OpenSSL  1.0.1c
 All Classes Files Functions Variables Typedefs Enumerations Enumerator Macros
bn_lcl.h
Go to the documentation of this file.
1 /* crypto/bn/bn_lcl.h */
2 /* Copyright (C) 1995-1998 Eric Young ([email protected])
3  * All rights reserved.
4  *
5  * This package is an SSL implementation written
6  * by Eric Young ([email protected]).
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 ([email protected]).
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 ([email protected])"
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 ([email protected])"
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  */
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
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  * ([email protected]). This product includes software written by Tim
108  * Hudson ([email protected]).
109  *
110  */
111 
112 #ifndef HEADER_BN_LCL_H
113 #define HEADER_BN_LCL_H
114 
115 #include <openssl/bn.h>
116 
117 #ifdef __cplusplus
118 extern "C" {
119 #endif
120 
121 
122 /*
123  * BN_window_bits_for_exponent_size -- macro for sliding window mod_exp functions
124  *
125  *
126  * For window size 'w' (w >= 2) and a random 'b' bits exponent,
127  * the number of multiplications is a constant plus on average
128  *
129  * 2^(w-1) + (b-w)/(w+1);
130  *
131  * here 2^(w-1) is for precomputing the table (we actually need
132  * entries only for windows that have the lowest bit set), and
133  * (b-w)/(w+1) is an approximation for the expected number of
134  * w-bit windows, not counting the first one.
135  *
136  * Thus we should use
137  *
138  * w >= 6 if b > 671
139  * w = 5 if 671 > b > 239
140  * w = 4 if 239 > b > 79
141  * w = 3 if 79 > b > 23
142  * w <= 2 if 23 > b
143  *
144  * (with draws in between). Very small exponents are often selected
145  * with low Hamming weight, so we use w = 1 for b <= 23.
146  */
147 #if 1
148 #define BN_window_bits_for_exponent_size(b) \
149  ((b) > 671 ? 6 : \
150  (b) > 239 ? 5 : \
151  (b) > 79 ? 4 : \
152  (b) > 23 ? 3 : 1)
153 #else
154 /* Old SSLeay/OpenSSL table.
155  * Maximum window size was 5, so this table differs for b==1024;
156  * but it coincides for other interesting values (b==160, b==512).
157  */
158 #define BN_window_bits_for_exponent_size(b) \
159  ((b) > 255 ? 5 : \
160  (b) > 127 ? 4 : \
161  (b) > 17 ? 3 : 1)
162 #endif
163 
164 
165 
166 /* BN_mod_exp_mont_conttime is based on the assumption that the
167  * L1 data cache line width of the target processor is at least
168  * the following value.
169  */
170 #define MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH ( 64 )
171 #define MOD_EXP_CTIME_MIN_CACHE_LINE_MASK (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - 1)
172 
173 /* Window sizes optimized for fixed window size modular exponentiation
174  * algorithm (BN_mod_exp_mont_consttime).
175  *
176  * To achieve the security goals of BN_mode_exp_mont_consttime, the
177  * maximum size of the window must not exceed
178  * log_2(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH).
179  *
180  * Window size thresholds are defined for cache line sizes of 32 and 64,
181  * cache line sizes where log_2(32)=5 and log_2(64)=6 respectively. A
182  * window size of 7 should only be used on processors that have a 128
183  * byte or greater cache line size.
184  */
185 #if MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 64
186 
187 # define BN_window_bits_for_ctime_exponent_size(b) \
188  ((b) > 937 ? 6 : \
189  (b) > 306 ? 5 : \
190  (b) > 89 ? 4 : \
191  (b) > 22 ? 3 : 1)
192 # define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (6)
193 
194 #elif MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 32
195 
196 # define BN_window_bits_for_ctime_exponent_size(b) \
197  ((b) > 306 ? 5 : \
198  (b) > 89 ? 4 : \
199  (b) > 22 ? 3 : 1)
200 # define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (5)
201 
202 #endif
203 
204 
205 /* Pentium pro 16,16,16,32,64 */
206 /* Alpha 16,16,16,16.64 */
207 #define BN_MULL_SIZE_NORMAL (16) /* 32 */
208 #define BN_MUL_RECURSIVE_SIZE_NORMAL (16) /* 32 less than */
209 #define BN_SQR_RECURSIVE_SIZE_NORMAL (16) /* 32 */
210 #define BN_MUL_LOW_RECURSIVE_SIZE_NORMAL (32) /* 32 */
211 #define BN_MONT_CTX_SET_SIZE_WORD (64) /* 32 */
212 
213 #if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) && !defined(PEDANTIC)
214 /*
215  * BN_UMULT_HIGH section.
216  *
217  * No, I'm not trying to overwhelm you when stating that the
218  * product of N-bit numbers is 2*N bits wide:-) No, I don't expect
219  * you to be impressed when I say that if the compiler doesn't
220  * support 2*N integer type, then you have to replace every N*N
221  * multiplication with 4 (N/2)*(N/2) accompanied by some shifts
222  * and additions which unavoidably results in severe performance
223  * penalties. Of course provided that the hardware is capable of
224  * producing 2*N result... That's when you normally start
225  * considering assembler implementation. However! It should be
226  * pointed out that some CPUs (most notably Alpha, PowerPC and
227  * upcoming IA-64 family:-) provide *separate* instruction
228  * calculating the upper half of the product placing the result
229  * into a general purpose register. Now *if* the compiler supports
230  * inline assembler, then it's not impossible to implement the
231  * "bignum" routines (and have the compiler optimize 'em)
232  * exhibiting "native" performance in C. That's what BN_UMULT_HIGH
233  * macro is about:-)
234  *
236  */
237 # if defined(__alpha) && (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT))
238 # if defined(__DECC)
239 # include <c_asm.h>
240 # define BN_UMULT_HIGH(a,b) (BN_ULONG)asm("umulh %a0,%a1,%v0",(a),(b))
241 # elif defined(__GNUC__) && __GNUC__>=2
242 # define BN_UMULT_HIGH(a,b) ({ \
243  register BN_ULONG ret; \
244  asm ("umulh %1,%2,%0" \
245  : "=r"(ret) \
246  : "r"(a), "r"(b)); \
247  ret; })
248 # endif /* compiler */
249 # elif defined(_ARCH_PPC) && defined(__64BIT__) && defined(SIXTY_FOUR_BIT_LONG)
250 # if defined(__GNUC__) && __GNUC__>=2
251 # define BN_UMULT_HIGH(a,b) ({ \
252  register BN_ULONG ret; \
253  asm ("mulhdu %0,%1,%2" \
254  : "=r"(ret) \
255  : "r"(a), "r"(b)); \
256  ret; })
257 # endif /* compiler */
258 # elif (defined(__x86_64) || defined(__x86_64__)) && \
259  (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT))
260 # if defined(__GNUC__) && __GNUC__>=2
261 # define BN_UMULT_HIGH(a,b) ({ \
262  register BN_ULONG ret,discard; \
263  asm ("mulq %3" \
264  : "=a"(discard),"=d"(ret) \
265  : "a"(a), "g"(b) \
266  : "cc"); \
267  ret; })
268 # define BN_UMULT_LOHI(low,high,a,b) \
269  asm ("mulq %3" \
270  : "=a"(low),"=d"(high) \
271  : "a"(a),"g"(b) \
272  : "cc");
273 # endif
274 # elif (defined(_M_AMD64) || defined(_M_X64)) && defined(SIXTY_FOUR_BIT)
275 # if defined(_MSC_VER) && _MSC_VER>=1400
276  unsigned __int64 __umulh (unsigned __int64 a,unsigned __int64 b);
277  unsigned __int64 _umul128 (unsigned __int64 a,unsigned __int64 b,
278  unsigned __int64 *h);
279 # pragma intrinsic(__umulh,_umul128)
280 # define BN_UMULT_HIGH(a,b) __umulh((a),(b))
281 # define BN_UMULT_LOHI(low,high,a,b) ((low)=_umul128((a),(b),&(high)))
282 # endif
283 # elif defined(__mips) && (defined(SIXTY_FOUR_BIT) || defined(SIXTY_FOUR_BIT_LONG))
284 # if defined(__GNUC__) && __GNUC__>=2
285 # define BN_UMULT_HIGH(a,b) ({ \
286  register BN_ULONG ret; \
287  asm ("dmultu %1,%2" \
288  : "=h"(ret) \
289  : "r"(a), "r"(b) : "l"); \
290  ret; })
291 # define BN_UMULT_LOHI(low,high,a,b) \
292  asm ("dmultu %2,%3" \
293  : "=l"(low),"=h"(high) \
294  : "r"(a), "r"(b));
295 # endif
296 # endif /* cpu */
297 #endif /* OPENSSL_NO_ASM */
298 
299 /*************************************************************
300  * Using the long long type
301  */
302 #define Lw(t) (((BN_ULONG)(t))&BN_MASK2)
303 #define Hw(t) (((BN_ULONG)((t)>>BN_BITS2))&BN_MASK2)
304 
305 #ifdef BN_DEBUG_RAND
306 #define bn_clear_top2max(a) \
307  { \
308  int ind = (a)->dmax - (a)->top; \
309  BN_ULONG *ftl = &(a)->d[(a)->top-1]; \
310  for (; ind != 0; ind--) \
311  *(++ftl) = 0x0; \
312  }
313 #else
314 #define bn_clear_top2max(a)
315 #endif
316 
317 #ifdef BN_LLONG
318 #define mul_add(r,a,w,c) { \
319  BN_ULLONG t; \
320  t=(BN_ULLONG)w * (a) + (r) + (c); \
321  (r)= Lw(t); \
322  (c)= Hw(t); \
323  }
324 
325 #define mul(r,a,w,c) { \
326  BN_ULLONG t; \
327  t=(BN_ULLONG)w * (a) + (c); \
328  (r)= Lw(t); \
329  (c)= Hw(t); \
330  }
331 
332 #define sqr(r0,r1,a) { \
333  BN_ULLONG t; \
334  t=(BN_ULLONG)(a)*(a); \
335  (r0)=Lw(t); \
336  (r1)=Hw(t); \
337  }
338 
339 #elif defined(BN_UMULT_LOHI)
340 #define mul_add(r,a,w,c) { \
341  BN_ULONG high,low,ret,tmp=(a); \
342  ret = (r); \
343  BN_UMULT_LOHI(low,high,w,tmp); \
344  ret += (c); \
345  (c) = (ret<(c))?1:0; \
346  (c) += high; \
347  ret += low; \
348  (c) += (ret<low)?1:0; \
349  (r) = ret; \
350  }
351 
352 #define mul(r,a,w,c) { \
353  BN_ULONG high,low,ret,ta=(a); \
354  BN_UMULT_LOHI(low,high,w,ta); \
355  ret = low + (c); \
356  (c) = high; \
357  (c) += (ret<low)?1:0; \
358  (r) = ret; \
359  }
360 
361 #define sqr(r0,r1,a) { \
362  BN_ULONG tmp=(a); \
363  BN_UMULT_LOHI(r0,r1,tmp,tmp); \
364  }
365 
366 #elif defined(BN_UMULT_HIGH)
367 #define mul_add(r,a,w,c) { \
368  BN_ULONG high,low,ret,tmp=(a); \
369  ret = (r); \
370  high= BN_UMULT_HIGH(w,tmp); \
371  ret += (c); \
372  low = (w) * tmp; \
373  (c) = (ret<(c))?1:0; \
374  (c) += high; \
375  ret += low; \
376  (c) += (ret<low)?1:0; \
377  (r) = ret; \
378  }
379 
380 #define mul(r,a,w,c) { \
381  BN_ULONG high,low,ret,ta=(a); \
382  low = (w) * ta; \
383  high= BN_UMULT_HIGH(w,ta); \
384  ret = low + (c); \
385  (c) = high; \
386  (c) += (ret<low)?1:0; \
387  (r) = ret; \
388  }
389 
390 #define sqr(r0,r1,a) { \
391  BN_ULONG tmp=(a); \
392  (r0) = tmp * tmp; \
393  (r1) = BN_UMULT_HIGH(tmp,tmp); \
394  }
395 
396 #else
397 /*************************************************************
398  * No long long type
399  */
400 
401 #define LBITS(a) ((a)&BN_MASK2l)
402 #define HBITS(a) (((a)>>BN_BITS4)&BN_MASK2l)
403 #define L2HBITS(a) (((a)<<BN_BITS4)&BN_MASK2)
404 
405 #define LLBITS(a) ((a)&BN_MASKl)
406 #define LHBITS(a) (((a)>>BN_BITS2)&BN_MASKl)
407 #define LL2HBITS(a) ((BN_ULLONG)((a)&BN_MASKl)<<BN_BITS2)
408 
409 #define mul64(l,h,bl,bh) \
410  { \
411  BN_ULONG m,m1,lt,ht; \
412  \
413  lt=l; \
414  ht=h; \
415  m =(bh)*(lt); \
416  lt=(bl)*(lt); \
417  m1=(bl)*(ht); \
418  ht =(bh)*(ht); \
419  m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS((BN_ULONG)1); \
420  ht+=HBITS(m); \
421  m1=L2HBITS(m); \
422  lt=(lt+m1)&BN_MASK2; if (lt < m1) ht++; \
423  (l)=lt; \
424  (h)=ht; \
425  }
426 
427 #define sqr64(lo,ho,in) \
428  { \
429  BN_ULONG l,h,m; \
430  \
431  h=(in); \
432  l=LBITS(h); \
433  h=HBITS(h); \
434  m =(l)*(h); \
435  l*=l; \
436  h*=h; \
437  h+=(m&BN_MASK2h1)>>(BN_BITS4-1); \
438  m =(m&BN_MASK2l)<<(BN_BITS4+1); \
439  l=(l+m)&BN_MASK2; if (l < m) h++; \
440  (lo)=l; \
441  (ho)=h; \
442  }
443 
444 #define mul_add(r,a,bl,bh,c) { \
445  BN_ULONG l,h; \
446  \
447  h= (a); \
448  l=LBITS(h); \
449  h=HBITS(h); \
450  mul64(l,h,(bl),(bh)); \
451  \
452  /* non-multiply part */ \
453  l=(l+(c))&BN_MASK2; if (l < (c)) h++; \
454  (c)=(r); \
455  l=(l+(c))&BN_MASK2; if (l < (c)) h++; \
456  (c)=h&BN_MASK2; \
457  (r)=l; \
458  }
459 
460 #define mul(r,a,bl,bh,c) { \
461  BN_ULONG l,h; \
462  \
463  h= (a); \
464  l=LBITS(h); \
465  h=HBITS(h); \
466  mul64(l,h,(bl),(bh)); \
467  \
468  /* non-multiply part */ \
469  l+=(c); if ((l&BN_MASK2) < (c)) h++; \
470  (c)=h&BN_MASK2; \
471  (r)=l&BN_MASK2; \
472  }
473 #endif /* !BN_LLONG */
474 
475 #if defined(OPENSSL_DOING_MAKEDEPEND) && defined(OPENSSL_FIPS)
476 #undef bn_div_words
477 #endif
478 
479 void bn_mul_normal(BN_ULONG *r,BN_ULONG *a,int na,BN_ULONG *b,int nb);
480 void bn_mul_comba8(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b);
481 void bn_mul_comba4(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b);
482 void bn_sqr_normal(BN_ULONG *r, const BN_ULONG *a, int n, BN_ULONG *tmp);
483 void bn_sqr_comba8(BN_ULONG *r,const BN_ULONG *a);
484 void bn_sqr_comba4(BN_ULONG *r,const BN_ULONG *a);
485 int bn_cmp_words(const BN_ULONG *a,const BN_ULONG *b,int n);
486 int bn_cmp_part_words(const BN_ULONG *a, const BN_ULONG *b,
487  int cl, int dl);
488 void bn_mul_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2,
489  int dna,int dnb,BN_ULONG *t);
490 void bn_mul_part_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,
491  int n,int tna,int tnb,BN_ULONG *t);
492 void bn_sqr_recursive(BN_ULONG *r,const BN_ULONG *a, int n2, BN_ULONG *t);
493 void bn_mul_low_normal(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b, int n);
494 void bn_mul_low_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2,
495  BN_ULONG *t);
496 void bn_mul_high(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,BN_ULONG *l,int n2,
497  BN_ULONG *t);
498 BN_ULONG bn_add_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
499  int cl, int dl);
500 BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
501  int cl, int dl);
502 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);
503 
504 #ifdef __cplusplus
505 }
506 #endif
507 
508 #endif