|
1 /* |
|
2 * mpi-priv.h - Private header file for MPI |
|
3 * Arbitrary precision integer arithmetic library |
|
4 * |
|
5 * NOTE WELL: the content of this header file is NOT part of the "public" |
|
6 * API for the MPI library, and may change at any time. |
|
7 * Application programs that use libmpi should NOT include this header file. |
|
8 * |
|
9 * This Source Code Form is subject to the terms of the Mozilla Public |
|
10 * License, v. 2.0. If a copy of the MPL was not distributed with this |
|
11 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ |
|
12 #ifndef _MPI_PRIV_H_ |
|
13 #define _MPI_PRIV_H_ 1 |
|
14 |
|
15 #include "mpi.h" |
|
16 #include <stdlib.h> |
|
17 #include <string.h> |
|
18 #include <ctype.h> |
|
19 |
|
20 #if MP_DEBUG |
|
21 #include <stdio.h> |
|
22 |
|
23 #define DIAG(T,V) {fprintf(stderr,T);mp_print(V,stderr);fputc('\n',stderr);} |
|
24 #else |
|
25 #define DIAG(T,V) |
|
26 #endif |
|
27 |
|
28 /* If we aren't using a wired-in logarithm table, we need to include |
|
29 the math library to get the log() function |
|
30 */ |
|
31 |
|
32 /* {{{ s_logv_2[] - log table for 2 in various bases */ |
|
33 |
|
34 #if MP_LOGTAB |
|
35 /* |
|
36 A table of the logs of 2 for various bases (the 0 and 1 entries of |
|
37 this table are meaningless and should not be referenced). |
|
38 |
|
39 This table is used to compute output lengths for the mp_toradix() |
|
40 function. Since a number n in radix r takes up about log_r(n) |
|
41 digits, we estimate the output size by taking the least integer |
|
42 greater than log_r(n), where: |
|
43 |
|
44 log_r(n) = log_2(n) * log_r(2) |
|
45 |
|
46 This table, therefore, is a table of log_r(2) for 2 <= r <= 36, |
|
47 which are the output bases supported. |
|
48 */ |
|
49 |
|
50 extern const float s_logv_2[]; |
|
51 #define LOG_V_2(R) s_logv_2[(R)] |
|
52 |
|
53 #else |
|
54 |
|
55 /* |
|
56 If MP_LOGTAB is not defined, use the math library to compute the |
|
57 logarithms on the fly. Otherwise, use the table. |
|
58 Pick which works best for your system. |
|
59 */ |
|
60 |
|
61 #include <math.h> |
|
62 #define LOG_V_2(R) (log(2.0)/log(R)) |
|
63 |
|
64 #endif /* if MP_LOGTAB */ |
|
65 |
|
66 /* }}} */ |
|
67 |
|
68 /* {{{ Digit arithmetic macros */ |
|
69 |
|
70 /* |
|
71 When adding and multiplying digits, the results can be larger than |
|
72 can be contained in an mp_digit. Thus, an mp_word is used. These |
|
73 macros mask off the upper and lower digits of the mp_word (the |
|
74 mp_word may be more than 2 mp_digits wide, but we only concern |
|
75 ourselves with the low-order 2 mp_digits) |
|
76 */ |
|
77 |
|
78 #define CARRYOUT(W) (mp_digit)((W)>>DIGIT_BIT) |
|
79 #define ACCUM(W) (mp_digit)(W) |
|
80 |
|
81 #define MP_MIN(a,b) (((a) < (b)) ? (a) : (b)) |
|
82 #define MP_MAX(a,b) (((a) > (b)) ? (a) : (b)) |
|
83 #define MP_HOWMANY(a,b) (((a) + (b) - 1)/(b)) |
|
84 #define MP_ROUNDUP(a,b) (MP_HOWMANY(a,b) * (b)) |
|
85 |
|
86 /* }}} */ |
|
87 |
|
88 /* {{{ Comparison constants */ |
|
89 |
|
90 #define MP_LT -1 |
|
91 #define MP_EQ 0 |
|
92 #define MP_GT 1 |
|
93 |
|
94 /* }}} */ |
|
95 |
|
96 /* {{{ private function declarations */ |
|
97 |
|
98 /* |
|
99 If MP_MACRO is false, these will be defined as actual functions; |
|
100 otherwise, suitable macro definitions will be used. This works |
|
101 around the fact that ANSI C89 doesn't support an 'inline' keyword |
|
102 (although I hear C9x will ... about bloody time). At present, the |
|
103 macro definitions are identical to the function bodies, but they'll |
|
104 expand in place, instead of generating a function call. |
|
105 |
|
106 I chose these particular functions to be made into macros because |
|
107 some profiling showed they are called a lot on a typical workload, |
|
108 and yet they are primarily housekeeping. |
|
109 */ |
|
110 #if MP_MACRO == 0 |
|
111 void s_mp_setz(mp_digit *dp, mp_size count); /* zero digits */ |
|
112 void s_mp_copy(const mp_digit *sp, mp_digit *dp, mp_size count); /* copy */ |
|
113 void *s_mp_alloc(size_t nb, size_t ni); /* general allocator */ |
|
114 void s_mp_free(void *ptr); /* general free function */ |
|
115 extern unsigned long mp_allocs; |
|
116 extern unsigned long mp_frees; |
|
117 extern unsigned long mp_copies; |
|
118 #else |
|
119 |
|
120 /* Even if these are defined as macros, we need to respect the settings |
|
121 of the MP_MEMSET and MP_MEMCPY configuration options... |
|
122 */ |
|
123 #if MP_MEMSET == 0 |
|
124 #define s_mp_setz(dp, count) \ |
|
125 {int ix;for(ix=0;ix<(count);ix++)(dp)[ix]=0;} |
|
126 #else |
|
127 #define s_mp_setz(dp, count) memset(dp, 0, (count) * sizeof(mp_digit)) |
|
128 #endif /* MP_MEMSET */ |
|
129 |
|
130 #if MP_MEMCPY == 0 |
|
131 #define s_mp_copy(sp, dp, count) \ |
|
132 {int ix;for(ix=0;ix<(count);ix++)(dp)[ix]=(sp)[ix];} |
|
133 #else |
|
134 #define s_mp_copy(sp, dp, count) memcpy(dp, sp, (count) * sizeof(mp_digit)) |
|
135 #endif /* MP_MEMCPY */ |
|
136 |
|
137 #define s_mp_alloc(nb, ni) calloc(nb, ni) |
|
138 #define s_mp_free(ptr) {if(ptr) free(ptr);} |
|
139 #endif /* MP_MACRO */ |
|
140 |
|
141 mp_err s_mp_grow(mp_int *mp, mp_size min); /* increase allocated size */ |
|
142 mp_err s_mp_pad(mp_int *mp, mp_size min); /* left pad with zeroes */ |
|
143 |
|
144 #if MP_MACRO == 0 |
|
145 void s_mp_clamp(mp_int *mp); /* clip leading zeroes */ |
|
146 #else |
|
147 #define s_mp_clamp(mp)\ |
|
148 { mp_size used = MP_USED(mp); \ |
|
149 while (used > 1 && DIGIT(mp, used - 1) == 0) --used; \ |
|
150 MP_USED(mp) = used; \ |
|
151 } |
|
152 #endif /* MP_MACRO */ |
|
153 |
|
154 void s_mp_exch(mp_int *a, mp_int *b); /* swap a and b in place */ |
|
155 |
|
156 mp_err s_mp_lshd(mp_int *mp, mp_size p); /* left-shift by p digits */ |
|
157 void s_mp_rshd(mp_int *mp, mp_size p); /* right-shift by p digits */ |
|
158 mp_err s_mp_mul_2d(mp_int *mp, mp_digit d); /* multiply by 2^d in place */ |
|
159 void s_mp_div_2d(mp_int *mp, mp_digit d); /* divide by 2^d in place */ |
|
160 void s_mp_mod_2d(mp_int *mp, mp_digit d); /* modulo 2^d in place */ |
|
161 void s_mp_div_2(mp_int *mp); /* divide by 2 in place */ |
|
162 mp_err s_mp_mul_2(mp_int *mp); /* multiply by 2 in place */ |
|
163 mp_err s_mp_norm(mp_int *a, mp_int *b, mp_digit *pd); |
|
164 /* normalize for division */ |
|
165 mp_err s_mp_add_d(mp_int *mp, mp_digit d); /* unsigned digit addition */ |
|
166 mp_err s_mp_sub_d(mp_int *mp, mp_digit d); /* unsigned digit subtract */ |
|
167 mp_err s_mp_mul_d(mp_int *mp, mp_digit d); /* unsigned digit multiply */ |
|
168 mp_err s_mp_div_d(mp_int *mp, mp_digit d, mp_digit *r); |
|
169 /* unsigned digit divide */ |
|
170 mp_err s_mp_reduce(mp_int *x, const mp_int *m, const mp_int *mu); |
|
171 /* Barrett reduction */ |
|
172 mp_err s_mp_add(mp_int *a, const mp_int *b); /* magnitude addition */ |
|
173 mp_err s_mp_add_3arg(const mp_int *a, const mp_int *b, mp_int *c); |
|
174 mp_err s_mp_sub(mp_int *a, const mp_int *b); /* magnitude subtract */ |
|
175 mp_err s_mp_sub_3arg(const mp_int *a, const mp_int *b, mp_int *c); |
|
176 mp_err s_mp_add_offset(mp_int *a, mp_int *b, mp_size offset); |
|
177 /* a += b * RADIX^offset */ |
|
178 mp_err s_mp_mul(mp_int *a, const mp_int *b); /* magnitude multiply */ |
|
179 #if MP_SQUARE |
|
180 mp_err s_mp_sqr(mp_int *a); /* magnitude square */ |
|
181 #else |
|
182 #define s_mp_sqr(a) s_mp_mul(a, a) |
|
183 #endif |
|
184 mp_err s_mp_div(mp_int *rem, mp_int *div, mp_int *quot); /* magnitude div */ |
|
185 mp_err s_mp_exptmod(const mp_int *a, const mp_int *b, const mp_int *m, mp_int *c); |
|
186 mp_err s_mp_2expt(mp_int *a, mp_digit k); /* a = 2^k */ |
|
187 int s_mp_cmp(const mp_int *a, const mp_int *b); /* magnitude comparison */ |
|
188 int s_mp_cmp_d(const mp_int *a, mp_digit d); /* magnitude digit compare */ |
|
189 int s_mp_ispow2(const mp_int *v); /* is v a power of 2? */ |
|
190 int s_mp_ispow2d(mp_digit d); /* is d a power of 2? */ |
|
191 |
|
192 int s_mp_tovalue(char ch, int r); /* convert ch to value */ |
|
193 char s_mp_todigit(mp_digit val, int r, int low); /* convert val to digit */ |
|
194 int s_mp_outlen(int bits, int r); /* output length in bytes */ |
|
195 mp_digit s_mp_invmod_radix(mp_digit P); /* returns (P ** -1) mod RADIX */ |
|
196 mp_err s_mp_invmod_odd_m( const mp_int *a, const mp_int *m, mp_int *c); |
|
197 mp_err s_mp_invmod_2d( const mp_int *a, mp_size k, mp_int *c); |
|
198 mp_err s_mp_invmod_even_m(const mp_int *a, const mp_int *m, mp_int *c); |
|
199 |
|
200 #ifdef NSS_USE_COMBA |
|
201 |
|
202 #define IS_POWER_OF_2(a) ((a) && !((a) & ((a)-1))) |
|
203 |
|
204 void s_mp_mul_comba_4(const mp_int *A, const mp_int *B, mp_int *C); |
|
205 void s_mp_mul_comba_8(const mp_int *A, const mp_int *B, mp_int *C); |
|
206 void s_mp_mul_comba_16(const mp_int *A, const mp_int *B, mp_int *C); |
|
207 void s_mp_mul_comba_32(const mp_int *A, const mp_int *B, mp_int *C); |
|
208 |
|
209 void s_mp_sqr_comba_4(const mp_int *A, mp_int *B); |
|
210 void s_mp_sqr_comba_8(const mp_int *A, mp_int *B); |
|
211 void s_mp_sqr_comba_16(const mp_int *A, mp_int *B); |
|
212 void s_mp_sqr_comba_32(const mp_int *A, mp_int *B); |
|
213 |
|
214 #endif /* end NSS_USE_COMBA */ |
|
215 |
|
216 /* ------ mpv functions, operate on arrays of digits, not on mp_int's ------ */ |
|
217 #if defined (__OS2__) && defined (__IBMC__) |
|
218 #define MPI_ASM_DECL __cdecl |
|
219 #else |
|
220 #define MPI_ASM_DECL |
|
221 #endif |
|
222 |
|
223 #ifdef MPI_AMD64 |
|
224 |
|
225 mp_digit MPI_ASM_DECL s_mpv_mul_set_vec64(mp_digit*, mp_digit *, mp_size, mp_digit); |
|
226 mp_digit MPI_ASM_DECL s_mpv_mul_add_vec64(mp_digit*, const mp_digit*, mp_size, mp_digit); |
|
227 |
|
228 /* c = a * b */ |
|
229 #define s_mpv_mul_d(a, a_len, b, c) \ |
|
230 ((mp_digit *)c)[a_len] = s_mpv_mul_set_vec64(c, a, a_len, b) |
|
231 |
|
232 /* c += a * b */ |
|
233 #define s_mpv_mul_d_add(a, a_len, b, c) \ |
|
234 ((mp_digit *)c)[a_len] = s_mpv_mul_add_vec64(c, a, a_len, b) |
|
235 |
|
236 |
|
237 #else |
|
238 |
|
239 void MPI_ASM_DECL s_mpv_mul_d(const mp_digit *a, mp_size a_len, |
|
240 mp_digit b, mp_digit *c); |
|
241 void MPI_ASM_DECL s_mpv_mul_d_add(const mp_digit *a, mp_size a_len, |
|
242 mp_digit b, mp_digit *c); |
|
243 |
|
244 #endif |
|
245 |
|
246 void MPI_ASM_DECL s_mpv_mul_d_add_prop(const mp_digit *a, |
|
247 mp_size a_len, mp_digit b, |
|
248 mp_digit *c); |
|
249 void MPI_ASM_DECL s_mpv_sqr_add_prop(const mp_digit *a, |
|
250 mp_size a_len, |
|
251 mp_digit *sqrs); |
|
252 |
|
253 mp_err MPI_ASM_DECL s_mpv_div_2dx1d(mp_digit Nhi, mp_digit Nlo, |
|
254 mp_digit divisor, mp_digit *quot, mp_digit *rem); |
|
255 |
|
256 /* c += a * b * (MP_RADIX ** offset); */ |
|
257 #define s_mp_mul_d_add_offset(a, b, c, off) \ |
|
258 (s_mpv_mul_d_add_prop(MP_DIGITS(a), MP_USED(a), b, MP_DIGITS(c) + off), MP_OKAY) |
|
259 |
|
260 typedef struct { |
|
261 mp_int N; /* modulus N */ |
|
262 mp_digit n0prime; /* n0' = - (n0 ** -1) mod MP_RADIX */ |
|
263 } mp_mont_modulus; |
|
264 |
|
265 mp_err s_mp_mul_mont(const mp_int *a, const mp_int *b, mp_int *c, |
|
266 mp_mont_modulus *mmm); |
|
267 mp_err s_mp_redc(mp_int *T, mp_mont_modulus *mmm); |
|
268 |
|
269 /* |
|
270 * s_mpi_getProcessorLineSize() returns the size in bytes of the cache line |
|
271 * if a cache exists, or zero if there is no cache. If more than one |
|
272 * cache line exists, it should return the smallest line size (which is |
|
273 * usually the L1 cache). |
|
274 * |
|
275 * mp_modexp uses this information to make sure that private key information |
|
276 * isn't being leaked through the cache. |
|
277 * |
|
278 * see mpcpucache.c for the implementation. |
|
279 */ |
|
280 unsigned long s_mpi_getProcessorLineSize(); |
|
281 |
|
282 /* }}} */ |
|
283 #endif |
|
284 |