bn_lcl.h revision 296465
1/* crypto/bn/bn_lcl.h */ 2/* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com) 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 */ 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 */ 111 112#ifndef HEADER_BN_LCL_H 113# define HEADER_BN_LCL_H 114 115# include <openssl/bn.h> 116 117#ifdef __cplusplus 118extern "C" { 119#endif 120 121/*- 122 * BN_window_bits_for_exponent_size -- macro for sliding window mod_exp functions 123 * 124 * 125 * For window size 'w' (w >= 2) and a random 'b' bits exponent, 126 * the number of multiplications is a constant plus on average 127 * 128 * 2^(w-1) + (b-w)/(w+1); 129 * 130 * here 2^(w-1) is for precomputing the table (we actually need 131 * entries only for windows that have the lowest bit set), and 132 * (b-w)/(w+1) is an approximation for the expected number of 133 * w-bit windows, not counting the first one. 134 * 135 * Thus we should use 136 * 137 * w >= 6 if b > 671 138 * w = 5 if 671 > b > 239 139 * w = 4 if 239 > b > 79 140 * w = 3 if 79 > b > 23 141 * w <= 2 if 23 > b 142 * 143 * (with draws in between). Very small exponents are often selected 144 * with low Hamming weight, so we use w = 1 for b <= 23. 145 */ 146# if 1 147# define BN_window_bits_for_exponent_size(b) \ 148 ((b) > 671 ? 6 : \ 149 (b) > 239 ? 5 : \ 150 (b) > 79 ? 4 : \ 151 (b) > 23 ? 3 : 1) 152# else 153/* 154 * Old SSLeay/OpenSSL table. Maximum window size was 5, so this table differs 155 * for b==1024; but it coincides for other interesting values (b==160, 156 * 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 * BN_mod_exp_mont_conttime is based on the assumption that the L1 data cache 166 * line width of the target processor is at least the following value. 167 */ 168# define MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH ( 64 ) 169# define MOD_EXP_CTIME_MIN_CACHE_LINE_MASK (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - 1) 170 171/* 172 * Window sizes optimized for fixed window size modular exponentiation 173 * algorithm (BN_mod_exp_mont_consttime). To achieve the security goals of 174 * BN_mode_exp_mont_consttime, the maximum size of the window must not exceed 175 * log_2(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH). Window size thresholds are 176 * defined for cache line sizes of 32 and 64, cache line sizes where 177 * log_2(32)=5 and log_2(64)=6 respectively. A window size of 7 should only be 178 * used on processors that have a 128 byte or greater cache line size. 179 */ 180# if MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 64 181 182# define BN_window_bits_for_ctime_exponent_size(b) \ 183 ((b) > 937 ? 6 : \ 184 (b) > 306 ? 5 : \ 185 (b) > 89 ? 4 : \ 186 (b) > 22 ? 3 : 1) 187# define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (6) 188 189# elif MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 32 190 191# define BN_window_bits_for_ctime_exponent_size(b) \ 192 ((b) > 306 ? 5 : \ 193 (b) > 89 ? 4 : \ 194 (b) > 22 ? 3 : 1) 195# define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (5) 196 197# endif 198 199/* Pentium pro 16,16,16,32,64 */ 200/* Alpha 16,16,16,16.64 */ 201# define BN_MULL_SIZE_NORMAL (16)/* 32 */ 202# define BN_MUL_RECURSIVE_SIZE_NORMAL (16)/* 32 less than */ 203# define BN_SQR_RECURSIVE_SIZE_NORMAL (16)/* 32 */ 204# define BN_MUL_LOW_RECURSIVE_SIZE_NORMAL (32)/* 32 */ 205# define BN_MONT_CTX_SET_SIZE_WORD (64)/* 32 */ 206 207# if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) && !defined(PEDANTIC) 208/* 209 * BN_UMULT_HIGH section. 210 * 211 * No, I'm not trying to overwhelm you when stating that the 212 * product of N-bit numbers is 2*N bits wide:-) No, I don't expect 213 * you to be impressed when I say that if the compiler doesn't 214 * support 2*N integer type, then you have to replace every N*N 215 * multiplication with 4 (N/2)*(N/2) accompanied by some shifts 216 * and additions which unavoidably results in severe performance 217 * penalties. Of course provided that the hardware is capable of 218 * producing 2*N result... That's when you normally start 219 * considering assembler implementation. However! It should be 220 * pointed out that some CPUs (most notably Alpha, PowerPC and 221 * upcoming IA-64 family:-) provide *separate* instruction 222 * calculating the upper half of the product placing the result 223 * into a general purpose register. Now *if* the compiler supports 224 * inline assembler, then it's not impossible to implement the 225 * "bignum" routines (and have the compiler optimize 'em) 226 * exhibiting "native" performance in C. That's what BN_UMULT_HIGH 227 * macro is about:-) 228 * 229 * <appro@fy.chalmers.se> 230 */ 231# if defined(__alpha) && (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT)) 232# if defined(__DECC) 233# include <c_asm.h> 234# define BN_UMULT_HIGH(a,b) (BN_ULONG)asm("umulh %a0,%a1,%v0",(a),(b)) 235# elif defined(__GNUC__) 236# define BN_UMULT_HIGH(a,b) ({ \ 237 register BN_ULONG ret; \ 238 asm ("umulh %1,%2,%0" \ 239 : "=r"(ret) \ 240 : "r"(a), "r"(b)); \ 241 ret; }) 242# endif /* compiler */ 243# elif defined(_ARCH_PPC) && defined(__64BIT__) && defined(SIXTY_FOUR_BIT_LONG) 244# if defined(__GNUC__) 245# define BN_UMULT_HIGH(a,b) ({ \ 246 register BN_ULONG ret; \ 247 asm ("mulhdu %0,%1,%2" \ 248 : "=r"(ret) \ 249 : "r"(a), "r"(b)); \ 250 ret; }) 251# endif /* compiler */ 252# elif defined(__x86_64) && defined(SIXTY_FOUR_BIT_LONG) 253# if defined(__GNUC__) 254# define BN_UMULT_HIGH(a,b) ({ \ 255 register BN_ULONG ret,discard; \ 256 asm ("mulq %3" \ 257 : "=a"(discard),"=d"(ret) \ 258 : "a"(a), "g"(b) \ 259 : "cc"); \ 260 ret; }) 261# define BN_UMULT_LOHI(low,high,a,b) \ 262 asm ("mulq %3" \ 263 : "=a"(low),"=d"(high) \ 264 : "a"(a),"g"(b) \ 265 : "cc"); 266# endif 267# elif (defined(_M_AMD64) || defined(_M_X64)) && defined(SIXTY_FOUR_BIT) 268# if defined(_MSC_VER) && _MSC_VER>=1400 269unsigned __int64 __umulh(unsigned __int64 a, unsigned __int64 b); 270unsigned __int64 _umul128(unsigned __int64 a, unsigned __int64 b, 271 unsigned __int64 *h); 272# pragma intrinsic(__umulh,_umul128) 273# define BN_UMULT_HIGH(a,b) __umulh((a),(b)) 274# define BN_UMULT_LOHI(low,high,a,b) ((low)=_umul128((a),(b),&(high))) 275# endif 276# endif /* cpu */ 277# endif /* OPENSSL_NO_ASM */ 278 279/************************************************************* 280 * Using the long long type 281 */ 282# define Lw(t) (((BN_ULONG)(t))&BN_MASK2) 283# define Hw(t) (((BN_ULONG)((t)>>BN_BITS2))&BN_MASK2) 284 285# ifdef BN_DEBUG_RAND 286# define bn_clear_top2max(a) \ 287 { \ 288 int ind = (a)->dmax - (a)->top; \ 289 BN_ULONG *ftl = &(a)->d[(a)->top-1]; \ 290 for (; ind != 0; ind--) \ 291 *(++ftl) = 0x0; \ 292 } 293# else 294# define bn_clear_top2max(a) 295# endif 296 297# ifdef BN_LLONG 298# define mul_add(r,a,w,c) { \ 299 BN_ULLONG t; \ 300 t=(BN_ULLONG)w * (a) + (r) + (c); \ 301 (r)= Lw(t); \ 302 (c)= Hw(t); \ 303 } 304 305# define mul(r,a,w,c) { \ 306 BN_ULLONG t; \ 307 t=(BN_ULLONG)w * (a) + (c); \ 308 (r)= Lw(t); \ 309 (c)= Hw(t); \ 310 } 311 312# define sqr(r0,r1,a) { \ 313 BN_ULLONG t; \ 314 t=(BN_ULLONG)(a)*(a); \ 315 (r0)=Lw(t); \ 316 (r1)=Hw(t); \ 317 } 318 319# elif defined(BN_UMULT_LOHI) 320# define mul_add(r,a,w,c) { \ 321 BN_ULONG high,low,ret,tmp=(a); \ 322 ret = (r); \ 323 BN_UMULT_LOHI(low,high,w,tmp); \ 324 ret += (c); \ 325 (c) = (ret<(c))?1:0; \ 326 (c) += high; \ 327 ret += low; \ 328 (c) += (ret<low)?1:0; \ 329 (r) = ret; \ 330 } 331 332# define mul(r,a,w,c) { \ 333 BN_ULONG high,low,ret,ta=(a); \ 334 BN_UMULT_LOHI(low,high,w,ta); \ 335 ret = low + (c); \ 336 (c) = high; \ 337 (c) += (ret<low)?1:0; \ 338 (r) = ret; \ 339 } 340 341# define sqr(r0,r1,a) { \ 342 BN_ULONG tmp=(a); \ 343 BN_UMULT_LOHI(r0,r1,tmp,tmp); \ 344 } 345 346# elif defined(BN_UMULT_HIGH) 347# define mul_add(r,a,w,c) { \ 348 BN_ULONG high,low,ret,tmp=(a); \ 349 ret = (r); \ 350 high= BN_UMULT_HIGH(w,tmp); \ 351 ret += (c); \ 352 low = (w) * tmp; \ 353 (c) = (ret<(c))?1:0; \ 354 (c) += high; \ 355 ret += low; \ 356 (c) += (ret<low)?1:0; \ 357 (r) = ret; \ 358 } 359 360# define mul(r,a,w,c) { \ 361 BN_ULONG high,low,ret,ta=(a); \ 362 low = (w) * ta; \ 363 high= BN_UMULT_HIGH(w,ta); \ 364 ret = low + (c); \ 365 (c) = high; \ 366 (c) += (ret<low)?1:0; \ 367 (r) = ret; \ 368 } 369 370# define sqr(r0,r1,a) { \ 371 BN_ULONG tmp=(a); \ 372 (r0) = tmp * tmp; \ 373 (r1) = BN_UMULT_HIGH(tmp,tmp); \ 374 } 375 376# else 377/************************************************************* 378 * No long long type 379 */ 380 381# define LBITS(a) ((a)&BN_MASK2l) 382# define HBITS(a) (((a)>>BN_BITS4)&BN_MASK2l) 383# define L2HBITS(a) (((a)<<BN_BITS4)&BN_MASK2) 384 385# define LLBITS(a) ((a)&BN_MASKl) 386# define LHBITS(a) (((a)>>BN_BITS2)&BN_MASKl) 387# define LL2HBITS(a) ((BN_ULLONG)((a)&BN_MASKl)<<BN_BITS2) 388 389# define mul64(l,h,bl,bh) \ 390 { \ 391 BN_ULONG m,m1,lt,ht; \ 392 \ 393 lt=l; \ 394 ht=h; \ 395 m =(bh)*(lt); \ 396 lt=(bl)*(lt); \ 397 m1=(bl)*(ht); \ 398 ht =(bh)*(ht); \ 399 m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS((BN_ULONG)1); \ 400 ht+=HBITS(m); \ 401 m1=L2HBITS(m); \ 402 lt=(lt+m1)&BN_MASK2; if (lt < m1) ht++; \ 403 (l)=lt; \ 404 (h)=ht; \ 405 } 406 407# define sqr64(lo,ho,in) \ 408 { \ 409 BN_ULONG l,h,m; \ 410 \ 411 h=(in); \ 412 l=LBITS(h); \ 413 h=HBITS(h); \ 414 m =(l)*(h); \ 415 l*=l; \ 416 h*=h; \ 417 h+=(m&BN_MASK2h1)>>(BN_BITS4-1); \ 418 m =(m&BN_MASK2l)<<(BN_BITS4+1); \ 419 l=(l+m)&BN_MASK2; if (l < m) h++; \ 420 (lo)=l; \ 421 (ho)=h; \ 422 } 423 424# define mul_add(r,a,bl,bh,c) { \ 425 BN_ULONG l,h; \ 426 \ 427 h= (a); \ 428 l=LBITS(h); \ 429 h=HBITS(h); \ 430 mul64(l,h,(bl),(bh)); \ 431 \ 432 /* non-multiply part */ \ 433 l=(l+(c))&BN_MASK2; if (l < (c)) h++; \ 434 (c)=(r); \ 435 l=(l+(c))&BN_MASK2; if (l < (c)) h++; \ 436 (c)=h&BN_MASK2; \ 437 (r)=l; \ 438 } 439 440# define mul(r,a,bl,bh,c) { \ 441 BN_ULONG l,h; \ 442 \ 443 h= (a); \ 444 l=LBITS(h); \ 445 h=HBITS(h); \ 446 mul64(l,h,(bl),(bh)); \ 447 \ 448 /* non-multiply part */ \ 449 l+=(c); if ((l&BN_MASK2) < (c)) h++; \ 450 (c)=h&BN_MASK2; \ 451 (r)=l&BN_MASK2; \ 452 } 453# endif /* !BN_LLONG */ 454 455void bn_mul_normal(BN_ULONG *r, BN_ULONG *a, int na, BN_ULONG *b, int nb); 456void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b); 457void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b); 458void bn_sqr_normal(BN_ULONG *r, const BN_ULONG *a, int n, BN_ULONG *tmp); 459void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a); 460void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a); 461int bn_cmp_words(const BN_ULONG *a, const BN_ULONG *b, int n); 462int bn_cmp_part_words(const BN_ULONG *a, const BN_ULONG *b, int cl, int dl); 463void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2, 464 int dna, int dnb, BN_ULONG *t); 465void bn_mul_part_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, 466 int n, int tna, int tnb, BN_ULONG *t); 467void bn_sqr_recursive(BN_ULONG *r, const BN_ULONG *a, int n2, BN_ULONG *t); 468void bn_mul_low_normal(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n); 469void bn_mul_low_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2, 470 BN_ULONG *t); 471void bn_mul_high(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, BN_ULONG *l, int n2, 472 BN_ULONG *t); 473BN_ULONG bn_add_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, 474 int cl, int dl); 475BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, 476 int cl, int dl); 477int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, 478 const BN_ULONG *np, const BN_ULONG *n0, int num); 479 480#ifdef __cplusplus 481} 482#endif 483 484#endif 485