X-Git-Url: https://git.openssl.org/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fbn%2Fbn_exp.c;h=cfbaf2b66fd10f0c6a9c0831eadc25b81543b953;hp=c11e5afd3288fc82b72e656385cdf07c7832c315;hb=ca48ace5c52feb0082954fff00a44e1914b9e7e9;hpb=b6358c89a10128692875fb92921b663c4d079a1e diff --git a/crypto/bn/bn_exp.c b/crypto/bn/bn_exp.c index c11e5afd32..cfbaf2b66f 100644 --- a/crypto/bn/bn_exp.c +++ b/crypto/bn/bn_exp.c @@ -56,7 +56,7 @@ * [including the GNU Public Licence.] */ /* ==================================================================== - * Copyright (c) 1998-2000 The OpenSSL Project. All rights reserved. + * Copyright (c) 1998-2005 The OpenSSL Project. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions @@ -109,10 +109,41 @@ * */ +#define OPENSSL_FIPSAPI #include "cryptlib.h" #include "bn_lcl.h" +#include +#ifdef _WIN32 +# include +# ifndef alloca +# define alloca _alloca +# endif +#elif defined(__GNUC__) +# ifndef alloca +# define alloca(s) __builtin_alloca((s)) +# endif +#elif defined(__sun) +# include +#endif + +#undef RSAZ_ENABLED +#if defined(OPENSSL_BN_ASM_MONT) && \ + (defined(__x86_64) || defined(__x86_64__) || \ + defined(_M_AMD64) || defined(_M_X64)) +# include "rsaz_exp.h" +# define RSAZ_ENABLED +#endif + +#undef SPARC_T4_MONT +#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc)) +# include "sparc_arch.h" +extern unsigned int OPENSSL_sparcv9cap_P[]; +# define SPARC_T4_MONT +#endif + +/* maximum precomputation table size for *variable* sliding windows */ #define TABLE_SIZE 32 /* this one works - simple but works */ @@ -121,12 +152,20 @@ int BN_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, BN_CTX *ctx) int i,bits,ret=0; BIGNUM *v,*rr; + if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) + { + /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ + BNerr(BN_F_BN_EXP,ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); + return -1; + } + BN_CTX_start(ctx); if ((r == a) || (r == p)) rr = BN_CTX_get(ctx); else rr = r; - if ((v = BN_CTX_get(ctx)) == NULL) goto err; + v = BN_CTX_get(ctx); + if (rr == NULL || v == NULL) goto err; if (BN_copy(v,a) == NULL) goto err; bits=BN_num_bits(p); @@ -205,7 +244,7 @@ int BN_mod_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, const BIGNUM *m, if (BN_is_odd(m)) { # ifdef MONT_EXP_WORD - if (a->top == 1 && !a->neg) + if (a->top == 1 && !a->neg && (BN_get_flags(p, BN_FLG_CONSTTIME) == 0)) { BN_ULONG A = a->d[0]; ret=BN_mod_exp_mont_word(r,A,p,m,ctx,NULL); @@ -231,11 +270,19 @@ int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, const BIGNUM *m, BN_CTX *ctx) { int i,j,bits,ret=0,wstart,wend,window,wvalue; - int start=1,ts=0; + int start=1; BIGNUM *aa; - BIGNUM val[TABLE_SIZE]; + /* Table of variables obtained from 'ctx' */ + BIGNUM *val[TABLE_SIZE]; BN_RECP_CTX recp; + if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) + { + /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ + BNerr(BN_F_BN_MOD_EXP_RECP,ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); + return -1; + } + bits=BN_num_bits(p); if (bits == 0) @@ -245,7 +292,9 @@ int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, } BN_CTX_start(ctx); - if ((aa = BN_CTX_get(ctx)) == NULL) goto err; + aa = BN_CTX_get(ctx); + val[0] = BN_CTX_get(ctx); + if(!aa || !val[0]) goto err; BN_RECP_CTX_init(&recp); if (m->neg) @@ -260,11 +309,8 @@ int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, if (BN_RECP_CTX_set(&recp,m,ctx) <= 0) goto err; } - BN_init(&(val[0])); - ts=1; - - if (!BN_nnmod(&(val[0]),a,m,ctx)) goto err; /* 1 */ - if (BN_is_zero(&(val[0]))) + if (!BN_nnmod(val[0],a,m,ctx)) goto err; /* 1 */ + if (BN_is_zero(val[0])) { BN_zero(r); ret = 1; @@ -274,16 +320,16 @@ int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, window = BN_window_bits_for_exponent_size(bits); if (window > 1) { - if (!BN_mod_mul_reciprocal(aa,&(val[0]),&(val[0]),&recp,ctx)) + if (!BN_mod_mul_reciprocal(aa,val[0],val[0],&recp,ctx)) goto err; /* 2 */ j=1<<(window-1); for (i=1; i>1]),&recp,ctx)) + if (!BN_mod_mul_reciprocal(r,r,val[wvalue>>1],&recp,ctx)) goto err; /* move the 'window' down further */ @@ -347,8 +393,6 @@ int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, ret=1; err: BN_CTX_end(ctx); - for (i=0; ineg || BN_ucmp(a,m) >= 0) { - if (!BN_nnmod(&(val[0]),a,m,ctx)) + if (!BN_nnmod(val[0],a,m,ctx)) goto err; - aa= &(val[0]); + aa= val[0]; } else aa=a; @@ -414,20 +462,20 @@ int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, ret = 1; goto err; } - if (!BN_to_montgomery(&(val[0]),aa,mont,ctx)) goto err; /* 1 */ + if (!BN_to_montgomery(val[0],aa,mont,ctx)) goto err; /* 1 */ window = BN_window_bits_for_exponent_size(bits); if (window > 1) { - if (!BN_mod_mul_montgomery(d,&(val[0]),&(val[0]),mont,ctx)) goto err; /* 2 */ + if (!BN_mod_mul_montgomery(d,val[0],val[0],mont,ctx)) goto err; /* 2 */ j=1<<(window-1); for (i=1; itop; /* borrow j */ + if (m->d[j-1] & (((BN_ULONG)1)<<(BN_BITS2-1))) + { + if (bn_wexpand(r,j) == NULL) goto err; + /* 2^(top*BN_BITS2) - m */ + r->d[0] = (0-m->d[0])&BN_MASK2; + for(i=1;id[i] = (~m->d[i])&BN_MASK2; + r->top = j; + } + else +#endif if (!BN_to_montgomery(r,BN_value_one(),mont,ctx)) goto err; for (;;) { @@ -480,7 +540,7 @@ int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, } /* wvalue will be an odd number < 2^window */ - if (!BN_mod_mul_montgomery(r,r,&(val[wvalue>>1]),mont,ctx)) + if (!BN_mod_mul_montgomery(r,r,val[wvalue>>1],mont,ctx)) goto err; /* move the 'window' down further */ @@ -489,17 +549,517 @@ int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, start=0; if (wstart < 0) break; } +#if defined(SPARC_T4_MONT) + if (OPENSSL_sparcv9cap_P[0]&(SPARCV9_VIS3|SPARCV9_PREFER_FPU)) + { + j = mont->N.top; /* borrow j */ + val[0]->d[0] = 1; /* borrow val[0] */ + for (i=1;id[i] = 0; + val[0]->top = j; + if (!BN_mod_mul_montgomery(rr,r,val[0],mont,ctx)) goto err; + } + else +#endif if (!BN_from_montgomery(rr,r,mont,ctx)) goto err; ret=1; err: if ((in_mont == NULL) && (mont != NULL)) BN_MONT_CTX_free(mont); BN_CTX_end(ctx); - for (i=0; i=0 && wordpos < a->top) + { + ret = a->d[wordpos]&BN_MASK2; + if (bitpos) + { + ret >>= bitpos; + if (++wordpos < a->top) + ret |= a->d[wordpos]<<(BN_BITS2-bitpos); + } + } + + return ret&BN_MASK2; +} +#endif + +/* BN_mod_exp_mont_consttime() stores the precomputed powers in a specific layout + * so that accessing any of these table values shows the same access pattern as far + * as cache lines are concerned. The following functions are used to transfer a BIGNUM + * from/to that table. */ + +static int MOD_EXP_CTIME_COPY_TO_PREBUF(const BIGNUM *b, int top, unsigned char *buf, int idx, int width) + { + size_t i, j; + + if (top > b->top) + top = b->top; /* this works because 'buf' is explicitly zeroed */ + for (i = 0, j=idx; i < top * sizeof b->d[0]; i++, j+=width) + { + buf[j] = ((unsigned char*)b->d)[i]; + } + + return 1; + } + +static int MOD_EXP_CTIME_COPY_FROM_PREBUF(BIGNUM *b, int top, unsigned char *buf, int idx, int width) + { + size_t i, j; + + if (bn_wexpand(b, top) == NULL) + return 0; + + for (i=0, j=idx; i < top * sizeof b->d[0]; i++, j+=width) + { + ((unsigned char*)b->d)[i] = buf[j]; + } + + b->top = top; + bn_correct_top(b); + return 1; + } + +/* Given a pointer value, compute the next address that is a cache line multiple. */ +#define MOD_EXP_CTIME_ALIGN(x_) \ + ((unsigned char*)(x_) + (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - (((size_t)(x_)) & (MOD_EXP_CTIME_MIN_CACHE_LINE_MASK)))) + +/* This variant of BN_mod_exp_mont() uses fixed windows and the special + * precomputation memory layout to limit data-dependency to a minimum + * to protect secret exponents (cf. the hyper-threading timing attacks + * pointed out by Colin Percival, + * http://www.daemonology.net/hyperthreading-considered-harmful/) + */ +int BN_mod_exp_mont_consttime(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, + const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont) + { + int i,bits,ret=0,window,wvalue; + int top; + BN_MONT_CTX *mont=NULL; + + int numPowers; + unsigned char *powerbufFree=NULL; + int powerbufLen = 0; + unsigned char *powerbuf=NULL; + BIGNUM tmp, am; +#if defined(SPARC_T4_MONT) + unsigned int t4=0; +#endif + + bn_check_top(a); + bn_check_top(p); + bn_check_top(m); + + top = m->top; + + if (!(m->d[0] & 1)) + { + BNerr(BN_F_BN_MOD_EXP_MONT_CONSTTIME,BN_R_CALLED_WITH_EVEN_MODULUS); + return(0); + } + bits=BN_num_bits(p); + if (bits == 0) + { + ret = BN_one(rr); + return ret; + } + + BN_CTX_start(ctx); + + /* Allocate a montgomery context if it was not supplied by the caller. + * If this is not done, things will break in the montgomery part. + */ + if (in_mont != NULL) + mont=in_mont; + else + { + if ((mont=BN_MONT_CTX_new()) == NULL) goto err; + if (!BN_MONT_CTX_set(mont,m,ctx)) goto err; + } + +#ifdef RSAZ_ENABLED + /* + * If the size of the operands allow it, perform the optimized + * RSAZ exponentiation. For further information see + * crypto/bn/rsaz_exp.c and accompanying assembly modules. + */ + if ((16 == a->top) && (16 == p->top) && (BN_num_bits(m) == 1024) + && rsaz_avx2_eligible()) + { + if (NULL == bn_wexpand(rr, 16)) goto err; + RSAZ_1024_mod_exp_avx2(rr->d, a->d, p->d, m->d, mont->RR.d, mont->n0[0]); + rr->top = 16; + rr->neg = 0; + bn_correct_top(rr); + ret = 1; + goto err; + } + else if ((8 == a->top) && (8 == p->top) && (BN_num_bits(m) == 512)) + { + if (NULL == bn_wexpand(rr,8)) goto err; + RSAZ_512_mod_exp(rr->d, a->d, p->d, m->d, mont->n0[0], mont->RR.d); + rr->top = 8; + rr->neg = 0; + bn_correct_top(rr); + ret = 1; + goto err; + } +#endif + + /* Get the window size to use with size of p. */ + window = BN_window_bits_for_ctime_exponent_size(bits); +#if defined(SPARC_T4_MONT) + if (window>=5 && (top&15)==0 && top<=64 && + (OPENSSL_sparcv9cap_P[1]&(CFR_MONTMUL|CFR_MONTSQR))== + (CFR_MONTMUL|CFR_MONTSQR) && + (t4=OPENSSL_sparcv9cap_P[0])) + window=5; + else +#endif +#if defined(OPENSSL_BN_ASM_MONT5) + if (window==6 && bits<=1024) window=5; /* ~5% improvement of 2048-bit RSA sign */ +#endif + (void)0; + + /* Allocate a buffer large enough to hold all of the pre-computed + * powers of am, am itself and tmp. + */ + numPowers = 1 << window; + powerbufLen = sizeof(m->d[0])*(top*numPowers + + ((2*top)>numPowers?(2*top):numPowers)); +#ifdef alloca + if (powerbufLen < 3072) + powerbufFree = alloca(powerbufLen+MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH); + else +#endif + if ((powerbufFree=(unsigned char*)OPENSSL_malloc(powerbufLen+MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH)) == NULL) + goto err; + + powerbuf = MOD_EXP_CTIME_ALIGN(powerbufFree); + memset(powerbuf, 0, powerbufLen); + +#ifdef alloca + if (powerbufLen < 3072) + powerbufFree = NULL; +#endif + + /* lay down tmp and am right after powers table */ + tmp.d = (BN_ULONG *)(powerbuf + sizeof(m->d[0])*top*numPowers); + am.d = tmp.d + top; + tmp.top = am.top = 0; + tmp.dmax = am.dmax = top; + tmp.neg = am.neg = 0; + tmp.flags = am.flags = BN_FLG_STATIC_DATA; + + /* prepare a^0 in Montgomery domain */ +#if 1 /* by Shay Gueron's suggestion */ + if (m->d[top-1] & (((BN_ULONG)1)<<(BN_BITS2-1))) + { + /* 2^(top*BN_BITS2) - m */ + tmp.d[0] = (0-m->d[0])&BN_MASK2; + for (i=1;id[i])&BN_MASK2; + tmp.top = top; + } + else +#endif + if (!BN_to_montgomery(&tmp,BN_value_one(),mont,ctx)) goto err; + + /* prepare a^1 in Montgomery domain */ + if (a->neg || BN_ucmp(a,m) >= 0) + { + if (!BN_mod(&am,a,m,ctx)) goto err; + if (!BN_to_montgomery(&am,&am,mont,ctx)) goto err; + } + else if (!BN_to_montgomery(&am,a,mont,ctx)) goto err; + +#if defined(SPARC_T4_MONT) + if (t4) + { + typedef int (*bn_pwr5_mont_f)(BN_ULONG *tp,const BN_ULONG *np, + const BN_ULONG *n0,const void *table,int power,int bits); + int bn_pwr5_mont_t4_8(BN_ULONG *tp,const BN_ULONG *np, + const BN_ULONG *n0,const void *table,int power,int bits); + int bn_pwr5_mont_t4_16(BN_ULONG *tp,const BN_ULONG *np, + const BN_ULONG *n0,const void *table,int power,int bits); + int bn_pwr5_mont_t4_24(BN_ULONG *tp,const BN_ULONG *np, + const BN_ULONG *n0,const void *table,int power,int bits); + int bn_pwr5_mont_t4_32(BN_ULONG *tp,const BN_ULONG *np, + const BN_ULONG *n0,const void *table,int power,int bits); + static const bn_pwr5_mont_f pwr5_funcs[4] = { + bn_pwr5_mont_t4_8, bn_pwr5_mont_t4_16, + bn_pwr5_mont_t4_24, bn_pwr5_mont_t4_32 }; + bn_pwr5_mont_f pwr5_worker = pwr5_funcs[top/16-1]; + + typedef int (*bn_mul_mont_f)(BN_ULONG *rp,const BN_ULONG *ap, + const void *bp,const BN_ULONG *np,const BN_ULONG *n0); + int bn_mul_mont_t4_8(BN_ULONG *rp,const BN_ULONG *ap, + const void *bp,const BN_ULONG *np,const BN_ULONG *n0); + int bn_mul_mont_t4_16(BN_ULONG *rp,const BN_ULONG *ap, + const void *bp,const BN_ULONG *np,const BN_ULONG *n0); + int bn_mul_mont_t4_24(BN_ULONG *rp,const BN_ULONG *ap, + const void *bp,const BN_ULONG *np,const BN_ULONG *n0); + int bn_mul_mont_t4_32(BN_ULONG *rp,const BN_ULONG *ap, + const void *bp,const BN_ULONG *np,const BN_ULONG *n0); + static const bn_mul_mont_f mul_funcs[4] = { + bn_mul_mont_t4_8, bn_mul_mont_t4_16, + bn_mul_mont_t4_24, bn_mul_mont_t4_32 }; + bn_mul_mont_f mul_worker = mul_funcs[top/16-1]; + + void bn_mul_mont_vis3(BN_ULONG *rp,const BN_ULONG *ap, + const void *bp,const BN_ULONG *np, + const BN_ULONG *n0,int num); + void bn_mul_mont_t4(BN_ULONG *rp,const BN_ULONG *ap, + const void *bp,const BN_ULONG *np, + const BN_ULONG *n0,int num); + void bn_mul_mont_gather5_t4(BN_ULONG *rp,const BN_ULONG *ap, + const void *table,const BN_ULONG *np, + const BN_ULONG *n0,int num,int power); + void bn_flip_n_scatter5_t4(const BN_ULONG *inp,size_t num, + void *table,size_t power); + void bn_gather5_t4(BN_ULONG *out,size_t num, + void *table,size_t power); + void bn_flip_t4(BN_ULONG *dst,BN_ULONG *src,size_t num); + + BN_ULONG *np=mont->N.d, *n0=mont->n0; + int stride = 5*(6-(top/16-1)); /* multiple of 5, but less than 32 */ + + /* BN_to_montgomery can contaminate words above .top + * [in BN_DEBUG[_DEBUG] build]... */ + for (i=am.top; iN.d,top); + + bits--; + for (wvalue=0, i=bits%5; i>=0; i--,bits--) + wvalue = (wvalue<<1)+BN_is_bit_set(p,bits); + bn_gather5_t4(tmp.d,top,powerbuf,wvalue); + + /* Scan the exponent one window at a time starting from the most + * significant bits. + */ + while (bits >= 0) + { + if (bits < stride) stride = bits+1; + bits -= stride; + wvalue = bn_get_bits(p,bits+1); + + if ((*pwr5_worker)(tmp.d,np,n0,powerbuf,wvalue,stride)) continue; + /* retry once and fall back */ + if ((*pwr5_worker)(tmp.d,np,n0,powerbuf,wvalue,stride)) continue; + + bits += stride-5; + wvalue >>= stride-5; + wvalue &= 31; + bn_mul_mont_t4(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont_t4(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont_t4(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont_t4(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont_t4(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont_gather5_t4(tmp.d,tmp.d,powerbuf,np,n0,top,wvalue); + } + + bn_flip_t4(tmp.d,tmp.d,top); + top *= 2; + /* back to 32-bit domain */ + tmp.top=top; + bn_correct_top(&tmp); + OPENSSL_cleanse(np,top*sizeof(BN_ULONG)); + } + else +#endif +#if defined(OPENSSL_BN_ASM_MONT5) + /* This optimization uses ideas from http://eprint.iacr.org/2011/239, + * specifically optimization of cache-timing attack countermeasures + * and pre-computation optimization. */ + + /* Dedicated window==4 case improves 512-bit RSA sign by ~15%, but as + * 512-bit RSA is hardly relevant, we omit it to spare size... */ + if (window==5) + { + void bn_mul_mont_gather5(BN_ULONG *rp,const BN_ULONG *ap, + const void *table,const BN_ULONG *np, + const BN_ULONG *n0,int num,int power); + void bn_scatter5(const BN_ULONG *inp,size_t num, + void *table,size_t power); + void bn_gather5(BN_ULONG *out,size_t num, + void *table,size_t power); + + BN_ULONG *np=mont->N.d, *n0=mont->n0; + + /* BN_to_montgomery can contaminate words above .top + * [in BN_DEBUG[_DEBUG] build]... */ + for (i=am.top; i=0; i--,bits--) + wvalue = (wvalue<<1)+BN_is_bit_set(p,bits); + bn_gather5(tmp.d,top,powerbuf,wvalue); + + /* Scan the exponent one window at a time starting from the most + * significant bits. + */ + while (bits >= 0) + { + for (wvalue=0, i=0; i<5; i++,bits--) + wvalue = (wvalue<<1)+BN_is_bit_set(p,bits); + + bn_mul_mont(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont(tmp.d,tmp.d,tmp.d,np,n0,top); + bn_mul_mont_gather5(tmp.d,tmp.d,powerbuf,np,n0,top,wvalue); + } + + tmp.top=top; + bn_correct_top(&tmp); + } + else +#endif + { + if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 0, numPowers)) goto err; + if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&am, top, powerbuf, 1, numPowers)) goto err; + + /* If the window size is greater than 1, then calculate + * val[i=2..2^winsize-1]. Powers are computed as a*a^(i-1) + * (even powers could instead be computed as (a^(i/2))^2 + * to use the slight performance advantage of sqr over mul). + */ + if (window > 1) + { + if (!BN_mod_mul_montgomery(&tmp,&am,&am,mont,ctx)) goto err; + if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 2, numPowers)) goto err; + for (i=3; i=0; i--,bits--) + wvalue = (wvalue<<1)+BN_is_bit_set(p,bits); + if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&tmp,top,powerbuf,wvalue,numPowers)) goto err; + + /* Scan the exponent one window at a time starting from the most + * significant bits. + */ + while (bits >= 0) + { + wvalue=0; /* The 'value' of the window */ + + /* Scan the window, squaring the result as we go */ + for (i=0; i 1) { - if (!BN_mod_mul(d,&(val[0]),&(val[0]),m,ctx)) + if (!BN_mod_mul(d,val[0],val[0],m,ctx)) goto err; /* 2 */ j=1<<(window-1); for (i=1; i>1]),m,ctx)) + if (!BN_mod_mul(r,r,val[wvalue>>1],m,ctx)) goto err; /* move the 'window' down further */ @@ -750,9 +1330,6 @@ int BN_mod_exp_simple(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, ret=1; err: BN_CTX_end(ctx); - for (i=0; i