1/*
2 * Copyright 2013-2016 The OpenSSL Project Authors. All Rights Reserved.
3 * Copyright (c) 2012, Intel Corporation. All Rights Reserved.
4 *
5 * Licensed under the OpenSSL license (the "License"). You may not use
6 * this file except in compliance with the License. You can obtain a copy
7 * in the file LICENSE in the source distribution or at
8 * https://www.openssl.org/source/license.html
9 *
10 * Originally written by Shay Gueron (1, 2), and Vlad Krasnov (1)
11 * (1) Intel Corporation, Israel Development Center, Haifa, Israel
12 * (2) University of Haifa, Israel
13 */
14
15#include "rsaz_exp.h"
16
17#if defined(RSAZ_ENABLED)
18
19#include <openssl/mem.h>
20
21#include "internal.h"
22#include "../../internal.h"
23
24
25// one is 1 in RSAZ's representation.
26alignas(64) static const BN_ULONG one[40] = {
27 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
28 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
29// two80 is 2^80 in RSAZ's representation. Note RSAZ uses base 2^29, so this is
30// 2^(29*2 + 22) = 2^80, not 2^(64*2 + 22).
31alignas(64) static const BN_ULONG two80[40] = {
32 0, 0, 1 << 22, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
33 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
34
35void RSAZ_1024_mod_exp_avx2(BN_ULONG result_norm[16],
36 const BN_ULONG base_norm[16],
37 const BN_ULONG exponent[16],
38 const BN_ULONG m_norm[16], const BN_ULONG RR[16],
39 BN_ULONG k0,
40 BN_ULONG storage[MOD_EXP_CTIME_STORAGE_LEN]) {
41 OPENSSL_STATIC_ASSERT(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH % 64 == 0,
42 "MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH is too small");
43 assert((uintptr_t)storage % 64 == 0);
44
45 BN_ULONG *a_inv, *m, *result, *table_s = storage + 40 * 3, *R2 = table_s;
46 // Note |R2| aliases |table_s|.
47 if (((((uintptr_t)storage & 4095) + 320) >> 12) != 0) {
48 result = storage;
49 a_inv = storage + 40;
50 m = storage + 40 * 2; // should not cross page
51 } else {
52 m = storage; // should not cross page
53 result = storage + 40;
54 a_inv = storage + 40 * 2;
55 }
56
57 rsaz_1024_norm2red_avx2(m, m_norm);
58 rsaz_1024_norm2red_avx2(a_inv, base_norm);
59 rsaz_1024_norm2red_avx2(R2, RR);
60
61 // Convert |R2| from the usual radix, giving R = 2^1024, to RSAZ's radix,
62 // giving R = 2^(36*29) = 2^1044.
63 rsaz_1024_mul_avx2(R2, R2, R2, m, k0);
64 // R2 = 2^2048 * 2^2048 / 2^1044 = 2^3052
65 rsaz_1024_mul_avx2(R2, R2, two80, m, k0);
66 // R2 = 2^3052 * 2^80 / 2^1044 = 2^2088 = (2^1044)^2
67
68 // table[0] = 1
69 rsaz_1024_mul_avx2(result, R2, one, m, k0);
70 // table[1] = a_inv^1
71 rsaz_1024_mul_avx2(a_inv, a_inv, R2, m, k0);
72
73 rsaz_1024_scatter5_avx2(table_s, result, 0);
74 rsaz_1024_scatter5_avx2(table_s, a_inv, 1);
75
76 // table[2] = a_inv^2
77 rsaz_1024_sqr_avx2(result, a_inv, m, k0, 1);
78 rsaz_1024_scatter5_avx2(table_s, result, 2);
79#if 0
80 // This is almost 2x smaller and less than 1% slower.
81 for (int index = 3; index < 32; index++) {
82 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
83 rsaz_1024_scatter5_avx2(table_s, result, index);
84 }
85#else
86 // table[4] = a_inv^4
87 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
88 rsaz_1024_scatter5_avx2(table_s, result, 4);
89 // table[8] = a_inv^8
90 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
91 rsaz_1024_scatter5_avx2(table_s, result, 8);
92 // table[16] = a_inv^16
93 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
94 rsaz_1024_scatter5_avx2(table_s, result, 16);
95 // table[17] = a_inv^17
96 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
97 rsaz_1024_scatter5_avx2(table_s, result, 17);
98
99 // table[3]
100 rsaz_1024_gather5_avx2(result, table_s, 2);
101 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
102 rsaz_1024_scatter5_avx2(table_s, result, 3);
103 // table[6]
104 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
105 rsaz_1024_scatter5_avx2(table_s, result, 6);
106 // table[12]
107 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
108 rsaz_1024_scatter5_avx2(table_s, result, 12);
109 // table[24]
110 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
111 rsaz_1024_scatter5_avx2(table_s, result, 24);
112 // table[25]
113 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
114 rsaz_1024_scatter5_avx2(table_s, result, 25);
115
116 // table[5]
117 rsaz_1024_gather5_avx2(result, table_s, 4);
118 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
119 rsaz_1024_scatter5_avx2(table_s, result, 5);
120 // table[10]
121 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
122 rsaz_1024_scatter5_avx2(table_s, result, 10);
123 // table[20]
124 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
125 rsaz_1024_scatter5_avx2(table_s, result, 20);
126 // table[21]
127 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
128 rsaz_1024_scatter5_avx2(table_s, result, 21);
129
130 // table[7]
131 rsaz_1024_gather5_avx2(result, table_s, 6);
132 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
133 rsaz_1024_scatter5_avx2(table_s, result, 7);
134 // table[14]
135 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
136 rsaz_1024_scatter5_avx2(table_s, result, 14);
137 // table[28]
138 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
139 rsaz_1024_scatter5_avx2(table_s, result, 28);
140 // table[29]
141 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
142 rsaz_1024_scatter5_avx2(table_s, result, 29);
143
144 // table[9]
145 rsaz_1024_gather5_avx2(result, table_s, 8);
146 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
147 rsaz_1024_scatter5_avx2(table_s, result, 9);
148 // table[18]
149 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
150 rsaz_1024_scatter5_avx2(table_s, result, 18);
151 // table[19]
152 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
153 rsaz_1024_scatter5_avx2(table_s, result, 19);
154
155 // table[11]
156 rsaz_1024_gather5_avx2(result, table_s, 10);
157 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
158 rsaz_1024_scatter5_avx2(table_s, result, 11);
159 // table[22]
160 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
161 rsaz_1024_scatter5_avx2(table_s, result, 22);
162 // table[23]
163 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
164 rsaz_1024_scatter5_avx2(table_s, result, 23);
165
166 // table[13]
167 rsaz_1024_gather5_avx2(result, table_s, 12);
168 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
169 rsaz_1024_scatter5_avx2(table_s, result, 13);
170 // table[26]
171 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
172 rsaz_1024_scatter5_avx2(table_s, result, 26);
173 // table[27]
174 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
175 rsaz_1024_scatter5_avx2(table_s, result, 27);
176
177 // table[15]
178 rsaz_1024_gather5_avx2(result, table_s, 14);
179 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
180 rsaz_1024_scatter5_avx2(table_s, result, 15);
181 // table[30]
182 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
183 rsaz_1024_scatter5_avx2(table_s, result, 30);
184 // table[31]
185 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
186 rsaz_1024_scatter5_avx2(table_s, result, 31);
187#endif
188
189 const uint8_t *p_str = (const uint8_t *)exponent;
190
191 // load first window
192 int wvalue = p_str[127] >> 3;
193 rsaz_1024_gather5_avx2(result, table_s, wvalue);
194
195 int index = 1014;
196 while (index > -1) { // Loop for the remaining 127 windows.
197
198 rsaz_1024_sqr_avx2(result, result, m, k0, 5);
199
200 uint16_t wvalue_16;
201 memcpy(&wvalue_16, &p_str[index / 8], sizeof(wvalue_16));
202 wvalue = wvalue_16;
203 wvalue = (wvalue >> (index % 8)) & 31;
204 index -= 5;
205
206 rsaz_1024_gather5_avx2(a_inv, table_s, wvalue); // Borrow |a_inv|.
207 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
208 }
209
210 // Square four times.
211 rsaz_1024_sqr_avx2(result, result, m, k0, 4);
212
213 wvalue = p_str[0] & 15;
214
215 rsaz_1024_gather5_avx2(a_inv, table_s, wvalue); // Borrow |a_inv|.
216 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
217
218 // Convert from Montgomery.
219 rsaz_1024_mul_avx2(result, result, one, m, k0);
220
221 rsaz_1024_red2norm_avx2(result_norm, result);
222
223 OPENSSL_cleanse(storage, MOD_EXP_CTIME_STORAGE_LEN * sizeof(BN_ULONG));
224}
225
226#endif // RSAZ_ENABLED
227