]>
Commit | Line | Data |
---|---|---|
d02b48c6 | 1 | /* crypto/bn/bn_lcl.h */ |
58964a49 | 2 | /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com) |
d02b48c6 RE |
3 | * All rights reserved. |
4 | * | |
5 | * This package is an SSL implementation written | |
6 | * by Eric Young (eay@cryptsoft.com). | |
7 | * The implementation was written so as to conform with Netscapes SSL. | |
8 | * | |
9 | * This library is free for commercial and non-commercial use as long as | |
10 | * the following conditions are aheared to. The following conditions | |
11 | * apply to all code found in this distribution, be it the RC4, RSA, | |
12 | * lhash, DES, etc., code; not just the SSL code. The SSL documentation | |
13 | * included with this distribution is covered by the same copyright terms | |
14 | * except that the holder is Tim Hudson (tjh@cryptsoft.com). | |
15 | * | |
16 | * Copyright remains Eric Young's, and as such any Copyright notices in | |
17 | * the code are not to be removed. | |
18 | * If this package is used in a product, Eric Young should be given attribution | |
19 | * as the author of the parts of the library used. | |
20 | * This can be in the form of a textual message at program startup or | |
21 | * in documentation (online or textual) provided with the package. | |
22 | * | |
23 | * Redistribution and use in source and binary forms, with or without | |
24 | * modification, are permitted provided that the following conditions | |
25 | * are met: | |
26 | * 1. Redistributions of source code must retain the copyright | |
27 | * notice, this list of conditions and the following disclaimer. | |
28 | * 2. Redistributions in binary form must reproduce the above copyright | |
29 | * notice, this list of conditions and the following disclaimer in the | |
30 | * documentation and/or other materials provided with the distribution. | |
31 | * 3. All advertising materials mentioning features or use of this software | |
32 | * must display the following acknowledgement: | |
33 | * "This product includes cryptographic software written by | |
34 | * Eric Young (eay@cryptsoft.com)" | |
35 | * The word 'cryptographic' can be left out if the rouines from the library | |
36 | * being used are not cryptographic related :-). | |
37 | * 4. If you include any Windows specific code (or a derivative thereof) from | |
38 | * the apps directory (application code) you must include an acknowledgement: | |
39 | * "This product includes software written by Tim Hudson (tjh@cryptsoft.com)" | |
40 | * | |
41 | * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND | |
42 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
43 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
44 | * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE | |
45 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
46 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS | |
47 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
48 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT | |
49 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | |
50 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | |
51 | * SUCH DAMAGE. | |
52 | * | |
53 | * The licence and distribution terms for any publically available version or | |
54 | * derivative of this code cannot be changed. i.e. this code cannot simply be | |
55 | * copied and put under another distribution licence | |
56 | * [including the GNU Public Licence.] | |
57 | */ | |
dc434bbc BM |
58 | /* ==================================================================== |
59 | * Copyright (c) 1998-2000 The OpenSSL Project. All rights reserved. | |
60 | * | |
61 | * Redistribution and use in source and binary forms, with or without | |
62 | * modification, are permitted provided that the following conditions | |
63 | * are met: | |
64 | * | |
65 | * 1. Redistributions of source code must retain the above copyright | |
66 | * notice, this list of conditions and the following disclaimer. | |
67 | * | |
68 | * 2. Redistributions in binary form must reproduce the above copyright | |
69 | * notice, this list of conditions and the following disclaimer in | |
70 | * the documentation and/or other materials provided with the | |
71 | * distribution. | |
72 | * | |
73 | * 3. All advertising materials mentioning features or use of this | |
74 | * software must display the following acknowledgment: | |
75 | * "This product includes software developed by the OpenSSL Project | |
76 | * for use in the OpenSSL Toolkit. (http://www.openssl.org/)" | |
77 | * | |
78 | * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to | |
79 | * endorse or promote products derived from this software without | |
80 | * prior written permission. For written permission, please contact | |
81 | * openssl-core@openssl.org. | |
82 | * | |
83 | * 5. Products derived from this software may not be called "OpenSSL" | |
84 | * nor may "OpenSSL" appear in their names without prior written | |
85 | * permission of the OpenSSL Project. | |
86 | * | |
87 | * 6. Redistributions of any form whatsoever must retain the following | |
88 | * acknowledgment: | |
89 | * "This product includes software developed by the OpenSSL Project | |
90 | * for use in the OpenSSL Toolkit (http://www.openssl.org/)" | |
91 | * | |
92 | * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY | |
93 | * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
94 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | |
95 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR | |
96 | * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
97 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT | |
98 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; | |
99 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
100 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, | |
101 | * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | |
102 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED | |
103 | * OF THE POSSIBILITY OF SUCH DAMAGE. | |
104 | * ==================================================================== | |
105 | * | |
106 | * This product includes cryptographic software written by Eric Young | |
107 | * (eay@cryptsoft.com). This product includes software written by Tim | |
108 | * Hudson (tjh@cryptsoft.com). | |
109 | * | |
110 | */ | |
d02b48c6 RE |
111 | |
112 | #ifndef HEADER_BN_LCL_H | |
113 | #define HEADER_BN_LCL_H | |
114 | ||
85bcf27c | 115 | #include "internal/bn_int.h" |
d02b48c6 RE |
116 | |
117 | #ifdef __cplusplus | |
118 | extern "C" { | |
119 | #endif | |
120 | ||
02a62d1a MC |
121 | /* Bignum consistency macros |
122 | * There is one "API" macro, bn_fix_top(), for stripping leading zeroes from | |
123 | * bignum data after direct manipulations on the data. There is also an | |
124 | * "internal" macro, bn_check_top(), for verifying that there are no leading | |
125 | * zeroes. Unfortunately, some auditing is required due to the fact that | |
126 | * bn_fix_top() has become an overabused duct-tape because bignum data is | |
127 | * occasionally passed around in an inconsistent state. So the following | |
128 | * changes have been made to sort this out; | |
129 | * - bn_fix_top()s implementation has been moved to bn_correct_top() | |
130 | * - if BN_DEBUG isn't defined, bn_fix_top() maps to bn_correct_top(), and | |
131 | * bn_check_top() is as before. | |
132 | * - if BN_DEBUG *is* defined; | |
133 | * - bn_check_top() tries to pollute unused words even if the bignum 'top' is | |
134 | * consistent. (ed: only if BN_DEBUG_RAND is defined) | |
135 | * - bn_fix_top() maps to bn_check_top() rather than "fixing" anything. | |
136 | * The idea is to have debug builds flag up inconsistent bignums when they | |
137 | * occur. If that occurs in a bn_fix_top(), we examine the code in question; if | |
138 | * the use of bn_fix_top() was appropriate (ie. it follows directly after code | |
139 | * that manipulates the bignum) it is converted to bn_correct_top(), and if it | |
140 | * was not appropriate, we convert it permanently to bn_check_top() and track | |
141 | * down the cause of the bug. Eventually, no internal code should be using the | |
142 | * bn_fix_top() macro. External applications and libraries should try this with | |
143 | * their own code too, both in terms of building against the openssl headers | |
144 | * with BN_DEBUG defined *and* linking with a version of OpenSSL built with it | |
145 | * defined. This not only improves external code, it provides more test | |
146 | * coverage for openssl's own code. | |
147 | */ | |
148 | ||
149 | #ifdef BN_DEBUG | |
150 | ||
151 | /* We only need assert() when debugging */ | |
152 | #include <assert.h> | |
153 | ||
154 | #ifdef BN_DEBUG_RAND | |
155 | /* To avoid "make update" cvs wars due to BN_DEBUG, use some tricks */ | |
156 | #ifndef RAND_pseudo_bytes | |
157 | int RAND_pseudo_bytes(unsigned char *buf,int num); | |
158 | #define BN_DEBUG_TRIX | |
159 | #endif | |
160 | #define bn_pollute(a) \ | |
161 | do { \ | |
162 | const BIGNUM *_bnum1 = (a); \ | |
163 | if(_bnum1->top < _bnum1->dmax) { \ | |
164 | unsigned char _tmp_char; \ | |
165 | /* We cast away const without the compiler knowing, any \ | |
166 | * *genuinely* constant variables that aren't mutable \ | |
167 | * wouldn't be constructed with top!=dmax. */ \ | |
168 | BN_ULONG *_not_const; \ | |
169 | memcpy(&_not_const, &_bnum1->d, sizeof(BN_ULONG*)); \ | |
170 | RAND_pseudo_bytes(&_tmp_char, 1); \ | |
171 | memset((unsigned char *)(_not_const + _bnum1->top), _tmp_char, \ | |
172 | (_bnum1->dmax - _bnum1->top) * sizeof(BN_ULONG)); \ | |
173 | } \ | |
174 | } while(0) | |
175 | #ifdef BN_DEBUG_TRIX | |
176 | #undef RAND_pseudo_bytes | |
177 | #endif | |
178 | #else | |
179 | #define bn_pollute(a) | |
180 | #endif | |
181 | #define bn_check_top(a) \ | |
182 | do { \ | |
183 | const BIGNUM *_bnum2 = (a); \ | |
184 | if (_bnum2 != NULL) { \ | |
185 | assert((_bnum2->top == 0) || \ | |
186 | (_bnum2->d[_bnum2->top - 1] != 0)); \ | |
187 | bn_pollute(_bnum2); \ | |
188 | } \ | |
189 | } while(0) | |
190 | ||
191 | #define bn_fix_top(a) bn_check_top(a) | |
192 | ||
193 | #define bn_check_size(bn, bits) bn_wcheck_size(bn, ((bits+BN_BITS2-1))/BN_BITS2) | |
194 | #define bn_wcheck_size(bn, words) \ | |
195 | do { \ | |
196 | const BIGNUM *_bnum2 = (bn); \ | |
9669d2e1 EK |
197 | assert((words) <= (_bnum2)->dmax && (words) >= (_bnum2)->top); \ |
198 | /* avoid unused variable warning with NDEBUG */ \ | |
199 | (void)(_bnum2); \ | |
02a62d1a MC |
200 | } while(0) |
201 | ||
202 | #else /* !BN_DEBUG */ | |
203 | ||
204 | #define bn_pollute(a) | |
205 | #define bn_check_top(a) | |
206 | #define bn_fix_top(a) bn_correct_top(a) | |
207 | #define bn_check_size(bn, bits) | |
208 | #define bn_wcheck_size(bn, words) | |
209 | ||
210 | #endif | |
211 | ||
212 | ||
213 | BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w); | |
214 | BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w); | |
215 | void bn_sqr_words(BN_ULONG *rp, const BN_ULONG *ap, int num); | |
216 | BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d); | |
217 | BN_ULONG bn_add_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,int num); | |
218 | BN_ULONG bn_sub_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,int num); | |
219 | ||
dc434bbc | 220 | |
19391879 MC |
221 | struct bignum_st |
222 | { | |
223 | BN_ULONG *d; /* Pointer to an array of 'BN_BITS2' bit chunks. */ | |
224 | int top; /* Index of last used d +1. */ | |
225 | /* The next are internal book keeping for bn_expand. */ | |
226 | int dmax; /* Size of the d array. */ | |
227 | int neg; /* one if the number is negative */ | |
228 | int flags; | |
229 | }; | |
230 | ||
231 | /* Used for montgomery multiplication */ | |
232 | struct bn_mont_ctx_st | |
233 | { | |
234 | int ri; /* number of bits in R */ | |
235 | BIGNUM RR; /* used to convert to montgomery form */ | |
236 | BIGNUM N; /* The modulus */ | |
237 | BIGNUM Ni; /* R*(1/R mod N) - N*Ni = 1 | |
238 | * (Ni is only stored for bignum algorithm) */ | |
239 | BN_ULONG n0[2];/* least significant word(s) of Ni; | |
240 | (type changed with 0.9.9, was "BN_ULONG n0;" before) */ | |
241 | int flags; | |
242 | }; | |
243 | ||
244 | /* Used for reciprocal division/mod functions | |
245 | * It cannot be shared between threads | |
246 | */ | |
247 | struct bn_recp_ctx_st | |
248 | { | |
249 | BIGNUM N; /* the divisor */ | |
250 | BIGNUM Nr; /* the reciprocal */ | |
251 | int num_bits; | |
252 | int shift; | |
253 | int flags; | |
254 | }; | |
255 | ||
256 | /* Used for slow "generation" functions. */ | |
257 | struct bn_gencb_st | |
258 | { | |
259 | unsigned int ver; /* To handle binary (in)compatibility */ | |
260 | void *arg; /* callback-specific data */ | |
261 | union | |
262 | { | |
263 | /* if(ver==1) - handles old style callbacks */ | |
264 | void (*cb_1)(int, int, void *); | |
265 | /* if(ver==2) - new callback style */ | |
266 | int (*cb_2)(int, int, BN_GENCB *); | |
267 | } cb; | |
268 | }; | |
269 | ||
270 | ||
dc434bbc BM |
271 | /* |
272 | * BN_window_bits_for_exponent_size -- macro for sliding window mod_exp functions | |
273 | * | |
274 | * | |
275 | * For window size 'w' (w >= 2) and a random 'b' bits exponent, | |
276 | * the number of multiplications is a constant plus on average | |
277 | * | |
278 | * 2^(w-1) + (b-w)/(w+1); | |
279 | * | |
280 | * here 2^(w-1) is for precomputing the table (we actually need | |
281 | * entries only for windows that have the lowest bit set), and | |
282 | * (b-w)/(w+1) is an approximation for the expected number of | |
283 | * w-bit windows, not counting the first one. | |
284 | * | |
285 | * Thus we should use | |
286 | * | |
287 | * w >= 6 if b > 671 | |
288 | * w = 5 if 671 > b > 239 | |
289 | * w = 4 if 239 > b > 79 | |
290 | * w = 3 if 79 > b > 23 | |
291 | * w <= 2 if 23 > b | |
292 | * | |
293 | * (with draws in between). Very small exponents are often selected | |
294 | * with low Hamming weight, so we use w = 1 for b <= 23. | |
295 | */ | |
296 | #if 1 | |
297 | #define BN_window_bits_for_exponent_size(b) \ | |
298 | ((b) > 671 ? 6 : \ | |
299 | (b) > 239 ? 5 : \ | |
300 | (b) > 79 ? 4 : \ | |
301 | (b) > 23 ? 3 : 1) | |
302 | #else | |
303 | /* Old SSLeay/OpenSSL table. | |
304 | * Maximum window size was 5, so this table differs for b==1024; | |
305 | * but it coincides for other interesting values (b==160, b==512). | |
306 | */ | |
307 | #define BN_window_bits_for_exponent_size(b) \ | |
308 | ((b) > 255 ? 5 : \ | |
309 | (b) > 127 ? 4 : \ | |
310 | (b) > 17 ? 3 : 1) | |
311 | #endif | |
312 | ||
313 | ||
314 | ||
46a64376 BM |
315 | /* BN_mod_exp_mont_conttime is based on the assumption that the |
316 | * L1 data cache line width of the target processor is at least | |
317 | * the following value. | |
318 | */ | |
319 | #define MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH ( 64 ) | |
320 | #define MOD_EXP_CTIME_MIN_CACHE_LINE_MASK (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - 1) | |
321 | ||
322 | /* Window sizes optimized for fixed window size modular exponentiation | |
323 | * algorithm (BN_mod_exp_mont_consttime). | |
324 | * | |
325 | * To achieve the security goals of BN_mode_exp_mont_consttime, the | |
326 | * maximum size of the window must not exceed | |
327 | * log_2(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH). | |
328 | * | |
329 | * Window size thresholds are defined for cache line sizes of 32 and 64, | |
330 | * cache line sizes where log_2(32)=5 and log_2(64)=6 respectively. A | |
331 | * window size of 7 should only be used on processors that have a 128 | |
332 | * byte or greater cache line size. | |
333 | */ | |
334 | #if MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 64 | |
335 | ||
336 | # define BN_window_bits_for_ctime_exponent_size(b) \ | |
337 | ((b) > 937 ? 6 : \ | |
338 | (b) > 306 ? 5 : \ | |
339 | (b) > 89 ? 4 : \ | |
340 | (b) > 22 ? 3 : 1) | |
341 | # define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (6) | |
342 | ||
343 | #elif MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 32 | |
344 | ||
345 | # define BN_window_bits_for_ctime_exponent_size(b) \ | |
346 | ((b) > 306 ? 5 : \ | |
347 | (b) > 89 ? 4 : \ | |
348 | (b) > 22 ? 3 : 1) | |
349 | # define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (5) | |
350 | ||
351 | #endif | |
352 | ||
353 | ||
dfeab068 RE |
354 | /* Pentium pro 16,16,16,32,64 */ |
355 | /* Alpha 16,16,16,16.64 */ | |
27eb622b DSH |
356 | #define BN_MULL_SIZE_NORMAL (16) /* 32 */ |
357 | #define BN_MUL_RECURSIVE_SIZE_NORMAL (16) /* 32 less than */ | |
358 | #define BN_SQR_RECURSIVE_SIZE_NORMAL (16) /* 32 */ | |
359 | #define BN_MUL_LOW_RECURSIVE_SIZE_NORMAL (32) /* 32 */ | |
360 | #define BN_MONT_CTX_SET_SIZE_WORD (64) /* 32 */ | |
dfeab068 | 361 | |
8d00f342 AP |
362 | /* 2011-02-22 SMS. |
363 | * In various places, a size_t variable or a type cast to size_t was | |
364 | * used to perform integer-only operations on pointers. This failed on | |
365 | * VMS with 64-bit pointers (CC /POINTER_SIZE = 64) because size_t is | |
366 | * still only 32 bits. What's needed in these cases is an integer type | |
367 | * with the same size as a pointer, which size_t is not certain to be. | |
368 | * The only fix here is VMS-specific. | |
369 | */ | |
370 | #if defined(OPENSSL_SYS_VMS) | |
371 | # if __INITIAL_POINTER_SIZE == 64 | |
372 | # define PTR_SIZE_INT long long | |
373 | # else /* __INITIAL_POINTER_SIZE == 64 */ | |
374 | # define PTR_SIZE_INT int | |
375 | # endif /* __INITIAL_POINTER_SIZE == 64 [else] */ | |
376 | #elif !defined(PTR_SIZE_INT) /* defined(OPENSSL_SYS_VMS) */ | |
377 | # define PTR_SIZE_INT size_t | |
378 | #endif /* defined(OPENSSL_SYS_VMS) [else] */ | |
379 | ||
cf1b7d96 | 380 | #if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) && !defined(PEDANTIC) |
fb81ac5e AP |
381 | /* |
382 | * BN_UMULT_HIGH section. | |
383 | * | |
384 | * No, I'm not trying to overwhelm you when stating that the | |
385 | * product of N-bit numbers is 2*N bits wide:-) No, I don't expect | |
386 | * you to be impressed when I say that if the compiler doesn't | |
387 | * support 2*N integer type, then you have to replace every N*N | |
388 | * multiplication with 4 (N/2)*(N/2) accompanied by some shifts | |
389 | * and additions which unavoidably results in severe performance | |
390 | * penalties. Of course provided that the hardware is capable of | |
391 | * producing 2*N result... That's when you normally start | |
392 | * considering assembler implementation. However! It should be | |
393 | * pointed out that some CPUs (most notably Alpha, PowerPC and | |
394 | * upcoming IA-64 family:-) provide *separate* instruction | |
395 | * calculating the upper half of the product placing the result | |
396 | * into a general purpose register. Now *if* the compiler supports | |
397 | * inline assembler, then it's not impossible to implement the | |
398 | * "bignum" routines (and have the compiler optimize 'em) | |
399 | * exhibiting "native" performance in C. That's what BN_UMULT_HIGH | |
400 | * macro is about:-) | |
401 | * | |
402 | * <appro@fy.chalmers.se> | |
403 | */ | |
404 | # if defined(__alpha) && (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT)) | |
405 | # if defined(__DECC) | |
406 | # include <c_asm.h> | |
407 | # define BN_UMULT_HIGH(a,b) (BN_ULONG)asm("umulh %a0,%a1,%v0",(a),(b)) | |
4f385652 | 408 | # elif defined(__GNUC__) && __GNUC__>=2 |
fb81ac5e AP |
409 | # define BN_UMULT_HIGH(a,b) ({ \ |
410 | register BN_ULONG ret; \ | |
411 | asm ("umulh %1,%2,%0" \ | |
412 | : "=r"(ret) \ | |
413 | : "r"(a), "r"(b)); \ | |
414 | ret; }) | |
415 | # endif /* compiler */ | |
416 | # elif defined(_ARCH_PPC) && defined(__64BIT__) && defined(SIXTY_FOUR_BIT_LONG) | |
4f385652 | 417 | # if defined(__GNUC__) && __GNUC__>=2 |
fb81ac5e AP |
418 | # define BN_UMULT_HIGH(a,b) ({ \ |
419 | register BN_ULONG ret; \ | |
420 | asm ("mulhdu %0,%1,%2" \ | |
421 | : "=r"(ret) \ | |
422 | : "r"(a), "r"(b)); \ | |
423 | ret; }) | |
424 | # endif /* compiler */ | |
122396f2 AP |
425 | # elif (defined(__x86_64) || defined(__x86_64__)) && \ |
426 | (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT)) | |
4f385652 | 427 | # if defined(__GNUC__) && __GNUC__>=2 |
2f98abbc AP |
428 | # define BN_UMULT_HIGH(a,b) ({ \ |
429 | register BN_ULONG ret,discard; \ | |
430 | asm ("mulq %3" \ | |
431 | : "=a"(discard),"=d"(ret) \ | |
432 | : "a"(a), "g"(b) \ | |
433 | : "cc"); \ | |
434 | ret; }) | |
435 | # define BN_UMULT_LOHI(low,high,a,b) \ | |
436 | asm ("mulq %3" \ | |
437 | : "=a"(low),"=d"(high) \ | |
438 | : "a"(a),"g"(b) \ | |
439 | : "cc"); | |
440 | # endif | |
11de71b0 AP |
441 | # elif (defined(_M_AMD64) || defined(_M_X64)) && defined(SIXTY_FOUR_BIT) |
442 | # if defined(_MSC_VER) && _MSC_VER>=1400 | |
443 | unsigned __int64 __umulh (unsigned __int64 a,unsigned __int64 b); | |
444 | unsigned __int64 _umul128 (unsigned __int64 a,unsigned __int64 b, | |
445 | unsigned __int64 *h); | |
446 | # pragma intrinsic(__umulh,_umul128) | |
447 | # define BN_UMULT_HIGH(a,b) __umulh((a),(b)) | |
448 | # define BN_UMULT_LOHI(low,high,a,b) ((low)=_umul128((a),(b),&(high))) | |
449 | # endif | |
4f385652 AP |
450 | # elif defined(__mips) && (defined(SIXTY_FOUR_BIT) || defined(SIXTY_FOUR_BIT_LONG)) |
451 | # if defined(__GNUC__) && __GNUC__>=2 | |
a58fdc7a AP |
452 | # if __GNUC__>=4 && __GNUC_MINOR__>=4 /* "h" constraint is no more since 4.4 */ |
453 | # define BN_UMULT_HIGH(a,b) (((__uint128_t)(a)*(b))>>64) | |
454 | # define BN_UMULT_LOHI(low,high,a,b) ({ \ | |
455 | __uint128_t ret=(__uint128_t)(a)*(b); \ | |
456 | (high)=ret>>64; (low)=ret; }) | |
457 | # else | |
458 | # define BN_UMULT_HIGH(a,b) ({ \ | |
4f385652 AP |
459 | register BN_ULONG ret; \ |
460 | asm ("dmultu %1,%2" \ | |
461 | : "=h"(ret) \ | |
462 | : "r"(a), "r"(b) : "l"); \ | |
463 | ret; }) | |
a58fdc7a | 464 | # define BN_UMULT_LOHI(low,high,a,b)\ |
4f385652 AP |
465 | asm ("dmultu %2,%3" \ |
466 | : "=l"(low),"=h"(high) \ | |
467 | : "r"(a), "r"(b)); | |
a58fdc7a | 468 | # endif |
4f385652 | 469 | # endif |
039081b8 AP |
470 | # elif defined(__aarch64__) && defined(SIXTY_FOUR_BIT_LONG) |
471 | # if defined(__GNUC__) && __GNUC__>=2 | |
472 | # define BN_UMULT_HIGH(a,b) ({ \ | |
473 | register BN_ULONG ret; \ | |
474 | asm ("umulh %0,%1,%2" \ | |
475 | : "=r"(ret) \ | |
476 | : "r"(a), "r"(b)); \ | |
477 | ret; }) | |
478 | # endif | |
fb81ac5e | 479 | # endif /* cpu */ |
cf1b7d96 | 480 | #endif /* OPENSSL_NO_ASM */ |
fb81ac5e | 481 | |
d02b48c6 RE |
482 | /************************************************************* |
483 | * Using the long long type | |
484 | */ | |
485 | #define Lw(t) (((BN_ULONG)(t))&BN_MASK2) | |
486 | #define Hw(t) (((BN_ULONG)((t)>>BN_BITS2))&BN_MASK2) | |
487 | ||
a8aa764d | 488 | #ifdef BN_DEBUG_RAND |
19b8d06a BM |
489 | #define bn_clear_top2max(a) \ |
490 | { \ | |
90a617e0 | 491 | int ind = (a)->dmax - (a)->top; \ |
19b8d06a | 492 | BN_ULONG *ftl = &(a)->d[(a)->top-1]; \ |
90a617e0 | 493 | for (; ind != 0; ind--) \ |
19b8d06a BM |
494 | *(++ftl) = 0x0; \ |
495 | } | |
a8aa764d GT |
496 | #else |
497 | #define bn_clear_top2max(a) | |
498 | #endif | |
19b8d06a | 499 | |
d02b48c6 RE |
500 | #ifdef BN_LLONG |
501 | #define mul_add(r,a,w,c) { \ | |
502 | BN_ULLONG t; \ | |
503 | t=(BN_ULLONG)w * (a) + (r) + (c); \ | |
58964a49 | 504 | (r)= Lw(t); \ |
d02b48c6 RE |
505 | (c)= Hw(t); \ |
506 | } | |
507 | ||
508 | #define mul(r,a,w,c) { \ | |
509 | BN_ULLONG t; \ | |
510 | t=(BN_ULLONG)w * (a) + (c); \ | |
58964a49 | 511 | (r)= Lw(t); \ |
d02b48c6 RE |
512 | (c)= Hw(t); \ |
513 | } | |
514 | ||
fb81ac5e AP |
515 | #define sqr(r0,r1,a) { \ |
516 | BN_ULLONG t; \ | |
517 | t=(BN_ULLONG)(a)*(a); \ | |
518 | (r0)=Lw(t); \ | |
aff0825c | 519 | (r1)=Hw(t); \ |
fb81ac5e AP |
520 | } |
521 | ||
11de71b0 AP |
522 | #elif defined(BN_UMULT_LOHI) |
523 | #define mul_add(r,a,w,c) { \ | |
524 | BN_ULONG high,low,ret,tmp=(a); \ | |
525 | ret = (r); \ | |
526 | BN_UMULT_LOHI(low,high,w,tmp); \ | |
527 | ret += (c); \ | |
528 | (c) = (ret<(c))?1:0; \ | |
529 | (c) += high; \ | |
530 | ret += low; \ | |
531 | (c) += (ret<low)?1:0; \ | |
532 | (r) = ret; \ | |
533 | } | |
534 | ||
535 | #define mul(r,a,w,c) { \ | |
536 | BN_ULONG high,low,ret,ta=(a); \ | |
537 | BN_UMULT_LOHI(low,high,w,ta); \ | |
538 | ret = low + (c); \ | |
539 | (c) = high; \ | |
540 | (c) += (ret<low)?1:0; \ | |
541 | (r) = ret; \ | |
542 | } | |
543 | ||
544 | #define sqr(r0,r1,a) { \ | |
545 | BN_ULONG tmp=(a); \ | |
546 | BN_UMULT_LOHI(r0,r1,tmp,tmp); \ | |
547 | } | |
548 | ||
fb81ac5e AP |
549 | #elif defined(BN_UMULT_HIGH) |
550 | #define mul_add(r,a,w,c) { \ | |
551 | BN_ULONG high,low,ret,tmp=(a); \ | |
552 | ret = (r); \ | |
553 | high= BN_UMULT_HIGH(w,tmp); \ | |
554 | ret += (c); \ | |
555 | low = (w) * tmp; \ | |
556 | (c) = (ret<(c))?1:0; \ | |
557 | (c) += high; \ | |
558 | ret += low; \ | |
559 | (c) += (ret<low)?1:0; \ | |
560 | (r) = ret; \ | |
561 | } | |
562 | ||
563 | #define mul(r,a,w,c) { \ | |
564 | BN_ULONG high,low,ret,ta=(a); \ | |
565 | low = (w) * ta; \ | |
566 | high= BN_UMULT_HIGH(w,ta); \ | |
567 | ret = low + (c); \ | |
568 | (c) = high; \ | |
569 | (c) += (ret<low)?1:0; \ | |
570 | (r) = ret; \ | |
571 | } | |
572 | ||
573 | #define sqr(r0,r1,a) { \ | |
574 | BN_ULONG tmp=(a); \ | |
575 | (r0) = tmp * tmp; \ | |
576 | (r1) = BN_UMULT_HIGH(tmp,tmp); \ | |
577 | } | |
578 | ||
d02b48c6 RE |
579 | #else |
580 | /************************************************************* | |
581 | * No long long type | |
582 | */ | |
583 | ||
584 | #define LBITS(a) ((a)&BN_MASK2l) | |
585 | #define HBITS(a) (((a)>>BN_BITS4)&BN_MASK2l) | |
2f98abbc | 586 | #define L2HBITS(a) (((a)<<BN_BITS4)&BN_MASK2) |
d02b48c6 RE |
587 | |
588 | #define LLBITS(a) ((a)&BN_MASKl) | |
589 | #define LHBITS(a) (((a)>>BN_BITS2)&BN_MASKl) | |
590 | #define LL2HBITS(a) ((BN_ULLONG)((a)&BN_MASKl)<<BN_BITS2) | |
591 | ||
592 | #define mul64(l,h,bl,bh) \ | |
593 | { \ | |
594 | BN_ULONG m,m1,lt,ht; \ | |
595 | \ | |
596 | lt=l; \ | |
597 | ht=h; \ | |
598 | m =(bh)*(lt); \ | |
599 | lt=(bl)*(lt); \ | |
600 | m1=(bl)*(ht); \ | |
601 | ht =(bh)*(ht); \ | |
316bfb77 | 602 | m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS((BN_ULONG)1); \ |
d02b48c6 RE |
603 | ht+=HBITS(m); \ |
604 | m1=L2HBITS(m); \ | |
58964a49 | 605 | lt=(lt+m1)&BN_MASK2; if (lt < m1) ht++; \ |
d02b48c6 RE |
606 | (l)=lt; \ |
607 | (h)=ht; \ | |
608 | } | |
609 | ||
610 | #define sqr64(lo,ho,in) \ | |
611 | { \ | |
612 | BN_ULONG l,h,m; \ | |
613 | \ | |
614 | h=(in); \ | |
615 | l=LBITS(h); \ | |
616 | h=HBITS(h); \ | |
617 | m =(l)*(h); \ | |
618 | l*=l; \ | |
619 | h*=h; \ | |
620 | h+=(m&BN_MASK2h1)>>(BN_BITS4-1); \ | |
621 | m =(m&BN_MASK2l)<<(BN_BITS4+1); \ | |
58964a49 | 622 | l=(l+m)&BN_MASK2; if (l < m) h++; \ |
d02b48c6 RE |
623 | (lo)=l; \ |
624 | (ho)=h; \ | |
625 | } | |
626 | ||
627 | #define mul_add(r,a,bl,bh,c) { \ | |
628 | BN_ULONG l,h; \ | |
629 | \ | |
630 | h= (a); \ | |
631 | l=LBITS(h); \ | |
632 | h=HBITS(h); \ | |
633 | mul64(l,h,(bl),(bh)); \ | |
634 | \ | |
635 | /* non-multiply part */ \ | |
58964a49 | 636 | l=(l+(c))&BN_MASK2; if (l < (c)) h++; \ |
d02b48c6 | 637 | (c)=(r); \ |
58964a49 | 638 | l=(l+(c))&BN_MASK2; if (l < (c)) h++; \ |
d02b48c6 | 639 | (c)=h&BN_MASK2; \ |
58964a49 | 640 | (r)=l; \ |
d02b48c6 RE |
641 | } |
642 | ||
643 | #define mul(r,a,bl,bh,c) { \ | |
644 | BN_ULONG l,h; \ | |
645 | \ | |
646 | h= (a); \ | |
647 | l=LBITS(h); \ | |
648 | h=HBITS(h); \ | |
649 | mul64(l,h,(bl),(bh)); \ | |
650 | \ | |
651 | /* non-multiply part */ \ | |
652 | l+=(c); if ((l&BN_MASK2) < (c)) h++; \ | |
653 | (c)=h&BN_MASK2; \ | |
654 | (r)=l&BN_MASK2; \ | |
655 | } | |
775c63fc | 656 | #endif /* !BN_LLONG */ |
d02b48c6 | 657 | |
19391879 MC |
658 | void BN_init(BIGNUM *a); |
659 | void BN_RECP_CTX_init(BN_RECP_CTX *recp); | |
660 | void BN_MONT_CTX_init(BN_MONT_CTX *ctx); | |
661 | ||
fb92ba64 UM |
662 | void bn_mul_normal(BN_ULONG *r,BN_ULONG *a,int na,BN_ULONG *b,int nb); |
663 | void bn_mul_comba8(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b); | |
664 | void bn_mul_comba4(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b); | |
cbd48ba6 RL |
665 | void bn_sqr_normal(BN_ULONG *r, const BN_ULONG *a, int n, BN_ULONG *tmp); |
666 | void bn_sqr_comba8(BN_ULONG *r,const BN_ULONG *a); | |
667 | void bn_sqr_comba4(BN_ULONG *r,const BN_ULONG *a); | |
668 | int bn_cmp_words(const BN_ULONG *a,const BN_ULONG *b,int n); | |
52a1bab2 UM |
669 | int bn_cmp_part_words(const BN_ULONG *a, const BN_ULONG *b, |
670 | int cl, int dl); | |
6343829a GT |
671 | void bn_mul_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2, |
672 | int dna,int dnb,BN_ULONG *t); | |
673 | void bn_mul_part_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b, | |
674 | int n,int tna,int tnb,BN_ULONG *t); | |
675 | void bn_sqr_recursive(BN_ULONG *r,const BN_ULONG *a, int n2, BN_ULONG *t); | |
676 | void bn_mul_low_normal(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b, int n); | |
677 | void bn_mul_low_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2, | |
678 | BN_ULONG *t); | |
679 | void bn_mul_high(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,BN_ULONG *l,int n2, | |
680 | BN_ULONG *t); | |
d5c21afd | 681 | BN_ULONG bn_add_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, |
6343829a GT |
682 | int cl, int dl); |
683 | BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, | |
684 | int cl, int dl); | |
685 | int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np,const BN_ULONG *n0, int num); | |
58964a49 | 686 | |
879bd6e3 DSH |
687 | BIGNUM *int_bn_mod_inverse(BIGNUM *in, |
688 | const BIGNUM *a, const BIGNUM *n, BN_CTX *ctx, int *noinv); | |
689 | ||
e46a059e FLM |
690 | int bn_probable_prime_dh(BIGNUM *rnd, int bits, |
691 | const BIGNUM *add, const BIGNUM *rem, BN_CTX *ctx); | |
982c42cb FLM |
692 | int bn_probable_prime_dh_retry(BIGNUM *rnd, int bits, BN_CTX *ctx); |
693 | int bn_probable_prime_dh_coprime(BIGNUM *rnd, int bits, BN_CTX *ctx); | |
e46a059e | 694 | |
d02b48c6 RE |
695 | #ifdef __cplusplus |
696 | } | |
697 | #endif | |
698 | ||
699 | #endif |