1/*
2 * Copyright 2013-2022 The OpenSSL Project Authors. All Rights Reserved.
3 * Copyright (c) 2012, Intel Corporation. All Rights Reserved.
4 *
5 * Licensed under the Apache License 2.0 (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 <openssl/opensslconf.h>
16#include "rsaz_exp.h"
17
18#ifndef RSAZ_ENABLED
19NON_EMPTY_TRANSLATION_UNIT
20#else
21
22/*
23 * See crypto/bn/asm/rsaz-avx2.pl for further details.
24 */
25void rsaz_1024_norm2red_avx2(void *red, const void *norm);
26void rsaz_1024_mul_avx2(void *ret, const void *a, const void *b,
27                        const void *n, BN_ULONG k);
28void rsaz_1024_sqr_avx2(void *ret, const void *a, const void *n, BN_ULONG k,
29                        int cnt);
30void rsaz_1024_scatter5_avx2(void *tbl, const void *val, int i);
31void rsaz_1024_gather5_avx2(void *val, const void *tbl, int i);
32void rsaz_1024_red2norm_avx2(void *norm, const void *red);
33
34#if defined(__GNUC__)
35# define ALIGN64        __attribute__((aligned(64)))
36#elif defined(_MSC_VER)
37# define ALIGN64        __declspec(align(64))
38#elif defined(__SUNPRO_C)
39# define ALIGN64
40# pragma align 64(one,two80)
41#else
42/* not fatal, might hurt performance a little */
43# define ALIGN64
44#endif
45
46ALIGN64 static const BN_ULONG one[40] = {
47    1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
48    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
49};
50
51ALIGN64 static const BN_ULONG two80[40] = {
52    0, 0, 1 << 22, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
53    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
54};
55
56void RSAZ_1024_mod_exp_avx2(BN_ULONG result_norm[16],
57                            const BN_ULONG base_norm[16],
58                            const BN_ULONG exponent[16],
59                            const BN_ULONG m_norm[16], const BN_ULONG RR[16],
60                            BN_ULONG k0)
61{
62    unsigned char storage[320 * 3 + 32 * 9 * 16 + 64]; /* 5.5KB */
63    unsigned char *p_str = storage + (64 - ((size_t)storage % 64));
64    unsigned char *a_inv, *m, *result;
65    unsigned char *table_s = p_str + 320 * 3;
66    unsigned char *R2 = table_s; /* borrow */
67    int index;
68    int wvalue;
69    BN_ULONG tmp[16];
70
71    if ((((size_t)p_str & 4095) + 320) >> 12) {
72        result = p_str;
73        a_inv = p_str + 320;
74        m = p_str + 320 * 2;    /* should not cross page */
75    } else {
76        m = p_str;              /* should not cross page */
77        result = p_str + 320;
78        a_inv = p_str + 320 * 2;
79    }
80
81    rsaz_1024_norm2red_avx2(m, m_norm);
82    rsaz_1024_norm2red_avx2(a_inv, base_norm);
83    rsaz_1024_norm2red_avx2(R2, RR);
84
85    rsaz_1024_mul_avx2(R2, R2, R2, m, k0);
86    rsaz_1024_mul_avx2(R2, R2, two80, m, k0);
87
88    /* table[0] = 1 */
89    rsaz_1024_mul_avx2(result, R2, one, m, k0);
90    /* table[1] = a_inv^1 */
91    rsaz_1024_mul_avx2(a_inv, a_inv, R2, m, k0);
92
93    rsaz_1024_scatter5_avx2(table_s, result, 0);
94    rsaz_1024_scatter5_avx2(table_s, a_inv, 1);
95
96    /* table[2] = a_inv^2 */
97    rsaz_1024_sqr_avx2(result, a_inv, m, k0, 1);
98    rsaz_1024_scatter5_avx2(table_s, result, 2);
99#if 0
100    /* this is almost 2x smaller and less than 1% slower */
101    for (index = 3; index < 32; index++) {
102        rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
103        rsaz_1024_scatter5_avx2(table_s, result, index);
104    }
105#else
106    /* table[4] = a_inv^4 */
107    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
108    rsaz_1024_scatter5_avx2(table_s, result, 4);
109    /* table[8] = a_inv^8 */
110    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
111    rsaz_1024_scatter5_avx2(table_s, result, 8);
112    /* table[16] = a_inv^16 */
113    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
114    rsaz_1024_scatter5_avx2(table_s, result, 16);
115    /* table[17] = a_inv^17 */
116    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
117    rsaz_1024_scatter5_avx2(table_s, result, 17);
118
119    /* table[3] */
120    rsaz_1024_gather5_avx2(result, table_s, 2);
121    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
122    rsaz_1024_scatter5_avx2(table_s, result, 3);
123    /* table[6] */
124    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
125    rsaz_1024_scatter5_avx2(table_s, result, 6);
126    /* table[12] */
127    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
128    rsaz_1024_scatter5_avx2(table_s, result, 12);
129    /* table[24] */
130    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
131    rsaz_1024_scatter5_avx2(table_s, result, 24);
132    /* table[25] */
133    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
134    rsaz_1024_scatter5_avx2(table_s, result, 25);
135
136    /* table[5] */
137    rsaz_1024_gather5_avx2(result, table_s, 4);
138    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
139    rsaz_1024_scatter5_avx2(table_s, result, 5);
140    /* table[10] */
141    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
142    rsaz_1024_scatter5_avx2(table_s, result, 10);
143    /* table[20] */
144    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
145    rsaz_1024_scatter5_avx2(table_s, result, 20);
146    /* table[21] */
147    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
148    rsaz_1024_scatter5_avx2(table_s, result, 21);
149
150    /* table[7] */
151    rsaz_1024_gather5_avx2(result, table_s, 6);
152    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
153    rsaz_1024_scatter5_avx2(table_s, result, 7);
154    /* table[14] */
155    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
156    rsaz_1024_scatter5_avx2(table_s, result, 14);
157    /* table[28] */
158    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
159    rsaz_1024_scatter5_avx2(table_s, result, 28);
160    /* table[29] */
161    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
162    rsaz_1024_scatter5_avx2(table_s, result, 29);
163
164    /* table[9] */
165    rsaz_1024_gather5_avx2(result, table_s, 8);
166    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
167    rsaz_1024_scatter5_avx2(table_s, result, 9);
168    /* table[18] */
169    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
170    rsaz_1024_scatter5_avx2(table_s, result, 18);
171    /* table[19] */
172    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
173    rsaz_1024_scatter5_avx2(table_s, result, 19);
174
175    /* table[11] */
176    rsaz_1024_gather5_avx2(result, table_s, 10);
177    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
178    rsaz_1024_scatter5_avx2(table_s, result, 11);
179    /* table[22] */
180    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
181    rsaz_1024_scatter5_avx2(table_s, result, 22);
182    /* table[23] */
183    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
184    rsaz_1024_scatter5_avx2(table_s, result, 23);
185
186    /* table[13] */
187    rsaz_1024_gather5_avx2(result, table_s, 12);
188    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
189    rsaz_1024_scatter5_avx2(table_s, result, 13);
190    /* table[26] */
191    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
192    rsaz_1024_scatter5_avx2(table_s, result, 26);
193    /* table[27] */
194    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
195    rsaz_1024_scatter5_avx2(table_s, result, 27);
196
197    /* table[15] */
198    rsaz_1024_gather5_avx2(result, table_s, 14);
199    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
200    rsaz_1024_scatter5_avx2(table_s, result, 15);
201    /* table[30] */
202    rsaz_1024_sqr_avx2(result, result, m, k0, 1);
203    rsaz_1024_scatter5_avx2(table_s, result, 30);
204    /* table[31] */
205    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
206    rsaz_1024_scatter5_avx2(table_s, result, 31);
207#endif
208
209    /* load first window */
210    p_str = (unsigned char *)exponent;
211    wvalue = p_str[127] >> 3;
212    rsaz_1024_gather5_avx2(result, table_s, wvalue);
213
214    index = 1014;
215
216    while (index > -1) {        /* loop for the remaining 127 windows */
217
218        rsaz_1024_sqr_avx2(result, result, m, k0, 5);
219
220        wvalue = (p_str[(index / 8) + 1] << 8) | p_str[index / 8];
221        wvalue = (wvalue >> (index % 8)) & 31;
222        index -= 5;
223
224        rsaz_1024_gather5_avx2(a_inv, table_s, wvalue); /* borrow a_inv */
225        rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
226    }
227
228    /* square four times */
229    rsaz_1024_sqr_avx2(result, result, m, k0, 4);
230
231    wvalue = p_str[0] & 15;
232
233    rsaz_1024_gather5_avx2(a_inv, table_s, wvalue); /* borrow a_inv */
234    rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
235
236    /* from Montgomery */
237    rsaz_1024_mul_avx2(result, result, one, m, k0);
238
239    rsaz_1024_red2norm_avx2(result_norm, result);
240
241    bn_reduce_once_in_place(result_norm, /*carry=*/0, m_norm, tmp, 16);
242
243    OPENSSL_cleanse(storage, sizeof(storage));
244    OPENSSL_cleanse(tmp, sizeof(tmp));
245}
246
247/*
248 * See crypto/bn/rsaz-x86_64.pl for further details.
249 */
250void rsaz_512_mul(void *ret, const void *a, const void *b, const void *n,
251                  BN_ULONG k);
252void rsaz_512_mul_scatter4(void *ret, const void *a, const void *n,
253                           BN_ULONG k, const void *tbl, unsigned int power);
254void rsaz_512_mul_gather4(void *ret, const void *a, const void *tbl,
255                          const void *n, BN_ULONG k, unsigned int power);
256void rsaz_512_mul_by_one(void *ret, const void *a, const void *n, BN_ULONG k);
257void rsaz_512_sqr(void *ret, const void *a, const void *n, BN_ULONG k,
258                  int cnt);
259void rsaz_512_scatter4(void *tbl, const BN_ULONG *val, int power);
260void rsaz_512_gather4(BN_ULONG *val, const void *tbl, int power);
261
262void RSAZ_512_mod_exp(BN_ULONG result[8],
263                      const BN_ULONG base[8], const BN_ULONG exponent[8],
264                      const BN_ULONG m[8], BN_ULONG k0, const BN_ULONG RR[8])
265{
266    unsigned char storage[16 * 8 * 8 + 64 * 2 + 64]; /* 1.2KB */
267    unsigned char *table = storage + (64 - ((size_t)storage % 64));
268    BN_ULONG *a_inv = (BN_ULONG *)(table + 16 * 8 * 8);
269    BN_ULONG *temp = (BN_ULONG *)(table + 16 * 8 * 8 + 8 * 8);
270    unsigned char *p_str = (unsigned char *)exponent;
271    int index;
272    unsigned int wvalue;
273    BN_ULONG tmp[8];
274
275    /* table[0] = 1_inv */
276    temp[0] = 0 - m[0];
277    temp[1] = ~m[1];
278    temp[2] = ~m[2];
279    temp[3] = ~m[3];
280    temp[4] = ~m[4];
281    temp[5] = ~m[5];
282    temp[6] = ~m[6];
283    temp[7] = ~m[7];
284    rsaz_512_scatter4(table, temp, 0);
285
286    /* table [1] = a_inv^1 */
287    rsaz_512_mul(a_inv, base, RR, m, k0);
288    rsaz_512_scatter4(table, a_inv, 1);
289
290    /* table [2] = a_inv^2 */
291    rsaz_512_sqr(temp, a_inv, m, k0, 1);
292    rsaz_512_scatter4(table, temp, 2);
293
294    for (index = 3; index < 16; index++)
295        rsaz_512_mul_scatter4(temp, a_inv, m, k0, table, index);
296
297    /* load first window */
298    wvalue = p_str[63];
299
300    rsaz_512_gather4(temp, table, wvalue >> 4);
301    rsaz_512_sqr(temp, temp, m, k0, 4);
302    rsaz_512_mul_gather4(temp, temp, table, m, k0, wvalue & 0xf);
303
304    for (index = 62; index >= 0; index--) {
305        wvalue = p_str[index];
306
307        rsaz_512_sqr(temp, temp, m, k0, 4);
308        rsaz_512_mul_gather4(temp, temp, table, m, k0, wvalue >> 4);
309
310        rsaz_512_sqr(temp, temp, m, k0, 4);
311        rsaz_512_mul_gather4(temp, temp, table, m, k0, wvalue & 0x0f);
312    }
313
314    /* from Montgomery */
315    rsaz_512_mul_by_one(result, temp, m, k0);
316
317    bn_reduce_once_in_place(result, /*carry=*/0, m, tmp, 8);
318
319    OPENSSL_cleanse(storage, sizeof(storage));
320    OPENSSL_cleanse(tmp, sizeof(tmp));
321}
322
323#endif
324