/src/openssl30/crypto/bn/rsaz_exp_x2.c
Line  | Count  | Source (jump to first uncovered line)  | 
1  |  | /*  | 
2  |  |  * Copyright 2020-2025 The OpenSSL Project Authors. All Rights Reserved.  | 
3  |  |  * Copyright (c) 2020, 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  |  |  *  | 
11  |  |  * Originally written by Ilya Albrekht, Sergey Kirillov and Andrey Matyukov  | 
12  |  |  * Intel Corporation  | 
13  |  |  *  | 
14  |  |  */  | 
15  |  |  | 
16  |  | #include <openssl/opensslconf.h>  | 
17  |  | #include <openssl/crypto.h>  | 
18  |  | #include "rsaz_exp.h"  | 
19  |  |  | 
20  |  | #ifndef RSAZ_ENABLED  | 
21  |  | NON_EMPTY_TRANSLATION_UNIT  | 
22  |  | #else  | 
23  |  | # include <assert.h>  | 
24  |  | # include <string.h>  | 
25  |  |  | 
26  |  | # if defined(__GNUC__)  | 
27  | 0  | #  define ALIGN64 __attribute__((aligned(64)))  | 
28  |  | # elif defined(_MSC_VER)  | 
29  |  | #  define ALIGN64 __declspec(align(64))  | 
30  |  | # else  | 
31  |  | #  define ALIGN64  | 
32  |  | # endif  | 
33  |  |  | 
34  |  | # define ALIGN_OF(ptr, boundary) \  | 
35  | 0  |     ((unsigned char *)(ptr) + (boundary - (((size_t)(ptr)) & (boundary - 1))))  | 
36  |  |  | 
37  |  | /* Internal radix */  | 
38  | 0  | # define DIGIT_SIZE (52)  | 
39  |  | /* 52-bit mask */  | 
40  | 0  | # define DIGIT_MASK ((uint64_t)0xFFFFFFFFFFFFF)  | 
41  |  |  | 
42  | 0  | # define BITS2WORD8_SIZE(x)  (((x) + 7) >> 3)  | 
43  | 0  | # define BITS2WORD64_SIZE(x) (((x) + 63) >> 6)  | 
44  |  |  | 
45  |  | static ossl_inline uint64_t get_digit52(const uint8_t *in, int in_len);  | 
46  |  | static ossl_inline void put_digit52(uint8_t *out, int out_len, uint64_t digit);  | 
47  |  | static void to_words52(BN_ULONG *out, int out_len, const BN_ULONG *in,  | 
48  |  |                        int in_bitsize);  | 
49  |  | static void from_words52(BN_ULONG *bn_out, int out_bitsize, const BN_ULONG *in);  | 
50  |  | static ossl_inline void set_bit(BN_ULONG *a, int idx);  | 
51  |  |  | 
52  |  | /* Number of |digit_size|-bit digits in |bitsize|-bit value */  | 
53  |  | static ossl_inline int number_of_digits(int bitsize, int digit_size)  | 
54  | 0  | { | 
55  | 0  |     return (bitsize + digit_size - 1) / digit_size;  | 
56  | 0  | }  | 
57  |  |  | 
58  |  | typedef void (*AMM52)(BN_ULONG *res, const BN_ULONG *base,  | 
59  |  |                       const BN_ULONG *exp, const BN_ULONG *m, BN_ULONG k0);  | 
60  |  | typedef void (*EXP52_x2)(BN_ULONG *res, const BN_ULONG *base,  | 
61  |  |                          const BN_ULONG *exp[2], const BN_ULONG *m,  | 
62  |  |                          const BN_ULONG *rr, const BN_ULONG k0[2]);  | 
63  |  |  | 
64  |  | /*  | 
65  |  |  * For details of the methods declared below please refer to  | 
66  |  |  *    crypto/bn/asm/rsaz-avx512.pl  | 
67  |  |  *  | 
68  |  |  * Naming notes:  | 
69  |  |  *  amm = Almost Montgomery Multiplication  | 
70  |  |  *  ams = Almost Montgomery Squaring  | 
71  |  |  *  52x20 - data represented as array of 20 digits in 52-bit radix  | 
72  |  |  *  _x1_/_x2_ - 1 or 2 independent inputs/outputs  | 
73  |  |  *  _256 suffix - uses 256-bit (AVX512VL) registers  | 
74  |  |  */  | 
75  |  |  | 
76  |  | /*AMM = Almost Montgomery Multiplication. */  | 
77  |  | void ossl_rsaz_amm52x20_x1_256(BN_ULONG *res, const BN_ULONG *base,  | 
78  |  |                                const BN_ULONG *exp, const BN_ULONG *m,  | 
79  |  |                                BN_ULONG k0);  | 
80  |  | static void RSAZ_exp52x20_x2_256(BN_ULONG *res, const BN_ULONG *base,  | 
81  |  |                                  const BN_ULONG *exp[2], const BN_ULONG *m,  | 
82  |  |                                  const BN_ULONG *rr, const BN_ULONG k0[2]);  | 
83  |  | void ossl_rsaz_amm52x20_x2_256(BN_ULONG *out, const BN_ULONG *a,  | 
84  |  |                                const BN_ULONG *b, const BN_ULONG *m,  | 
85  |  |                                const BN_ULONG k0[2]);  | 
86  |  | void ossl_extract_multiplier_2x20_win5(BN_ULONG *red_Y,  | 
87  |  |                                        const BN_ULONG *red_table,  | 
88  |  |                                        int red_table_idx, int tbl_idx);  | 
89  |  |  | 
90  |  | /*  | 
91  |  |  * Dual Montgomery modular exponentiation using prime moduli of the  | 
92  |  |  * same bit size, optimized with AVX512 ISA.  | 
93  |  |  *  | 
94  |  |  * Input and output parameters for each exponentiation are independent and  | 
95  |  |  * denoted here by index |i|, i = 1..2.  | 
96  |  |  *  | 
97  |  |  * Input and output are all in regular 2^64 radix.  | 
98  |  |  *  | 
99  |  |  * Each moduli shall be |factor_size| bit size.  | 
100  |  |  *  | 
101  |  |  * NOTE: currently only 2x1024 case is supported.  | 
102  |  |  *  | 
103  |  |  *  [out] res|i|      - result of modular exponentiation: array of qword values  | 
104  |  |  *                      in regular (2^64) radix. Size of array shall be enough  | 
105  |  |  *                      to hold |factor_size| bits.  | 
106  |  |  *  [in]  base|i|     - base  | 
107  |  |  *  [in]  exp|i|      - exponent  | 
108  |  |  *  [in]  m|i|        - moduli  | 
109  |  |  *  [in]  rr|i|       - Montgomery parameter RR = R^2 mod m|i|  | 
110  |  |  *  [in]  k0_|i|      - Montgomery parameter k0 = -1/m|i| mod 2^64  | 
111  |  |  *  [in]  factor_size - moduli bit size  | 
112  |  |  *  | 
113  |  |  * \return 0 in case of failure,  | 
114  |  |  *         1 in case of success.  | 
115  |  |  */  | 
116  |  | int ossl_rsaz_mod_exp_avx512_x2(BN_ULONG *res1,  | 
117  |  |                                 const BN_ULONG *base1,  | 
118  |  |                                 const BN_ULONG *exp1,  | 
119  |  |                                 const BN_ULONG *m1,  | 
120  |  |                                 const BN_ULONG *rr1,  | 
121  |  |                                 BN_ULONG k0_1,  | 
122  |  |                                 BN_ULONG *res2,  | 
123  |  |                                 const BN_ULONG *base2,  | 
124  |  |                                 const BN_ULONG *exp2,  | 
125  |  |                                 const BN_ULONG *m2,  | 
126  |  |                                 const BN_ULONG *rr2,  | 
127  |  |                                 BN_ULONG k0_2,  | 
128  |  |                                 int factor_size)  | 
129  | 0  | { | 
130  | 0  |     int ret = 0;  | 
131  |  |  | 
132  |  |     /*  | 
133  |  |      * Number of word-size (BN_ULONG) digits to store exponent in redundant  | 
134  |  |      * representation.  | 
135  |  |      */  | 
136  | 0  |     int exp_digits = number_of_digits(factor_size + 2, DIGIT_SIZE);  | 
137  | 0  |     int coeff_pow = 4 * (DIGIT_SIZE * exp_digits - factor_size);  | 
138  | 0  |     BN_ULONG *base1_red, *m1_red, *rr1_red;  | 
139  | 0  |     BN_ULONG *base2_red, *m2_red, *rr2_red;  | 
140  | 0  |     BN_ULONG *coeff_red;  | 
141  | 0  |     BN_ULONG *storage = NULL;  | 
142  | 0  |     BN_ULONG *storage_aligned = NULL;  | 
143  | 0  |     BN_ULONG storage_len_bytes = 7 * exp_digits * sizeof(BN_ULONG);  | 
144  |  |  | 
145  |  |     /* AMM = Almost Montgomery Multiplication */  | 
146  | 0  |     AMM52 amm = NULL;  | 
147  |  |     /* Dual (2-exps in parallel) exponentiation */  | 
148  | 0  |     EXP52_x2 exp_x2 = NULL;  | 
149  |  | 
  | 
150  | 0  |     const BN_ULONG *exp[2] = {0}; | 
151  | 0  |     BN_ULONG k0[2] = {0}; | 
152  |  |  | 
153  |  |     /* Only 1024-bit factor size is supported now */  | 
154  | 0  |     switch (factor_size) { | 
155  | 0  |     case 1024:  | 
156  | 0  |         amm = ossl_rsaz_amm52x20_x1_256;  | 
157  | 0  |         exp_x2 = RSAZ_exp52x20_x2_256;  | 
158  | 0  |         break;  | 
159  | 0  |     default:  | 
160  | 0  |         goto err;  | 
161  | 0  |     }  | 
162  |  |  | 
163  | 0  |     storage = (BN_ULONG *)OPENSSL_malloc(storage_len_bytes + 64);  | 
164  | 0  |     if (storage == NULL)  | 
165  | 0  |         goto err;  | 
166  | 0  |     storage_aligned = (BN_ULONG *)ALIGN_OF(storage, 64);  | 
167  |  |  | 
168  |  |     /* Memory layout for red(undant) representations */  | 
169  | 0  |     base1_red = storage_aligned;  | 
170  | 0  |     base2_red = storage_aligned + 1 * exp_digits;  | 
171  | 0  |     m1_red    = storage_aligned + 2 * exp_digits;  | 
172  | 0  |     m2_red    = storage_aligned + 3 * exp_digits;  | 
173  | 0  |     rr1_red   = storage_aligned + 4 * exp_digits;  | 
174  | 0  |     rr2_red   = storage_aligned + 5 * exp_digits;  | 
175  | 0  |     coeff_red = storage_aligned + 6 * exp_digits;  | 
176  |  |  | 
177  |  |     /* Convert base_i, m_i, rr_i, from regular to 52-bit radix */  | 
178  | 0  |     to_words52(base1_red, exp_digits, base1, factor_size);  | 
179  | 0  |     to_words52(base2_red, exp_digits, base2, factor_size);  | 
180  | 0  |     to_words52(m1_red, exp_digits, m1, factor_size);  | 
181  | 0  |     to_words52(m2_red, exp_digits, m2, factor_size);  | 
182  | 0  |     to_words52(rr1_red, exp_digits, rr1, factor_size);  | 
183  | 0  |     to_words52(rr2_red, exp_digits, rr2, factor_size);  | 
184  |  |  | 
185  |  |     /*  | 
186  |  |      * Compute target domain Montgomery converters RR' for each modulus  | 
187  |  |      * based on precomputed original domain's RR.  | 
188  |  |      *  | 
189  |  |      * RR -> RR' transformation steps:  | 
190  |  |      *  (1) coeff = 2^k  | 
191  |  |      *  (2) t = AMM(RR,RR) = RR^2 / R' mod m  | 
192  |  |      *  (3) RR' = AMM(t, coeff) = RR^2 * 2^k / R'^2 mod m  | 
193  |  |      * where  | 
194  |  |      *  k = 4 * (52 * digits52 - modlen)  | 
195  |  |      *  R  = 2^(64 * ceil(modlen/64)) mod m  | 
196  |  |      *  RR = R^2 mod M  | 
197  |  |      *  R' = 2^(52 * ceil(modlen/52)) mod m  | 
198  |  |      *  | 
199  |  |      *  modlen = 1024: k = 64, RR = 2^2048 mod m, RR' = 2^2080 mod m  | 
200  |  |      */  | 
201  | 0  |     memset(coeff_red, 0, exp_digits * sizeof(BN_ULONG));  | 
202  |  |     /* (1) in reduced domain representation */  | 
203  | 0  |     set_bit(coeff_red, 64 * (int)(coeff_pow / 52) + coeff_pow % 52);  | 
204  |  | 
  | 
205  | 0  |     amm(rr1_red, rr1_red, rr1_red, m1_red, k0_1);     /* (2) for m1 */  | 
206  | 0  |     amm(rr1_red, rr1_red, coeff_red, m1_red, k0_1);   /* (3) for m1 */  | 
207  |  | 
  | 
208  | 0  |     amm(rr2_red, rr2_red, rr2_red, m2_red, k0_2);     /* (2) for m2 */  | 
209  | 0  |     amm(rr2_red, rr2_red, coeff_red, m2_red, k0_2);   /* (3) for m2 */  | 
210  |  | 
  | 
211  | 0  |     exp[0] = exp1;  | 
212  | 0  |     exp[1] = exp2;  | 
213  |  | 
  | 
214  | 0  |     k0[0] = k0_1;  | 
215  | 0  |     k0[1] = k0_2;  | 
216  |  | 
  | 
217  | 0  |     exp_x2(rr1_red, base1_red, exp, m1_red, rr1_red, k0);  | 
218  |  |  | 
219  |  |     /* Convert rr_i back to regular radix */  | 
220  | 0  |     from_words52(res1, factor_size, rr1_red);  | 
221  | 0  |     from_words52(res2, factor_size, rr2_red);  | 
222  |  |  | 
223  |  |     /* bn_reduce_once_in_place expects number of BN_ULONG, not bit size */  | 
224  | 0  |     factor_size /= sizeof(BN_ULONG) * 8;  | 
225  |  | 
  | 
226  | 0  |     bn_reduce_once_in_place(res1, /*carry=*/0, m1, storage, factor_size);  | 
227  | 0  |     bn_reduce_once_in_place(res2, /*carry=*/0, m2, storage, factor_size);  | 
228  |  | 
  | 
229  | 0  |     ret = 1;  | 
230  | 0  | err:  | 
231  | 0  |     if (storage != NULL) { | 
232  | 0  |         OPENSSL_cleanse(storage, storage_len_bytes);  | 
233  | 0  |         OPENSSL_free(storage);  | 
234  | 0  |     }  | 
235  | 0  |     return ret;  | 
236  | 0  | }  | 
237  |  |  | 
238  |  | /*  | 
239  |  |  * Dual 1024-bit w-ary modular exponentiation using prime moduli of the same  | 
240  |  |  * bit size using Almost Montgomery Multiplication, optimized with AVX512_IFMA  | 
241  |  |  * ISA.  | 
242  |  |  *  | 
243  |  |  * The parameter w (window size) = 5.  | 
244  |  |  *  | 
245  |  |  *  [out] res      - result of modular exponentiation: 2x20 qword  | 
246  |  |  *                   values in 2^52 radix.  | 
247  |  |  *  [in]  base     - base (2x20 qword values in 2^52 radix)  | 
248  |  |  *  [in]  exp      - array of 2 pointers to 16 qword values in 2^64 radix.  | 
249  |  |  *                   Exponent is not converted to redundant representation.  | 
250  |  |  *  [in]  m        - moduli (2x20 qword values in 2^52 radix)  | 
251  |  |  *  [in]  rr       - Montgomery parameter for 2 moduli: RR = 2^2080 mod m.  | 
252  |  |  *                   (2x20 qword values in 2^52 radix)  | 
253  |  |  *  [in]  k0       - Montgomery parameter for 2 moduli: k0 = -1/m mod 2^64  | 
254  |  |  *  | 
255  |  |  * \return (void).  | 
256  |  |  */  | 
257  |  | static void RSAZ_exp52x20_x2_256(BN_ULONG *out,          /* [2][20] */  | 
258  |  |                                  const BN_ULONG *base,   /* [2][20] */  | 
259  |  |                                  const BN_ULONG *exp[2], /* 2x16    */  | 
260  |  |                                  const BN_ULONG *m,      /* [2][20] */  | 
261  |  |                                  const BN_ULONG *rr,     /* [2][20] */  | 
262  |  |                                  const BN_ULONG k0[2])  | 
263  | 0  | { | 
264  | 0  | # define BITSIZE_MODULUS (1024)  | 
265  | 0  | # define EXP_WIN_SIZE (5)  | 
266  | 0  | # define EXP_WIN_MASK ((1U << EXP_WIN_SIZE) - 1)  | 
267  |  | /*  | 
268  |  |  * Number of digits (64-bit words) in redundant representation to handle  | 
269  |  |  * modulus bits  | 
270  |  |  */  | 
271  | 0  | # define RED_DIGITS (20)  | 
272  | 0  | # define EXP_DIGITS (16)  | 
273  | 0  | # define DAMM ossl_rsaz_amm52x20_x2_256  | 
274  |  | /*  | 
275  |  |  * Squaring is done using multiplication now. That can be a subject of  | 
276  |  |  * optimization in future.  | 
277  |  |  */  | 
278  | 0  | # define DAMS(r,a,m,k0) \  | 
279  | 0  |               ossl_rsaz_amm52x20_x2_256((r),(a),(a),(m),(k0))  | 
280  |  |  | 
281  |  |     /* Allocate stack for red(undant) result Y and multiplier X */  | 
282  | 0  |     ALIGN64 BN_ULONG red_Y[2][RED_DIGITS];  | 
283  | 0  |     ALIGN64 BN_ULONG red_X[2][RED_DIGITS];  | 
284  |  |  | 
285  |  |     /* Allocate expanded exponent */  | 
286  | 0  |     ALIGN64 BN_ULONG expz[2][EXP_DIGITS + 1];  | 
287  |  |  | 
288  |  |     /* Pre-computed table of base powers */  | 
289  | 0  |     ALIGN64 BN_ULONG red_table[1U << EXP_WIN_SIZE][2][RED_DIGITS];  | 
290  |  | 
  | 
291  | 0  |     int idx;  | 
292  |  | 
  | 
293  | 0  |     memset(red_Y, 0, sizeof(red_Y));  | 
294  | 0  |     memset(red_table, 0, sizeof(red_table));  | 
295  | 0  |     memset(red_X, 0, sizeof(red_X));  | 
296  |  |  | 
297  |  |     /*  | 
298  |  |      * Compute table of powers base^i, i = 0, ..., (2^EXP_WIN_SIZE) - 1  | 
299  |  |      *   table[0] = mont(x^0) = mont(1)  | 
300  |  |      *   table[1] = mont(x^1) = mont(x)  | 
301  |  |      */  | 
302  | 0  |     red_X[0][0] = 1;  | 
303  | 0  |     red_X[1][0] = 1;  | 
304  | 0  |     DAMM(red_table[0][0], (const BN_ULONG*)red_X, rr, m, k0);  | 
305  | 0  |     DAMM(red_table[1][0], base,  rr, m, k0);  | 
306  |  | 
  | 
307  | 0  |     for (idx = 1; idx < (int)((1U << EXP_WIN_SIZE) / 2); idx++) { | 
308  | 0  |         DAMS(red_table[2 * idx + 0][0], red_table[1 * idx][0], m, k0);  | 
309  | 0  |         DAMM(red_table[2 * idx + 1][0], red_table[2 * idx][0], red_table[1][0], m, k0);  | 
310  | 0  |     }  | 
311  |  |  | 
312  |  |     /* Copy and expand exponents */  | 
313  | 0  |     memcpy(expz[0], exp[0], EXP_DIGITS * sizeof(BN_ULONG));  | 
314  | 0  |     expz[0][EXP_DIGITS] = 0;  | 
315  | 0  |     memcpy(expz[1], exp[1], EXP_DIGITS * sizeof(BN_ULONG));  | 
316  | 0  |     expz[1][EXP_DIGITS] = 0;  | 
317  |  |  | 
318  |  |     /* Exponentiation */  | 
319  | 0  |     { | 
320  | 0  |         const int rem = BITSIZE_MODULUS % EXP_WIN_SIZE;  | 
321  | 0  |         BN_ULONG table_idx_mask = EXP_WIN_MASK;  | 
322  |  | 
  | 
323  | 0  |         int exp_bit_no = BITSIZE_MODULUS - rem;  | 
324  | 0  |         int exp_chunk_no = exp_bit_no / 64;  | 
325  | 0  |         int exp_chunk_shift = exp_bit_no % 64;  | 
326  |  | 
  | 
327  | 0  |         BN_ULONG red_table_idx_0, red_table_idx_1;  | 
328  |  |  | 
329  |  |         /*  | 
330  |  |          * If rem == 0, then  | 
331  |  |          *      exp_bit_no = modulus_bitsize - exp_win_size  | 
332  |  |          * However, this isn't possible because rem is { 1024, 1536, 2048 } % 5 | 
333  |  |          * which is { 4, 1, 3 } respectively. | 
334  |  |          *  | 
335  |  |          * If this assertion ever fails the fix above is easy.  | 
336  |  |          */  | 
337  | 0  |         OPENSSL_assert(rem != 0);  | 
338  |  |  | 
339  |  |         /* Process 1-st exp window - just init result */  | 
340  | 0  |         red_table_idx_0 = expz[0][exp_chunk_no];  | 
341  | 0  |         red_table_idx_1 = expz[1][exp_chunk_no];  | 
342  |  |         /*  | 
343  |  |          * The function operates with fixed moduli sizes divisible by 64,  | 
344  |  |          * thus table index here is always in supported range [0, EXP_WIN_SIZE).  | 
345  |  |          */  | 
346  | 0  |         red_table_idx_0 >>= exp_chunk_shift;  | 
347  | 0  |         red_table_idx_1 >>= exp_chunk_shift;  | 
348  |  | 
  | 
349  | 0  |         ossl_extract_multiplier_2x20_win5(red_Y[0], (const BN_ULONG*)red_table,  | 
350  | 0  |                                           (int)red_table_idx_0, 0);  | 
351  | 0  |         ossl_extract_multiplier_2x20_win5(red_Y[1], (const BN_ULONG*)red_table,  | 
352  | 0  |                                           (int)red_table_idx_1, 1);  | 
353  |  |  | 
354  |  |         /* Process other exp windows */  | 
355  | 0  |         for (exp_bit_no -= EXP_WIN_SIZE; exp_bit_no >= 0; exp_bit_no -= EXP_WIN_SIZE) { | 
356  |  |             /* Extract pre-computed multiplier from the table */  | 
357  | 0  |             { | 
358  | 0  |                 BN_ULONG T;  | 
359  |  | 
  | 
360  | 0  |                 exp_chunk_no = exp_bit_no / 64;  | 
361  | 0  |                 exp_chunk_shift = exp_bit_no % 64;  | 
362  | 0  |                 { | 
363  | 0  |                     red_table_idx_0 = expz[0][exp_chunk_no];  | 
364  | 0  |                     T = expz[0][exp_chunk_no + 1];  | 
365  |  | 
  | 
366  | 0  |                     red_table_idx_0 >>= exp_chunk_shift;  | 
367  |  |                     /*  | 
368  |  |                      * Get additional bits from then next quadword  | 
369  |  |                      * when 64-bit boundaries are crossed.  | 
370  |  |                      */  | 
371  | 0  |                     if (exp_chunk_shift > 64 - EXP_WIN_SIZE) { | 
372  | 0  |                         T <<= (64 - exp_chunk_shift);  | 
373  | 0  |                         red_table_idx_0 ^= T;  | 
374  | 0  |                     }  | 
375  | 0  |                     red_table_idx_0 &= table_idx_mask;  | 
376  |  | 
  | 
377  | 0  |                     ossl_extract_multiplier_2x20_win5(red_X[0],  | 
378  | 0  |                                                       (const BN_ULONG*)red_table,  | 
379  | 0  |                                                       (int)red_table_idx_0, 0);  | 
380  | 0  |                 }  | 
381  | 0  |                 { | 
382  | 0  |                     red_table_idx_1 = expz[1][exp_chunk_no];  | 
383  | 0  |                     T = expz[1][exp_chunk_no + 1];  | 
384  |  | 
  | 
385  | 0  |                     red_table_idx_1 >>= exp_chunk_shift;  | 
386  |  |                     /*  | 
387  |  |                      * Get additional bits from then next quadword  | 
388  |  |                      * when 64-bit boundaries are crossed.  | 
389  |  |                      */  | 
390  | 0  |                     if (exp_chunk_shift > 64 - EXP_WIN_SIZE) { | 
391  | 0  |                         T <<= (64 - exp_chunk_shift);  | 
392  | 0  |                         red_table_idx_1 ^= T;  | 
393  | 0  |                     }  | 
394  | 0  |                     red_table_idx_1 &= table_idx_mask;  | 
395  |  | 
  | 
396  | 0  |                     ossl_extract_multiplier_2x20_win5(red_X[1],  | 
397  | 0  |                                                       (const BN_ULONG*)red_table,  | 
398  | 0  |                                                       (int)red_table_idx_1, 1);  | 
399  | 0  |                 }  | 
400  | 0  |             }  | 
401  |  |  | 
402  |  |             /* Series of squaring */  | 
403  | 0  |             DAMS((BN_ULONG*)red_Y, (const BN_ULONG*)red_Y, m, k0);  | 
404  | 0  |             DAMS((BN_ULONG*)red_Y, (const BN_ULONG*)red_Y, m, k0);  | 
405  | 0  |             DAMS((BN_ULONG*)red_Y, (const BN_ULONG*)red_Y, m, k0);  | 
406  | 0  |             DAMS((BN_ULONG*)red_Y, (const BN_ULONG*)red_Y, m, k0);  | 
407  | 0  |             DAMS((BN_ULONG*)red_Y, (const BN_ULONG*)red_Y, m, k0);  | 
408  |  | 
  | 
409  | 0  |             DAMM((BN_ULONG*)red_Y, (const BN_ULONG*)red_Y, (const BN_ULONG*)red_X, m, k0);  | 
410  | 0  |         }  | 
411  | 0  |     }  | 
412  |  |  | 
413  |  |     /*  | 
414  |  |      *  | 
415  |  |      * NB: After the last AMM of exponentiation in Montgomery domain, the result  | 
416  |  |      * may be 1025-bit, but the conversion out of Montgomery domain performs an  | 
417  |  |      * AMM(x,1) which guarantees that the final result is less than |m|, so no  | 
418  |  |      * conditional subtraction is needed here. See "Efficient Software  | 
419  |  |      * Implementations of Modular Exponentiation" (by Shay Gueron) paper for details.  | 
420  |  |      */  | 
421  |  |  | 
422  |  |     /* Convert result back in regular 2^52 domain */  | 
423  | 0  |     memset(red_X, 0, sizeof(red_X));  | 
424  | 0  |     red_X[0][0] = 1;  | 
425  | 0  |     red_X[1][0] = 1;  | 
426  | 0  |     DAMM(out, (const BN_ULONG*)red_Y, (const BN_ULONG*)red_X, m, k0);  | 
427  |  |  | 
428  |  |     /* Clear exponents */  | 
429  | 0  |     OPENSSL_cleanse(expz, sizeof(expz));  | 
430  | 0  |     OPENSSL_cleanse(red_Y, sizeof(red_Y));  | 
431  |  | 
  | 
432  | 0  | # undef DAMS  | 
433  | 0  | # undef DAMM  | 
434  | 0  | # undef EXP_DIGITS  | 
435  | 0  | # undef RED_DIGITS  | 
436  | 0  | # undef EXP_WIN_MASK  | 
437  | 0  | # undef EXP_WIN_SIZE  | 
438  | 0  | # undef BITSIZE_MODULUS  | 
439  | 0  | }  | 
440  |  |  | 
441  |  | static ossl_inline uint64_t get_digit52(const uint8_t *in, int in_len)  | 
442  | 0  | { | 
443  | 0  |     uint64_t digit = 0;  | 
444  |  | 
  | 
445  | 0  |     assert(in != NULL);  | 
446  |  |  | 
447  | 0  |     for (; in_len > 0; in_len--) { | 
448  | 0  |         digit <<= 8;  | 
449  | 0  |         digit += (uint64_t)(in[in_len - 1]);  | 
450  | 0  |     }  | 
451  | 0  |     return digit;  | 
452  | 0  | }  | 
453  |  |  | 
454  |  | /*  | 
455  |  |  * Convert array of words in regular (base=2^64) representation to array of  | 
456  |  |  * words in redundant (base=2^52) one.  | 
457  |  |  */  | 
458  |  | static void to_words52(BN_ULONG *out, int out_len,  | 
459  |  |                        const BN_ULONG *in, int in_bitsize)  | 
460  | 0  | { | 
461  | 0  |     uint8_t *in_str = NULL;  | 
462  |  | 
  | 
463  | 0  |     assert(out != NULL);  | 
464  | 0  |     assert(in != NULL);  | 
465  |  |     /* Check destination buffer capacity */  | 
466  | 0  |     assert(out_len >= number_of_digits(in_bitsize, DIGIT_SIZE));  | 
467  |  |  | 
468  | 0  |     in_str = (uint8_t *)in;  | 
469  |  | 
  | 
470  | 0  |     for (; in_bitsize >= (2 * DIGIT_SIZE); in_bitsize -= (2 * DIGIT_SIZE), out += 2) { | 
471  | 0  |         uint64_t digit;  | 
472  |  | 
  | 
473  | 0  |         memcpy(&digit, in_str, sizeof(digit));  | 
474  | 0  |         out[0] = digit & DIGIT_MASK;  | 
475  | 0  |         in_str += 6;  | 
476  | 0  |         memcpy(&digit, in_str, sizeof(digit));  | 
477  | 0  |         out[1] = (digit >> 4) & DIGIT_MASK;  | 
478  | 0  |         in_str += 7;  | 
479  | 0  |         out_len -= 2;  | 
480  | 0  |     }  | 
481  |  | 
  | 
482  | 0  |     if (in_bitsize > DIGIT_SIZE) { | 
483  | 0  |         uint64_t digit = get_digit52(in_str, 7);  | 
484  |  | 
  | 
485  | 0  |         out[0] = digit & DIGIT_MASK;  | 
486  | 0  |         in_str += 6;  | 
487  | 0  |         in_bitsize -= DIGIT_SIZE;  | 
488  | 0  |         digit = get_digit52(in_str, BITS2WORD8_SIZE(in_bitsize));  | 
489  | 0  |         out[1] = digit >> 4;  | 
490  | 0  |         out += 2;  | 
491  | 0  |         out_len -= 2;  | 
492  | 0  |     } else if (in_bitsize > 0) { | 
493  | 0  |         out[0] = get_digit52(in_str, BITS2WORD8_SIZE(in_bitsize));  | 
494  | 0  |         out++;  | 
495  | 0  |         out_len--;  | 
496  | 0  |     }  | 
497  |  | 
  | 
498  | 0  |     memset(out, 0, out_len * sizeof(BN_ULONG));  | 
499  | 0  | }  | 
500  |  |  | 
501  |  | static ossl_inline void put_digit52(uint8_t *pStr, int strLen, uint64_t digit)  | 
502  | 0  | { | 
503  | 0  |     assert(pStr != NULL);  | 
504  |  |  | 
505  | 0  |     for (; strLen > 0; strLen--) { | 
506  | 0  |         *pStr++ = (uint8_t)(digit & 0xFF);  | 
507  | 0  |         digit >>= 8;  | 
508  | 0  |     }  | 
509  | 0  | }  | 
510  |  |  | 
511  |  | /*  | 
512  |  |  * Convert array of words in redundant (base=2^52) representation to array of  | 
513  |  |  * words in regular (base=2^64) one.  | 
514  |  |  */  | 
515  |  | static void from_words52(BN_ULONG *out, int out_bitsize, const BN_ULONG *in)  | 
516  | 0  | { | 
517  | 0  |     int i;  | 
518  | 0  |     int out_len = BITS2WORD64_SIZE(out_bitsize);  | 
519  |  | 
  | 
520  | 0  |     assert(out != NULL);  | 
521  | 0  |     assert(in != NULL);  | 
522  |  |  | 
523  | 0  |     for (i = 0; i < out_len; i++)  | 
524  | 0  |         out[i] = 0;  | 
525  |  | 
  | 
526  | 0  |     { | 
527  | 0  |         uint8_t *out_str = (uint8_t *)out;  | 
528  |  | 
  | 
529  | 0  |         for (; out_bitsize >= (2 * DIGIT_SIZE);  | 
530  | 0  |                out_bitsize -= (2 * DIGIT_SIZE), in += 2) { | 
531  | 0  |             uint64_t digit;  | 
532  |  | 
  | 
533  | 0  |             digit = in[0];  | 
534  | 0  |             memcpy(out_str, &digit, sizeof(digit));  | 
535  | 0  |             out_str += 6;  | 
536  | 0  |             digit = digit >> 48 | in[1] << 4;  | 
537  | 0  |             memcpy(out_str, &digit, sizeof(digit));  | 
538  | 0  |             out_str += 7;  | 
539  | 0  |         }  | 
540  |  | 
  | 
541  | 0  |         if (out_bitsize > DIGIT_SIZE) { | 
542  | 0  |             put_digit52(out_str, 7, in[0]);  | 
543  | 0  |             out_str += 6;  | 
544  | 0  |             out_bitsize -= DIGIT_SIZE;  | 
545  | 0  |             put_digit52(out_str, BITS2WORD8_SIZE(out_bitsize),  | 
546  | 0  |                         (in[1] << 4 | in[0] >> 48));  | 
547  | 0  |         } else if (out_bitsize) { | 
548  | 0  |             put_digit52(out_str, BITS2WORD8_SIZE(out_bitsize), in[0]);  | 
549  | 0  |         }  | 
550  | 0  |     }  | 
551  | 0  | }  | 
552  |  |  | 
553  |  | /*  | 
554  |  |  * Set bit at index |idx| in the words array |a|.  | 
555  |  |  * It does not do any boundaries checks, make sure the index is valid before  | 
556  |  |  * calling the function.  | 
557  |  |  */  | 
558  |  | static ossl_inline void set_bit(BN_ULONG *a, int idx)  | 
559  | 0  | { | 
560  | 0  |     assert(a != NULL);  | 
561  |  |  | 
562  | 0  |     { | 
563  | 0  |         int i, j;  | 
564  |  | 
  | 
565  | 0  |         i = idx / BN_BITS2;  | 
566  | 0  |         j = idx % BN_BITS2;  | 
567  | 0  |         a[i] |= (((BN_ULONG)1) << j);  | 
568  | 0  |     }  | 
569  | 0  | }  | 
570  |  |  | 
571  |  | #endif  |