13ccd3cbab, 0x02995b1b18c28a},
138      {1, 0, 0, 0}},
139     {{0x7874568e7295ef, 0x86b419fbe38d04, 0xdc0690a7550d9a, 0xd3966a44beac33},
140      {0x2b7280ec29132f, 0xbeaa3b6a032df3, 0xdc7dd88ae41200, 0xd25e2513e3a100},
141      {1, 0, 0, 0}},
142     {{0x924857eb2efafd, 0xac2bce41223190, 0x8edaa1445553fc, 0x825800fd3562d5},
143      {0x8d79148ea96621, 0x23a01c3dd9ed8d, 0xaf8b219f9416b5, 0xd8db0cc277daea},
144      {1, 0, 0, 0}},
145     {{0x76a9c3b1a700f0, 0xe9acd29bc7e691, 0x69212d1a6b0327, 0x6322e97fe154be},
146      {0x469fc5465d62aa, 0x8d41ed18883b05, 0x1f8eae66c52b88, 0xe4fcbe9325be51},
147      {1, 0, 0, 0}},
148     {{0x825fdf583cac16, 0x020b857c7b023a, 0x683c17744b0165, 0x14ffd0a2daf2f1},
149      {0x323b36184218f9, 0x4944ec4e3b47d4, 0xc15b3080841acf, 0x0bced4b01a28bb},
150      {1, 0, 0, 0}},
151     {{0x92ac22230df5c4, 0x52f33b4063eda8, 0xcb3f19870c0c93, 0x40064f2ba65233},
152      {0xfe16f0924f8992, 0x012da25af5b517, 0x1a57bb24f723a6, 0x06f8bc76760def},
153      {1, 0, 0, 0}},
154     {{0x4a7084f7817cb9, 0xbcab0738ee9a78, 0x3ec11e11d9c326, 0xdc0fe90e0f1aae},
155      {0xcf639ea5f98390, 0x5c350aa22ffb74, 0x9afae98a4047b7, 0x956ec2d617fc45},
156      {1, 0, 0, 0}},
157     {{0x4306d648c1be6a, 0x9247cd8bc9a462, 0xf5595e377d2f2e, 0xbd1c3caff1a52e},
158      {0x045e14472409d0, 0x29f3e17078f773, 0x745a602b2d4f7d, 0x191837685cdfbb},
159      {1, 0, 0, 0}},
160     {{0x5b6ee254a8cb79, 0x4953433f5e7026, 0xe21faeb1d1def4, 0xc4c225785c09de},
161      {0x307ce7bba1e518, 0x31b125b1036db8, 0x47e91868839e8f, 0xc765866e33b9f3},
162      {1, 0, 0, 0}},
163     {{0x3bfece24f96906, 0x4794da641e5093, 0xde5df64f95db26, 0x297ecd89714b05},
164      {0x701bd3ebb2c3aa, 0x7073b4f53cb1d5, 0x13c5665658af16, 0x9895089d66fe58},
165      {1, 0, 0, 0}},
166     {{0x0fef05f78c4790, 0x2d773633b05d2e, 0x94229c3a951c94, 0xbbbd70df4911bb},
167      {0xb2c6963d2c1168, 0x105f47a72b0d73, 0x9fdf6111614080, 0x7b7e94b39e67b0},
168      {1, 0, 0, 0}},
169     {{0xad1a7d6efbe2b3, 0xf012482c0da69d, 0x6b3bdf12438345, 0x40d7558d7aa4d9},
170      {0x8a09fffb5c6d3d, 0x9a356e5d9ffd38, 0x5973f15f4f9b1c, 0xdcd5f59f63c3ea},
171      {1, 0, 0, 0}},
172     {{0xacf39f4c5ca7ab, 0x4c8071cc5fd737, 0xc64e3602cd1184, 0x0acd4644c9abba},
173      {0x6c011a36d8bf6e, 0xfecd87ba24e32a, 0x19f6f56574fad8, 0x050b204ced9405},
174      {1, 0, 0, 0}},
175     {{0xed4f1cae7d9a96, 0x5ceef7ad94c40a, 0x778e4a3bf3ef9b, 0x7405783dc3b55e},
176      {0x32477c61b6e8c6, 0xb46a97570f018b, 0x91176d0a7e95d1, 0x3df90fbc4c7d0e},
177      {1, 0, 0, 0}}}};
178
179
180// Helper functions to convert field elements to/from internal representation
181
182static void p224_generic_to_felem(p224_felem out, const EC_FELEM *in) {
183  // |p224_felem|'s minimal representation uses four 56-bit words. |EC_FELEM|
184  // uses four 64-bit words. (The top-most word only has 32 bits.)
185  out[0] = in->words[0] & 0x00ffffffffffffff;
186  out[1] = ((in->words[0] >> 56) | (in->words[1] << 8)) & 0x00ffffffffffffff;
187  out[2] = ((in->words[1] >> 48) | (in->words[2] << 16)) & 0x00ffffffffffffff;
188  out[3] = ((in->words[2] >> 40) | (in->words[3] << 24)) & 0x00ffffffffffffff;
189}
190
191// Requires 0 <= in < 2*p (always call p224_felem_reduce first)
192static void p224_felem_to_generic(EC_FELEM *out, const p224_felem in) {
193  // Reduce to unique minimal representation.
194  static const int64_t two56 = ((p224_limb)1) << 56;
195  // 0 <= in < 2*p, p = 2^224 - 2^96 + 1
196  // if in > p , reduce in = in - 2^224 + 2^96 - 1
197  int64_t tmp[4], a;
198  tmp[0] = in[0];
199  tmp[1] = in[1];
200  tmp[2] = in[2];
201  tmp[3] = in[3];
202  // Case 1: a = 1 iff in >= 2^224
203  a = (in[3] >> 56);
204  tmp[0] -= a;
205  tmp[1] += a << 40;
206  tmp[3] &= 0x00ffffffffffffff;
207  // Case 2: a = 0 iff p <= in < 2^224, i.e., the high 128 bits are all 1 and
208  // the lower part is non-zero
209  a = ((in[3] & in[2] & (in[1] | 0x000000ffffffffff)) + 1) |
210      (((int64_t)(in[0] + (in[1] & 0x000000ffffffffff)) - 1) >> 63);
211  a &= 0x00ffffffffffffff;
212  // turn a into an all-one mask (if a = 0) or an all-zero mask
213  a = (a - 1) >> 63;
214  // subtract 2^224 - 2^96 + 1 if a is all-one
215  tmp[3] &= a ^ 0xffffffffffffffff;
216  tmp[2] &= a ^ 0xffffffffffffffff;
217  tmp[1] &= (a ^ 0xffffffffffffffff) | 0x000000ffffffffff;
218  tmp[0] -= 1 & a;
219
220  // eliminate negative coefficients: if tmp[0] is negative, tmp[1] must
221  // be non-zero, so we only need one step
222  a = tmp[0] >> 63;
223  tmp[0] += two56 & a;
224  tmp[1] -= 1 & a;
225
226  // carry 1 -> 2 -> 3
227  tmp[2] += tmp[1] >> 56;
228  tmp[1] &= 0x00ffffffffffffff;
229
230  tmp[3] += tmp[2] >> 56;
231  tmp[2] &= 0x00ffffffffffffff;
232
233  // Now 0 <= tmp < p
234  p224_felem tmp2;
235  tmp2[0] = tmp[0];
236  tmp2[1] = tmp[1];
237  tmp2[2] = tmp[2];
238  tmp2[3] = tmp[3];
239
240  // |p224_felem|'s minimal representation uses four 56-bit words. |EC_FELEM|
241  // uses four 64-bit words. (The top-most word only has 32 bits.)
242  out->words[0] = tmp2[0] | (tmp2[1] << 56);
243  out->words[1] = (tmp2[1] >> 8) | (tmp2[2] << 48);
244  out->words[2] = (tmp2[2] >> 16) | (tmp2[3] << 40);
245  out->words[3] = tmp2[3] >> 24;
246}
247
248
249// Field operations, using the internal representation of field elements.
250// NB! These operations are specific to our point multiplication and cannot be
251// expected to be correct in general - e.g., multiplication with a large scalar
252// will cause an overflow.
253
254static void p224_felem_assign(p224_felem out, const p224_felem in) {
255  out[0] = in[0];
256  out[1] = in[1];
257  out[2] = in[2];
258  out[3] = in[3];
259}
260
261// Sum two field elements: out += in
262static void p224_felem_sum(p224_felem out, const p224_felem in) {
263  out[0] += in[0];
264  out[1] += in[1];
265  out[2] += in[2];
266  out[3] += in[3];
267}
268
269// Subtract field elements: out -= in
270// Assumes in[i] < 2^57
271static void p224_felem_diff(p224_felem out, const p224_felem in) {
272  static const p224_limb two58p2 =
273      (((p224_limb)1) << 58) + (((p224_limb)1) << 2);
274  static const p224_limb two58m2 =
275      (((p224_limb)1) << 58) - (((p224_limb)1) << 2);
276  static const p224_limb two58m42m2 =
277      (((p224_limb)1) << 58) - (((p224_limb)1) << 42) - (((p224_limb)1) << 2);
278
279  // Add 0 mod 2^224-2^96+1 to ensure out > in
280  out[0] += two58p2;
281  out[1] += two58m42m2;
282  out[2] += two58m2;
283  out[3] += two58m2;
284
285  out[0] -= in[0];
286  out[1] -= in[1];
287  out[2] -= in[2];
288  out[3] -= in[3];
289}
290
291// Subtract in unreduced 128-bit mode: out -= in
292// Assumes in[i] < 2^119
293static void p224_widefelem_diff(p224_widefelem out, const p224_widefelem in) {
294  static const p224_widelimb two120 = ((p224_widelimb)1) << 120;
295  static const p224_widelimb two120m64 =
296      (((p224_widelimb)1) << 120) - (((p224_widelimb)1) << 64);
297  static const p224_widelimb two120m104m64 = (((p224_widelimb)1) << 120) -
298                                             (((p224_widelimb)1) << 104) -
299                                             (((p224_widelimb)1) << 64);
300
301  // Add 0 mod 2^224-2^96+1 to ensure out > in
302  out[0] += two120;
303  out[1] += two120m64;
304  out[2] += two120m64;
305  out[3] += two120;
306  out[4] += two120m104m64;
307  out[5] += two120m64;
308  out[6] += two120m64;
309
310  out[0] -= in[0];
311  out[1] -= in[1];
312  out[2] -= in[2];
313  out[3] -= in[3];
314  out[4] -= in[4];
315  out[5] -= in[5];
316  out[6] -= in[6];
317}
318
319// Subtract in mixed mode: out128 -= in64
320// in[i] < 2^63
321static void p224_felem_diff_128_64(p224_widefelem out, const p224_felem in) {
322  static const p224_widelimb two64p8 =
323      (((p224_widelimb)1) << 64) + (((p224_widelimb)1) << 8);
324  static const p224_widelimb two64m8 =
325      (((p224_widelimb)1) << 64) - (((p224_widelimb)1) << 8);
326  static const p224_widelimb two64m48m8 = (((p224_widelimb)1) << 64) -
327                                          (((p224_widelimb)1) << 48) -
328                                          (((p224_widelimb)1) << 8);
329
330  // Add 0 mod 2^224-2^96+1 to ensure out > in
331  out[0] += two64p8;
332  out[1] += two64m48m8;
333  out[2] += two64m8;
334  out[3] += two64m8;
335
336  out[0] -= in[0];
337  out[1] -= in[1];
338  out[2] -= in[2];
339  out[3] -= in[3];
340}
341
342// Multiply a field element by a scalar: out = out * scalar
343// The scalars we actually use are small, so results fit without overflow
344static void p224_felem_scalar(p224_felem out, const p224_limb scalar) {
345  out[0] *= scalar;
346  out[1] *= scalar;
347  out[2] *= scalar;
348  out[3] *= scalar;
349}
350
351// Multiply an unreduced field element by a scalar: out = out * scalar
352// The scalars we actually use are small, so results fit without overflow
353static void p224_widefelem_scalar(p224_widefelem out,
354                                  const p224_widelimb scalar) {
355  out[0] *= scalar;
356  out[1] *= scalar;
357  out[2] *= scalar;
358  out[3] *= scalar;
359  out[4] *= scalar;
360  out[5] *= scalar;
361  out[6] *= scalar;
362}
363
364// Square a field element: out = in^2
365static void p224_felem_square(p224_widefelem out, const p224_felem in) {
366  p224_limb tmp0, tmp1, tmp2;
367  tmp0 = 2 * in[0];
368  tmp1 = 2 * in[1];
369  tmp2 = 2 * in[2];
370  out[0] = ((p224_widelimb)in[0]) * in[0];
371  out[1] = ((p224_widelimb)in[0]) * tmp1;
372  out[2] = ((p224_widelimb)in[0]) * tmp2 + ((p224_widelimb)in[1]) * in[1];
373  out[3] = ((p224_widelimb)in[3]) * tmp0 + ((p224_widelimb)in[1]) * tmp2;
374  out[4] = ((p224_widelimb)in[3]) * tmp1 + ((p224_widelimb)in[2]) * in[2];
375  out[5] = ((p224_widelimb)in[3]) * tmp2;
376  out[6] = ((p224_widelimb)in[3]) * in[3];
377}
378
379// Multiply two field elements: out = in1 * in2
380static void p224_felem_mul(p224_widefelem out, const p224_felem in1,
381                           const p224_felem in2) {
382  out[0] = ((p224_widelimb)in1[0]) * in2[0];
383  out[1] = ((p224_widelimb)in1[0]) * in2[1] + ((p224_widelimb)in1[1]) * in2[0];
384  out[2] = ((p224_widelimb)in1[0]) * in2[2] + ((p224_widelimb)in1[1]) * in2[1] +
385           ((p224_widelimb)in1[2]) * in2[0];
386  out[3] = ((p224_widelimb)in1[0]) * in2[3] + ((p224_widelimb)in1[1]) * in2[2] +
387           ((p224_widelimb)in1[2]) * in2[1] + ((p224_widelimb)in1[3]) * in2[0];
388  out[4] = ((p224_widelimb)in1[1]) * in2[3] + ((p224_widelimb)in1[2]) * in2[2] +
389           ((p224_widelimb)in1[3]) * in2[1];
390  out[5] = ((p224_widelimb)in1[2]) * in2[3] + ((p224_widelimb)in1[3]) * in2[2];
391  out[6] = ((p224_widelimb)in1[3]) * in2[3];
392}
393
394// Reduce seven 128-bit coefficients to four 64-bit coefficients.
395// Requires in[i] < 2^126,
396// ensures out[0] < 2^56, out[1] < 2^56, out[2] < 2^56, out[3] <= 2^56 + 2^16
397static void p224_felem_reduce(p224_felem out, const p224_widefelem in) {
398  static const p224_widelimb two127p15 =
399      (((p224_widelimb)1) << 127) + (((p224_widelimb)1) << 15);
400  static const p224_widelimb two127m71 =
401      (((p224_widelimb)1) << 127) - (((p224_widelimb)1) << 71);
402  static const p224_widelimb two127m71m55 = (((p224_widelimb)1) << 127) -
403                                            (((p224_widelimb)1) << 71) -
404                                            (((p224_widelimb)1) << 55);
405  p224_widelimb output[5];
406
407  // Add 0 mod 2^224-2^96+1 to ensure all differences are positive
408  output[0] = in[0] + two127p15;
409  output[1] = in[1] + two127m71m55;
410  output[2] = in[2] + two127m71;
411  output[3] = in[3];
412  output[4] = in[4];
413
414  // Eliminate in[4], in[5], in[6]
415  output[4] += in[6] >> 16;
416  output[3] += (in[6] & 0xffff) << 40;
417  output[2] -= in[6];
418
419  output[3] += in[5] >> 16;
420  output[2] += (in[5] & 0xffff) << 40;
421  output[1] -= in[5];
422
423  output[2] += output[4] >> 16;
424  output[1] += (output[4] & 0xffff) << 40;
425  output[0] -= output[4];
426
427  // Carry 2 -> 3 -> 4
428  output[3] += output[2] >> 56;
429  output[2] &= 0x00ffffffffffffff;
430
431  output[4] = output[3] >> 56;
432  output[3] &= 0x00ffffffffffffff;
433
434  // Now output[2] < 2^56, output[3] < 2^56, output[4] < 2^72
435
436  // Eliminate output[4]
437  output[2] += output[4] >> 16;
438  // output[2] < 2^56 + 2^56 = 2^57
439  output[1] += (output[4] & 0xffff) << 40;
440  output[0] -= output[4];
441
442  // Carry 0 -> 1 -> 2 -> 3
443  output[1] += output[0] >> 56;
444  out[0] = output[0] & 0x00ffffffffffffff;
445
446  output[2] += output[1] >> 56;
447  // output[2] < 2^57 + 2^72
448  out[1] = output[1] & 0x00ffffffffffffff;
449  output[3] += output[2] >> 56;
450  // output[3] <= 2^56 + 2^16
451  out[2] = output[2] & 0x00ffffffffffffff;
452
453  // out[0] < 2^56, out[1] < 2^56, out[2] < 2^56,
454  // out[3] <= 2^56 + 2^16 (due to final carry),
455  // so out < 2*p
456  out[3] = output[3];
457}
458
459// Get negative value: out = -in
460// Requires in[i] < 2^63,
461// ensures out[0] < 2^56, out[1] < 2^56, out[2] < 2^56, out[3] <= 2^56 + 2^16
462static void p224_felem_neg(p224_felem out, const p224_felem in) {
463  p224_widefelem tmp = {0};
464  p224_felem_diff_128_64(tmp, in);
465  p224_felem_reduce(out, tmp);
466}
467
468// Zero-check: returns 1 if input is 0, and 0 otherwise. We know that field
469// elements are reduced to in < 2^225, so we only need to check three cases: 0,
470// 2^224 - 2^96 + 1, and 2^225 - 2^97 + 2
471static p224_limb p224_felem_is_zero(const p224_felem in) {
472  p224_limb zero = in[0] | in[1] | in[2] | in[3];
473  zero = (((int64_t)(zero)-1) >> 63) & 1;
474
475  p224_limb two224m96p1 = (in[0] ^ 1) | (in[1] ^ 0x00ffff0000000000) |
476                     (in[2] ^ 0x00ffffffffffffff) |
477                     (in[3] ^ 0x00ffffffffffffff);
478  two224m96p1 = (((int64_t)(two224m96p1)-1) >> 63) & 1;
479  p224_limb two225m97p2 = (in[0] ^ 2) | (in[1] ^ 0x00fffe0000000000) |
480                     (in[2] ^ 0x00ffffffffffffff) |
481                     (in[3] ^ 0x01ffffffffffffff);
482  two225m97p2 = (((int64_t)(two225m97p2)-1) >> 63) & 1;
483  return (zero | two224m96p1 | two225m97p2);
484}
485
486// Invert a field element
487// Computation chain copied from djb's code
488static void p224_felem_inv(p224_felem out, const p224_felem in) {
489  p224_felem ftmp, ftmp2, ftmp3, ftmp4;
490  p224_widefelem tmp;
491
492  p224_felem_square(tmp, in);
493  p224_felem_reduce(ftmp, tmp);  // 2
494  p224_felem_mul(tmp, in, ftmp);
495  p224_felem_reduce(ftmp, tmp);  // 2^2 - 1
496  p224_felem_square(tmp, ftmp);
497  p224_felem_reduce(ftmp, tmp);  // 2^3 - 2
498  p224_felem_mul(tmp, in, ftmp);
499  p224_felem_reduce(ftmp, tmp);  // 2^3 - 1
500  p224_felem_square(tmp, ftmp);
501  p224_felem_reduce(ftmp2, tmp);  // 2^4 - 2
502  p224_felem_square(tmp, ftmp2);
503  p224_felem_reduce(ftmp2, tmp);  // 2^5 - 4
504  p224_felem_square(tmp, ftmp2);
505  p224_felem_reduce(ftmp2, tmp);  // 2^6 - 8
506  p224_felem_mul(tmp, ftmp2, ftmp);
507  p224_felem_reduce(ftmp, tmp);  // 2^6 - 1
508  p224_felem_square(tmp, ftmp);
509  p224_felem_reduce(ftmp2, tmp);  // 2^7 - 2
510  for (size_t i = 0; i < 5; ++i) {  // 2^12 - 2^6
511    p224_felem_square(tmp, ftmp2);
512    p224_felem_reduce(ftmp2, tmp);
513  }
514  p224_felem_mul(tmp, ftmp2, ftmp);
515  p224_felem_reduce(ftmp2, tmp);  // 2^12 - 1
516  p224_felem_square(tmp, ftmp2);
517  p224_felem_reduce(ftmp3, tmp);  // 2^13 - 2
518  for (size_t i = 0; i < 11; ++i) {  // 2^24 - 2^12
519    p224_felem_square(tmp, ftmp3);
520    p224_felem_reduce(ftmp3, tmp);
521  }
522  p224_felem_mul(tmp, ftmp3, ftmp2);
523  p224_felem_reduce(ftmp2, tmp);  // 2^24 - 1
524  p224_felem_square(tmp, ftmp2);
525  p224_felem_reduce(ftmp3, tmp);  // 2^25 - 2
526  for (size_t i = 0; i < 23; ++i) {  // 2^48 - 2^24
527    p224_felem_square(tmp, ftmp3);
528    p224_felem_reduce(ftmp3, tmp);
529  }
530  p224_felem_mul(tmp, ftmp3, ftmp2);
531  p224_felem_reduce(ftmp3, tmp);  // 2^48 - 1
532  p224_felem_square(tmp, ftmp3);
533  p224_felem_reduce(ftmp4, tmp);  // 2^49 - 2
534  for (size_t i = 0; i < 47; ++i) {  // 2^96 - 2^48
535    p224_felem_square(tmp, ftmp4);
536    p224_felem_reduce(ftmp4, tmp);
537  }
538  p224_felem_mul(tmp, ftmp3, ftmp4);
539  p224_felem_reduce(ftmp3, tmp);  // 2^96 - 1
540  p224_felem_square(tmp, ftmp3);
541  p224_felem_reduce(ftmp4, tmp);  // 2^97 - 2
542  for (size_t i = 0; i < 23; ++i) {  // 2^120 - 2^24
543    p224_felem_square(tmp, ftmp4);
544    p224_felem_reduce(ftmp4, tmp);
545  }
546  p224_felem_mul(tmp, ftmp2, ftmp4);
547  p224_felem_reduce(ftmp2, tmp);  // 2^120 - 1
548  for (size_t i = 0; i < 6; ++i) {  // 2^126 - 2^6
549    p224_felem_square(tmp, ftmp2);
550    p224_felem_reduce(ftmp2, tmp);
551  }
552  p224_felem_mul(tmp, ftmp2, ftmp);
553  p224_felem_reduce(ftmp, tmp);  // 2^126 - 1
554  p224_felem_square(tmp, ftmp);
555  p224_felem_reduce(ftmp, tmp);  // 2^127 - 2
556  p224_felem_mul(tmp, ftmp, in);
557  p224_felem_reduce(ftmp, tmp);  // 2^127 - 1
558  for (size_t i = 0; i < 97; ++i) {  // 2^224 - 2^97
559    p224_felem_square(tmp, ftmp);
560    p224_felem_reduce(ftmp, tmp);
561  }
562  p224_felem_mul(tmp, ftmp, ftmp3);
563  p224_felem_reduce(out, tmp);  // 2^224 - 2^96 - 1
564}
565
566// Copy in constant time:
567// if icopy == 1, copy in to out,
568// if icopy == 0, copy out to itself.
569static void p224_copy_conditional(p224_felem out, const p224_felem in,
570                                  p224_limb icopy) {
571  // icopy is a (64-bit) 0 or 1, so copy is either all-zero or all-one
572  const p224_limb copy = -icopy;
573  for (size_t i = 0; i < 4; ++i) {
574    const p224_limb tmp = copy & (in[i] ^ out[i]);
575    out[i] ^= tmp;
576  }
577}
578
579// ELLIPTIC CURVE POINT OPERATIONS
580//
581// Points are represented in Jacobian projective coordinates:
582// (X, Y, Z) corresponds to the affine point (X/Z^2, Y/Z^3),
583// or to the point at infinity if Z == 0.
584
585// Double an elliptic curve point:
586// (X', Y', Z') = 2 * (X, Y, Z), where
587// X' = (3 * (X - Z^2) * (X + Z^2))^2 - 8 * X * Y^2
588// Y' = 3 * (X - Z^2) * (X + Z^2) * (4 * X * Y^2 - X') - 8 * Y^2
589// Z' = (Y + Z)^2 - Y^2 - Z^2 = 2 * Y * Z
590// Outputs can equal corresponding inputs, i.e., x_out == x_in is allowed,
591// while x_out == y_in is not (maybe this works, but it's not tested).
592static void p224_point_double(p224_felem x_out, p224_felem y_out,
593                              p224_felem z_out, const p224_felem x_in,
594                              const p224_felem y_in, const p224_felem z_in) {
595  p224_widefelem tmp, tmp2;
596  p224_felem delta, gamma, beta, alpha, ftmp, ftmp2;
597
598  p224_felem_assign(ftmp, x_in);
599  p224_felem_assign(ftmp2, x_in);
600
601  // delta = z^2
602  p224_felem_square(tmp, z_in);
603  p224_felem_reduce(delta, tmp);
604
605  // gamma = y^2
606  p224_felem_square(tmp, y_in);
607  p224_felem_reduce(gamma, tmp);
608
609  // beta = x*gamma
610  p224_felem_mul(tmp, x_in, gamma);
611  p224_felem_reduce(beta, tmp);
612
613  // alpha = 3*(x-delta)*(x+delta)
614  p224_felem_diff(ftmp, delta);
615  // ftmp[i] < 2^57 + 2^58 + 2 < 2^59
616  p224_felem_sum(ftmp2, delta);
617  // ftmp2[i] < 2^57 + 2^57 = 2^58
618  p224_felem_scalar(ftmp2, 3);
619  // ftmp2[i] < 3 * 2^58 < 2^60
620  p224_felem_mul(tmp, ftmp, ftmp2);
621  // tmp[i] < 2^60 * 2^59 * 4 = 2^121
622  p224_felem_reduce(alpha, tmp);
623
624  // x' = alpha^2 - 8*beta
625  p224_felem_square(tmp, alpha);
626  // tmp[i] < 4 * 2^57 * 2^57 = 2^116
627  p224_felem_assign(ftmp, beta);
628  p224_felem_scalar(ftmp, 8);
629  // ftmp[i] < 8 * 2^57 = 2^60
630  p224_felem_diff_128_64(tmp, ftmp);
631  // tmp[i] < 2^116 + 2^64 + 8 < 2^117
632  p224_felem_reduce(x_out, tmp);
633
634  // z' = (y + z)^2 - gamma - delta
635  p224_felem_sum(delta, gamma);
636  // delta[i] < 2^57 + 2^57 = 2^58
637  p224_felem_assign(ftmp, y_in);
638  p224_felem_sum(ftmp, z_in);
639  // ftmp[i] < 2^57 + 2^57 = 2^58
640  p224_felem_square(tmp, ftmp);
641  // tmp[i] < 4 * 2^58 * 2^58 = 2^118
642  p224_felem_diff_128_64(tmp, delta);
643  // tmp[i] < 2^118 + 2^64 + 8 < 2^119
644  p224_felem_reduce(z_out, tmp);
645
646  // y' = alpha*(4*beta - x') - 8*gamma^2
647  p224_felem_scalar(beta, 4);
648  // beta[i] < 4 * 2^57 = 2^59
649  p224_felem_diff(beta, x_out);
650  // beta[i] < 2^59 + 2^58 + 2 < 2^60
651  p224_felem_mul(tmp, alpha, beta);
652  // tmp[i] < 4 * 2^57 * 2^60 = 2^119
653  p224_felem_square(tmp2, gamma);
654  // tmp2[i] < 4 * 2^57 * 2^57 = 2^116
655  p224_widefelem_scalar(tmp2, 8);
656  // tmp2[i] < 8 * 2^116 = 2^119
657  p224_widefelem_diff(tmp, tmp2);
658  // tmp[i] < 2^119 + 2^120 < 2^121
659  p224_felem_reduce(y_out, tmp);
660}
661
662// Add two elliptic curve points:
663// (X_1, Y_1, Z_1) + (X_2, Y_2, Z_2) = (X_3, Y_3, Z_3), where
664// X_3 = (Z_1^3 * Y_2 - Z_2^3 * Y_1)^2 - (Z_1^2 * X_2 - Z_2^2 * X_1)^3 -
665// 2 * Z_2^2 * X_1 * (Z_1^2 * X_2 - Z_2^2 * X_1)^2
666// Y_3 = (Z_1^3 * Y_2 - Z_2^3 * Y_1) * (Z_2^2 * X_1 * (Z_1^2 * X_2 - Z_2^2 *
667// X_1)^2 - X_3) -
668//        Z_2^3 * Y_1 * (Z_1^2 * X_2 - Z_2^2 * X_1)^3
669// Z_3 = (Z_1^2 * X_2 - Z_2^2 * X_1) * (Z_1 * Z_2)
670//
671// This runs faster if 'mixed' is set, which requires Z_2 = 1 or Z_2 = 0.
672
673// This function is not entirely constant-time: it includes a branch for
674// checking whether the two input points are equal, (while not equal to the
675// point at infinity). This case never happens during single point
676// multiplication, so there is no timing leak for ECDH or ECDSA signing.
677static void p224_point_add(p224_felem x3, p224_felem y3, p224_felem z3,
678                           const p224_felem x1, const p224_felem y1,
679                           const p224_felem z1, const int mixed,
680                           const p224_felem x2, const p224_felem y2,
681                           const p224_felem z2) {
682  p224_felem ftmp, ftmp2, ftmp3, ftmp4, ftmp5, x_out, y_out, z_out;
683  p224_widefelem tmp, tmp2;
684  p224_limb z1_is_zero, z2_is_zero, x_equal, y_equal;
685
686  if (!mixed) {
687    // ftmp2 = z2^2
688    p224_felem_square(tmp, z2);
689    p224_felem_reduce(ftmp2, tmp);
690
691    // ftmp4 = z2^3
692    p224_felem_mul(tmp, ftmp2, z2);
693    p224_felem_reduce(ftmp4, tmp);
694
695    // ftmp4 = z2^3*y1
696    p224_felem_mul(tmp2, ftmp4, y1);
697    p224_felem_reduce(ftmp4, tmp2);
698
699    // ftmp2 = z2^2*x1
700    p224_felem_mul(tmp2, ftmp2, x1);
701    p224_felem_reduce(ftmp2, tmp2);
702  } else {
703    // We'll assume z2 = 1 (special case z2 = 0 is handled later)
704
705    // ftmp4 = z2^3*y1
706    p224_felem_assign(ftmp4, y1);
707
708    // ftmp2 = z2^2*x1
709    p224_felem_assign(ftmp2, x1);
710  }
711
712  // ftmp = z1^2
713  p224_felem_square(tmp, z1);
714  p224_felem_reduce(ftmp, tmp);
715
716  // ftmp3 = z1^3
717  p224_felem_mul(tmp, ftmp, z1);
718  p224_felem_reduce(ftmp3, tmp);
719
720  // tmp = z1^3*y2
721  p224_felem_mul(tmp, ftmp3, y2);
722  // tmp[i] < 4 * 2^57 * 2^57 = 2^116
723
724  // ftmp3 = z1^3*y2 - z2^3*y1
725  p224_felem_diff_128_64(tmp, ftmp4);
726  // tmp[i] < 2^116 + 2^64 + 8 < 2^117
727  p224_felem_reduce(ftmp3, tmp);
728
729  // tmp = z1^2*x2
730  p224_felem_mul(tmp, ftmp, x2);
731  // tmp[i] < 4 * 2^57 * 2^57 = 2^116
732
733  // ftmp = z1^2*x2 - z2^2*x1
734  p224_felem_diff_128_64(tmp, ftmp2);
735  // tmp[i] < 2^116 + 2^64 + 8 < 2^117
736  p224_felem_reduce(ftmp, tmp);
737
738  // The formulae are incorrect if the points are equal, so we check for this
739  // and do doubling if this happens.
740  x_equal = p224_felem_is_zero(ftmp);
741  y_equal = p224_felem_is_zero(ftmp3);
742  z1_is_zero = p224_felem_is_zero(z1);
743  z2_is_zero = p224_felem_is_zero(z2);
744  // In affine coordinates, (X_1, Y_1) == (X_2, Y_2)
745  p224_limb is_nontrivial_double =
746      x_equal & y_equal & (1 - z1_is_zero) & (1 - z2_is_zero);
747  if (constant_time_declassify_w(is_nontrivial_double)) {
748    p224_point_double(x3, y3, z3, x1, y1, z1);
749    return;
750  }
751
752  // ftmp5 = z1*z2
753  if (!mixed) {
754    p224_felem_mul(tmp, z1, z2);
755    p224_felem_reduce(ftmp5, tmp);
756  } else {
757    // special case z2 = 0 is handled later
758    p224_felem_assign(ftmp5, z1);
759  }
760
761  // z_out = (z1^2*x2 - z2^2*x1)*(z1*z2)
762  p224_felem_mul(tmp, ftmp, ftmp5);
763  p224_felem_reduce(z_out, tmp);
764
765  // ftmp = (z1^2*x2 - z2^2*x1)^2
766  p224_felem_assign(ftmp5, ftmp);
767  p224_felem_square(tmp, ftmp);
768  p224_felem_reduce(ftmp, tmp);
769
770  // ftmp5 = (z1^2*x2 - z2^2*x1)^3
771  p224_felem_mul(tmp, ftmp, ftmp5);
772  p224_felem_reduce(ftmp5, tmp);
773
774  // ftmp2 = z2^2*x1*(z1^2*x2 - z2^2*x1)^2
775  p224_felem_mul(tmp, ftmp2, ftmp);
776  p224_felem_reduce(ftmp2, tmp);
777
778  // tmp = z2^3*y1*(z1^2*x2 - z2^2*x1)^3
779  p224_felem_mul(tmp, ftmp4, ftmp5);
780  // tmp[i] < 4 * 2^57 * 2^57 = 2^116
781
782  // tmp2 = (z1^3*y2 - z2^3*y1)^2
783  p224_felem_square(tmp2, ftmp3);
784  // tmp2[i] < 4 * 2^57 * 2^57 < 2^116
785
786  // tmp2 = (z1^3*y2 - z2^3*y1)^2 - (z1^2*x2 - z2^2*x1)^3
787  p224_felem_diff_128_64(tmp2, ftmp5);
788  // tmp2[i] < 2^116 + 2^64 + 8 < 2^117
789
790  // ftmp5 = 2*z2^2*x1*(z1^2*x2 - z2^2*x1)^2
791  p224_felem_assign(ftmp5, ftmp2);
792  p224_felem_scalar(ftmp5, 2);
793  // ftmp5[i] < 2 * 2^57 = 2^58
794
795  /* x_out = (z1^3*y2 - z2^3*y1)^2 - (z1^2*x2 - z2^2*x1)^3 -
796     2*z2^2*x1*(z1^2*x2 - z2^2*x1)^2 */
797  p224_felem_diff_128_64(tmp2, ftmp5);
798  // tmp2[i] < 2^117 + 2^64 + 8 < 2^118
799  p224_felem_reduce(x_out, tmp2);
800
801  // ftmp2 = z2^2*x1*(z1^2*x2 - z2^2*x1)^2 - x_out
802  p224_felem_diff(ftmp2, x_out);
803  // ftmp2[i] < 2^57 + 2^58 + 2 < 2^59
804
805  // tmp2 = (z1^3*y2 - z2^3*y1)*(z2^2*x1*(z1^2*x2 - z2^2*x1)^2 - x_out)
806  p224_felem_mul(tmp2, ftmp3, ftmp2);
807  // tmp2[i] < 4 * 2^57 * 2^59 = 2^118
808
809  /* y_out = (z1^3*y2 - z2^3*y1)*(z2^2*x1*(z1^2*x2 - z2^2*x1)^2 - x_out) -
810     z2^3*y1*(z1^2*x2 - z2^2*x1)^3 */
811  p224_widefelem_diff(tmp2, tmp);
812  // tmp2[i] < 2^118 + 2^120 < 2^121
813  p224_felem_reduce(y_out, tmp2);
814
815  // the result (x_out, y_out, z_out) is incorrect if one of the inputs is
816  // the point at infinity, so we need to check for this separately
817
818  // if point 1 is at infinity, copy point 2 to output, and vice versa
819  p224_copy_conditional(x_out, x2, z1_is_zero);
820  p224_copy_conditional(x_out, x1, z2_is_zero);
821  p224_copy_conditional(y_out, y2, z1_is_zero);
822  p224_copy_conditional(y_out, y1, z2_is_zero);
823  p224_copy_conditional(z_out, z2, z1_is_zero);
824  p224_copy_conditional(z_out, z1, z2_is_zero);
825  p224_felem_assign(x3, x_out);
826  p224_felem_assign(y3, y_out);
827  p224_felem_assign(z3, z_out);
828}
829
830// p224_select_point selects the |idx|th point from a precomputation table and
831// copies it to out.
832static void p224_select_point(const uint64_t idx, size_t size,
833                              const p224_felem pre_comp[/*size*/][3],
834                              p224_felem out[3]) {
835  p224_limb *outlimbs = &out[0][0];
836  OPENSSL_memset(outlimbs, 0, 3 * sizeof(p224_felem));
837
838  for (size_t i = 0; i < size; i++) {
839    const p224_limb *inlimbs = &pre_comp[i][0][0];
840    static_assert(sizeof(uint64_t) <= sizeof(crypto_word_t),
841                  "crypto_word_t too small");
842    static_assert(sizeof(size_t) <= sizeof(crypto_word_t),
843                  "crypto_word_t too small");
844    // Without a value barrier, Clang adds a branch here.
845    uint64_t mask = value_barrier_w(constant_time_eq_w(i, idx));
846    for (size_t j = 0; j < 4 * 3; j++) {
847      outlimbs[j] |= inlimbs[j] & mask;
848    }
849  }
850}
851
852// p224_get_bit returns the |i|th bit in |in|.
853static crypto_word_t p224_get_bit(const EC_SCALAR *in, size_t i) {
854  if (i >= 224) {
855    return 0;
856  }
857  static_assert(sizeof(in->words[0]) == 8, "BN_ULONG is not 64-bit");
858  return (in->words[i >> 6] >> (i & 63)) & 1;
859}
860
861// Takes the Jacobian coordinates (X, Y, Z) of a point and returns
862// (X', Y') = (X/Z^2, Y/Z^3)
863static int ec_GFp_nistp224_point_get_affine_coordinates(
864    const EC_GROUP *group, const EC_JACOBIAN *point, EC_FELEM *x,
865    EC_FELEM *y) {
866  if (constant_time_declassify_int(
867          ec_GFp_simple_is_at_infinity(group, point))) {
868    OPENSSL_PUT_ERROR(EC, EC_R_POINT_AT_INFINITY);
869    return 0;
870  }
871
872  p224_felem z1, z2;
873  p224_widefelem tmp;
874  p224_generic_to_felem(z1, &point->Z);
875  p224_felem_inv(z2, z1);
876  p224_felem_square(tmp, z2);
877  p224_felem_reduce(z1, tmp);
878
879  if (x != NULL) {
880    p224_felem x_in, x_out;
881    p224_generic_to_felem(x_in, &point->X);
882    p224_felem_mul(tmp, x_in, z1);
883    p224_felem_reduce(x_out, tmp);
884    p224_felem_to_generic(x, x_out);
885  }
886
887  if (y != NULL) {
888    p224_felem y_in, y_out;
889    p224_generic_to_felem(y_in, &point->Y);
890    p224_felem_mul(tmp, z1, z2);
891    p224_felem_reduce(z1, tmp);
892    p224_felem_mul(tmp, y_in, z1);
893    p224_felem_reduce(y_out, tmp);
894    p224_felem_to_generic(y, y_out);
895  }
896
897  return 1;
898}
899
900static void ec_GFp_nistp224_add(const EC_GROUP *group, EC_JACOBIAN *r,
901                                const EC_JACOBIAN *a, const EC_JACOBIAN *b) {
902  p224_felem x1, y1, z1, x2, y2, z2;
903  p224_generic_to_felem(x1, &a->X);
904  p224_generic_to_felem(y1, &a->Y);
905  p224_generic_to_felem(z1, &a->Z);
906  p224_generic_to_felem(x2, &b->X);
907  p224_generic_to_felem(y2, &b->Y);
908  p224_generic_to_felem(z2, &b->Z);
909  p224_point_add(x1, y1, z1, x1, y1, z1, 0 /* both Jacobian */, x2, y2, z2);
910  // The outputs are already reduced, but still need to be contracted.
911  p224_felem_to_generic(&r->X, x1);
912  p224_felem_to_generic(&r->Y, y1);
913  p224_felem_to_generic(&r->Z, z1);
914}
915
916static void ec_GFp_nistp224_dbl(const EC_GROUP *group, EC_JACOBIAN *r,
917                                const EC_JACOBIAN *a) {
918  p224_felem x, y, z;
919  p224_generic_to_felem(x, &a->X);
920  p224_generic_to_felem(y, &a->Y);
921  p224_generic_to_felem(z, &a->Z);
922  p224_point_double(x, y, z, x, y, z);
923  // The outputs are already reduced, but still need to be contracted.
924  p224_felem_to_generic(&r->X, x);
925  p224_felem_to_generic(&r->Y, y);
926  p224_felem_to_generic(&r->Z, z);
927}
928
929static void ec_GFp_nistp224_make_precomp(p224_felem out[17][3],
930                                         const EC_JACOBIAN *p) {
931  OPENSSL_memset(out[0], 0, sizeof(p224_felem) * 3);
932
933  p224_generic_to_felem(out[1][0], &p->X);
934  p224_generic_to_felem(out[1][1], &p->Y);
935  p224_generic_to_felem(out[1][2], &p->Z);
936
937  for (size_t j = 2; j <= 16; ++j) {
938    if (j & 1) {
939      p224_point_add(out[j][0], out[j][1], out[j][2], out[1][0], out[1][1],
940                     out[1][2], 0, out[j - 1][0], out[j - 1][1], out[j - 1][2]);
941    } else {
942      p224_point_double(out[j][0], out[j][1], out[j][2], out[j / 2][0],
943                        out[j / 2][1], out[j / 2][2]);
944    }
945  }
946}
947
948static void ec_GFp_nistp224_point_mul(const EC_GROUP *group, EC_JACOBIAN *r,
949                                      const EC_JACOBIAN *p,
950                                      const EC_SCALAR *scalar) {
951  p224_felem p_pre_comp[17][3];
952  ec_GFp_nistp224_make_precomp(p_pre_comp, p);
953
954  // Set nq to the point at infinity.
955  p224_felem nq[3], tmp[4];
956  OPENSSL_memset(nq, 0, 3 * sizeof(p224_felem));
957
958  int skip = 1;  // Save two point operations in the first round.
959  for (size_t i = 220; i < 221; i--) {
960    if (!skip) {
961      p224_point_double(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2]);
962    }
963
964    // Add every 5 doublings.
965    if (i % 5 == 0) {
966      crypto_word_t bits = p224_get_bit(scalar, i + 4) << 5;
967      bits |= p224_get_bit(scalar, i + 3) << 4;
968      bits |= p224_get_bit(scalar, i + 2) << 3;
969      bits |= p224_get_bit(scalar, i + 1) << 2;
970      bits |= p224_get_bit(scalar, i) << 1;
971      bits |= p224_get_bit(scalar, i - 1);
972      crypto_word_t sign, digit;
973      ec_GFp_nistp_recode_scalar_bits(&sign, &digit, bits);
974
975      // Select the point to add or subtract.
976      p224_select_point(digit, 17, (const p224_felem(*)[3])p_pre_comp, tmp);
977      p224_felem_neg(tmp[3], tmp[1]);  // (X, -Y, Z) is the negative point
978      p224_copy_conditional(tmp[1], tmp[3], sign);
979
980      if (!skip) {
981        p224_point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 0 /* mixed */,
982                       tmp[0], tmp[1], tmp[2]);
983      } else {
984        OPENSSL_memcpy(nq, tmp, 3 * sizeof(p224_felem));
985        skip = 0;
986      }
987    }
988  }
989
990  // Reduce the output to its unique minimal representation.
991  p224_felem_to_generic(&r->X, nq[0]);
992  p224_felem_to_generic(&r->Y, nq[1]);
993  p224_felem_to_generic(&r->Z, nq[2]);
994}
995
996static void ec_GFp_nistp224_point_mul_base(const EC_GROUP *group,
997                                           EC_JACOBIAN *r,
998                                           const EC_SCALAR *scalar) {
999  // Set nq to the point at infinity.
1000  p224_felem nq[3], tmp[3];
1001  OPENSSL_memset(nq, 0, 3 * sizeof(p224_felem));
1002
1003  int skip = 1;  // Save two point operations in the first round.
1004  for (size_t i = 27; i < 28; i--) {
1005    // double
1006    if (!skip) {
1007      p224_point_double(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2]);
1008    }
1009
1010    // First, look 28 bits upwards.
1011    crypto_word_t bits = p224_get_bit(scalar, i + 196) << 3;
1012    bits |= p224_get_bit(scalar, i + 140) << 2;
1013    bits |= p224_get_bit(scalar, i + 84) << 1;
1014    bits |= p224_get_bit(scalar, i + 28);
1015    // Select the point to add, in constant time.
1016    p224_select_point(bits, 16, g_p224_pre_comp[1], tmp);
1017
1018    if (!skip) {
1019      p224_point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1 /* mixed */,
1020                     tmp[0], tmp[1], tmp[2]);
1021    } else {
1022      OPENSSL_memcpy(nq, tmp, 3 * sizeof(p224_felem));
1023      skip = 0;
1024    }
1025
1026    // Second, look at the current position/
1027    bits = p224_get_bit(scalar, i + 168) << 3;
1028    bits |= p224_get_bit(scalar, i + 112) << 2;
1029    bits |= p224_get_bit(scalar, i + 56) << 1;
1030    bits |= p224_get_bit(scalar, i);
1031    // Select the point to add, in constant time.
1032    p224_select_point(bits, 16, g_p224_pre_comp[0], tmp);
1033    p224_point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1 /* mixed */,
1034                   tmp[0], tmp[1], tmp[2]);
1035  }
1036
1037  // Reduce the output to its unique minimal representation.
1038  p224_felem_to_generic(&r->X, nq[0]);
1039  p224_felem_to_generic(&r->Y, nq[1]);
1040  p224_felem_to_generic(&r->Z, nq[2]);
1041}
1042
1043static void ec_GFp_nistp224_point_mul_public(const EC_GROUP *group,
1044                                             EC_JACOBIAN *r,
1045                                             const EC_SCALAR *g_scalar,
1046                                             const EC_JACOBIAN *p,
1047                                             const EC_SCALAR *p_scalar) {
1048  // TODO(davidben): If P-224 ECDSA verify performance ever matters, using
1049  // |ec_compute_wNAF| for |p_scalar| would likely be an easy improvement.
1050  p224_felem p_pre_comp[17][3];
1051  ec_GFp_nistp224_make_precomp(p_pre_comp, p);
1052
1053  // Set nq to the point at infinity.
1054  p224_felem nq[3], tmp[3];
1055  OPENSSL_memset(nq, 0, 3 * sizeof(p224_felem));
1056
1057  // Loop over both scalars msb-to-lsb, interleaving additions of multiples of
1058  // the generator (two in each of the last 28 rounds) and additions of p (every
1059  // 5th round).
1060  int skip = 1;  // Save two point operations in the first round.
1061  for (size_t i = 220; i < 221; i--) {
1062    if (!skip) {
1063      p224_point_double(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2]);
1064    }
1065
1066    // Add multiples of the generator.
1067    if (i <= 27) {
1068      // First, look 28 bits upwards.
1069      crypto_word_t bits = p224_get_bit(g_scalar, i + 196) << 3;
1070      bits |= p224_get_bit(g_scalar, i + 140) << 2;
1071      bits |= p224_get_bit(g_scalar, i + 84) << 1;
1072      bits |= p224_get_bit(g_scalar, i + 28);
1073
1074      size_t index = (size_t)bits;
1075      p224_point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1 /* mixed */,
1076                     g_p224_pre_comp[1][index][0], g_p224_pre_comp[1][index][1],
1077                     g_p224_pre_comp[1][index][2]);
1078      assert(!skip);
1079
1080      // Second, look at the current position.
1081      bits = p224_get_bit(g_scalar, i + 168) << 3;
1082      bits |= p224_get_bit(g_scalar, i + 112) << 2;
1083      bits |= p224_get_bit(g_scalar, i + 56) << 1;
1084      bits |= p224_get_bit(g_scalar, i);
1085      index = (size_t)bits;
1086      p224_point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1 /* mixed */,
1087                     g_p224_pre_comp[0][index][0], g_p224_pre_comp[0][index][1],
1088                     g_p224_pre_comp[0][index][2]);
1089    }
1090
1091    // Incorporate |p_scalar| every 5 doublings.
1092    if (i % 5 == 0) {
1093      crypto_word_t bits = p224_get_bit(p_scalar, i + 4) << 5;
1094      bits |= p224_get_bit(p_scalar, i + 3) << 4;
1095      bits |= p224_get_bit(p_scalar, i + 2) << 3;
1096      bits |= p224_get_bit(p_scalar, i + 1) << 2;
1097      bits |= p224_get_bit(p_scalar, i) << 1;
1098      bits |= p224_get_bit(p_scalar, i - 1);
1099      crypto_word_t sign, digit;
1100      ec_GFp_nistp_recode_scalar_bits(&sign, &digit, bits);
1101
1102      // Select the point to add or subtract.
1103      OPENSSL_memcpy(tmp, p_pre_comp[digit], 3 * sizeof(p224_felem));
1104      if (sign) {
1105        p224_felem_neg(tmp[1], tmp[1]);  // (X, -Y, Z) is the negative point
1106      }
1107
1108      if (!skip) {
1109        p224_point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 0 /* mixed */,
1110                       tmp[0], tmp[1], tmp[2]);
1111      } else {
1112        OPENSSL_memcpy(nq, tmp, 3 * sizeof(p224_felem));
1113        skip = 0;
1114      }
1115    }
1116  }
1117
1118  // Reduce the output to its unique minimal representation.
1119  p224_felem_to_generic(&r->X, nq[0]);
1120  p224_felem_to_generic(&r->Y, nq[1]);
1121  p224_felem_to_generic(&r->Z, nq[2]);
1122}
1123
1124static void ec_GFp_nistp224_felem_mul(const EC_GROUP *group, EC_FELEM *r,
1125                                      const EC_FELEM *a, const EC_FELEM *b) {
1126  p224_felem felem1, felem2;
1127  p224_widefelem wide;
1128  p224_generic_to_felem(felem1, a);
1129  p224_generic_to_felem(felem2, b);
1130  p224_felem_mul(wide, felem1, felem2);
1131  p224_felem_reduce(felem1, wide);
1132  p224_felem_to_generic(r, felem1);
1133}
1134
1135static void ec_GFp_nistp224_felem_sqr(const EC_GROUP *group, EC_FELEM *r,
1136                                      const EC_FELEM *a) {
1137  p224_felem felem;
1138  p224_generic_to_felem(felem, a);
1139  p224_widefelem wide;
1140  p224_felem_square(wide, felem);
1141  p224_felem_reduce(felem, wide);
1142  p224_felem_to_generic(r, felem);
1143}
1144
1145DEFINE_METHOD_FUNCTION(EC_METHOD, EC_GFp_nistp224_method) {
1146  out->point_get_affine_coordinates =
1147      ec_GFp_nistp224_point_get_affine_coordinates;
1148  out->add = ec_GFp_nistp224_add;
1149  out->dbl = ec_GFp_nistp224_dbl;
1150  out->mul = ec_GFp_nistp224_point_mul;
1151  out->mul_base = ec_GFp_nistp224_point_mul_base;
1152  out->mul_public = ec_GFp_nistp224_point_mul_public;
1153  out->felem_mul = ec_GFp_nistp224_felem_mul;
1154  out->felem_sqr = ec_GFp_nistp224_felem_sqr;
1155  out->felem_to_bytes = ec_GFp_simple_felem_to_bytes;
1156  out->felem_from_bytes = ec_GFp_simple_felem_from_bytes;
1157  out->scalar_inv0_montgomery = ec_simple_scalar_inv0_montgomery;
1158  out->scalar_to_montgomery_inv_vartime =
1159      ec_simple_scalar_to_montgomery_inv_vartime;
1160  out->cmp_x_coordinate = ec_GFp_simple_cmp_x_coordinate;
1161}
1162
1163#endif  // BORINGSSL_HAS_UINT128 && !SMALL
1164