X-Git-Url: https://git.openssl.org/gitweb/?a=blobdiff_plain;f=crypto%2Fbn%2Fbn_lcl.h;h=90787e576e88dfae14e7bd40dd469bd36313602d;hb=ad0d2579cf3a293a35a5b606afc5a97c71cf6ca7;hp=9c959921b49198fc7094cb16ed714ef5b38cf0d0;hpb=c9b51693dc8f9f198281a4f8a976cf5e118313b4;p=openssl.git diff --git a/crypto/bn/bn_lcl.h b/crypto/bn/bn_lcl.h index 9c959921b4..90787e576e 100644 --- a/crypto/bn/bn_lcl.h +++ b/crypto/bn/bn_lcl.h @@ -163,6 +163,45 @@ extern "C" { +/* BN_mod_exp_mont_conttime is based on the assumption that the + * L1 data cache line width of the target processor is at least + * the following value. + */ +#define MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH ( 64 ) +#define MOD_EXP_CTIME_MIN_CACHE_LINE_MASK (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - 1) + +/* Window sizes optimized for fixed window size modular exponentiation + * algorithm (BN_mod_exp_mont_consttime). + * + * To achieve the security goals of BN_mode_exp_mont_consttime, the + * maximum size of the window must not exceed + * log_2(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH). + * + * Window size thresholds are defined for cache line sizes of 32 and 64, + * cache line sizes where log_2(32)=5 and log_2(64)=6 respectively. A + * window size of 7 should only be used on processors that have a 128 + * byte or greater cache line size. + */ +#if MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 64 + +# define BN_window_bits_for_ctime_exponent_size(b) \ + ((b) > 937 ? 6 : \ + (b) > 306 ? 5 : \ + (b) > 89 ? 4 : \ + (b) > 22 ? 3 : 1) +# define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (6) + +#elif MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 32 + +# define BN_window_bits_for_ctime_exponent_size(b) \ + ((b) > 306 ? 5 : \ + (b) > 89 ? 4 : \ + (b) > 22 ? 3 : 1) +# define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (5) + +#endif + + /* Pentium pro 16,16,16,32,64 */ /* Alpha 16,16,16,16.64 */ #define BN_MULL_SIZE_NORMAL (16) /* 32 */ @@ -171,7 +210,25 @@ extern "C" { #define BN_MUL_LOW_RECURSIVE_SIZE_NORMAL (32) /* 32 */ #define BN_MONT_CTX_SET_SIZE_WORD (64) /* 32 */ -#if !defined(NO_ASM) && !defined(NO_INLINE_ASM) && !defined(PEDANTIC) +/* 2011-02-22 SMS. + * In various places, a size_t variable or a type cast to size_t was + * used to perform integer-only operations on pointers. This failed on + * VMS with 64-bit pointers (CC /POINTER_SIZE = 64) because size_t is + * still only 32 bits. What's needed in these cases is an integer type + * with the same size as a pointer, which size_t is not certain to be. + * The only fix here is VMS-specific. + */ +#if defined(OPENSSL_SYS_VMS) +# if __INITIAL_POINTER_SIZE == 64 +# define PTR_SIZE_INT long long +# else /* __INITIAL_POINTER_SIZE == 64 */ +# define PTR_SIZE_INT int +# endif /* __INITIAL_POINTER_SIZE == 64 [else] */ +#elif !defined(PTR_SIZE_INT) /* defined(OPENSSL_SYS_VMS) */ +# define PTR_SIZE_INT size_t +#endif /* defined(OPENSSL_SYS_VMS) [else] */ + +#if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) && !defined(PEDANTIC) /* * BN_UMULT_HIGH section. * @@ -199,7 +256,7 @@ extern "C" { # if defined(__DECC) # include # define BN_UMULT_HIGH(a,b) (BN_ULONG)asm("umulh %a0,%a1,%v0",(a),(b)) -# elif defined(__GNUC__) +# elif defined(__GNUC__) && __GNUC__>=2 # define BN_UMULT_HIGH(a,b) ({ \ register BN_ULONG ret; \ asm ("umulh %1,%2,%0" \ @@ -208,7 +265,7 @@ extern "C" { ret; }) # endif /* compiler */ # elif defined(_ARCH_PPC) && defined(__64BIT__) && defined(SIXTY_FOUR_BIT_LONG) -# if defined(__GNUC__) +# if defined(__GNUC__) && __GNUC__>=2 # define BN_UMULT_HIGH(a,b) ({ \ register BN_ULONG ret; \ asm ("mulhdu %0,%1,%2" \ @@ -216,8 +273,62 @@ extern "C" { : "r"(a), "r"(b)); \ ret; }) # endif /* compiler */ +# elif (defined(__x86_64) || defined(__x86_64__)) && \ + (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT)) +# if defined(__GNUC__) && __GNUC__>=2 +# define BN_UMULT_HIGH(a,b) ({ \ + register BN_ULONG ret,discard; \ + asm ("mulq %3" \ + : "=a"(discard),"=d"(ret) \ + : "a"(a), "g"(b) \ + : "cc"); \ + ret; }) +# define BN_UMULT_LOHI(low,high,a,b) \ + asm ("mulq %3" \ + : "=a"(low),"=d"(high) \ + : "a"(a),"g"(b) \ + : "cc"); +# endif +# elif (defined(_M_AMD64) || defined(_M_X64)) && defined(SIXTY_FOUR_BIT) +# if defined(_MSC_VER) && _MSC_VER>=1400 + unsigned __int64 __umulh (unsigned __int64 a,unsigned __int64 b); + unsigned __int64 _umul128 (unsigned __int64 a,unsigned __int64 b, + unsigned __int64 *h); +# pragma intrinsic(__umulh,_umul128) +# define BN_UMULT_HIGH(a,b) __umulh((a),(b)) +# define BN_UMULT_LOHI(low,high,a,b) ((low)=_umul128((a),(b),&(high))) +# endif +# elif defined(__mips) && (defined(SIXTY_FOUR_BIT) || defined(SIXTY_FOUR_BIT_LONG)) +# if defined(__GNUC__) && __GNUC__>=2 +# if __GNUC__>=4 && __GNUC_MINOR__>=4 /* "h" constraint is no more since 4.4 */ +# define BN_UMULT_HIGH(a,b) (((__uint128_t)(a)*(b))>>64) +# define BN_UMULT_LOHI(low,high,a,b) ({ \ + __uint128_t ret=(__uint128_t)(a)*(b); \ + (high)=ret>>64; (low)=ret; }) +# else +# define BN_UMULT_HIGH(a,b) ({ \ + register BN_ULONG ret; \ + asm ("dmultu %1,%2" \ + : "=h"(ret) \ + : "r"(a), "r"(b) : "l"); \ + ret; }) +# define BN_UMULT_LOHI(low,high,a,b)\ + asm ("dmultu %2,%3" \ + : "=l"(low),"=h"(high) \ + : "r"(a), "r"(b)); +# endif +# endif +# elif defined(__aarch64__) && defined(SIXTY_FOUR_BIT_LONG) +# if defined(__GNUC__) && __GNUC__>=2 +# define BN_UMULT_HIGH(a,b) ({ \ + register BN_ULONG ret; \ + asm ("umulh %0,%1,%2" \ + : "=r"(ret) \ + : "r"(a), "r"(b)); \ + ret; }) +# endif # endif /* cpu */ -#endif /* NO_ASM */ +#endif /* OPENSSL_NO_ASM */ /************************************************************* * Using the long long type @@ -225,44 +336,17 @@ extern "C" { #define Lw(t) (((BN_ULONG)(t))&BN_MASK2) #define Hw(t) (((BN_ULONG)((t)>>BN_BITS2))&BN_MASK2) -/* This is used for internal error checking and is not normally used */ -#ifdef BN_DEBUG -# include -# define bn_check_top(a) assert ((a)->top >= 0 && (a)->top <= (a)->dmax); -#else -# define bn_check_top(a) -#endif - -/* This macro is to add extra stuff for development checking */ -#ifdef BN_DEBUG -#define bn_set_max(r) ((r)->max=(r)->top,BN_set_flags((r),BN_FLG_STATIC_DATA)) -#else -#define bn_set_max(r) -#endif - -/* These macros are used to 'take' a section of a bignum for read only use */ -#define bn_set_low(r,a,n) \ +#ifdef BN_DEBUG_RAND +#define bn_clear_top2max(a) \ { \ - (r)->top=((a)->top > (n))?(n):(a)->top; \ - (r)->d=(a)->d; \ - (r)->neg=(a)->neg; \ - (r)->flags|=BN_FLG_STATIC_DATA; \ - bn_set_max(r); \ - } - -#define bn_set_high(r,a,n) \ - { \ - if ((a)->top > (n)) \ - { \ - (r)->top=(a)->top-n; \ - (r)->d= &((a)->d[n]); \ - } \ - else \ - (r)->top=0; \ - (r)->neg=(a)->neg; \ - (r)->flags|=BN_FLG_STATIC_DATA; \ - bn_set_max(r); \ + int ind = (a)->dmax - (a)->top; \ + BN_ULONG *ftl = &(a)->d[(a)->top-1]; \ + for (; ind != 0; ind--) \ + *(++ftl) = 0x0; \ } +#else +#define bn_clear_top2max(a) +#endif #ifdef BN_LLONG #define mul_add(r,a,w,c) { \ @@ -286,6 +370,33 @@ extern "C" { (r1)=Hw(t); \ } +#elif defined(BN_UMULT_LOHI) +#define mul_add(r,a,w,c) { \ + BN_ULONG high,low,ret,tmp=(a); \ + ret = (r); \ + BN_UMULT_LOHI(low,high,w,tmp); \ + ret += (c); \ + (c) = (ret<(c))?1:0; \ + (c) += high; \ + ret += low; \ + (c) += (ret>BN_BITS4)&BN_MASK2l) -#define L2HBITS(a) ((BN_ULONG)((a)&BN_MASK2l)<>BN_BITS2)&BN_MASKl) @@ -339,7 +450,7 @@ extern "C" { lt=(bl)*(lt); \ m1=(bl)*(ht); \ ht =(bh)*(ht); \ - m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS(1L); \ + m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS((BN_ULONG)1); \ ht+=HBITS(m); \ m1=L2HBITS(m); \ lt=(lt+m1)&BN_MASK2; if (lt < m1) ht++; \ @@ -398,19 +509,30 @@ extern "C" { void bn_mul_normal(BN_ULONG *r,BN_ULONG *a,int na,BN_ULONG *b,int nb); void bn_mul_comba8(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b); void bn_mul_comba4(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b); -void bn_sqr_normal(BN_ULONG *r, BN_ULONG *a, int n, BN_ULONG *tmp); -void bn_sqr_comba8(BN_ULONG *r,BN_ULONG *a); -void bn_sqr_comba4(BN_ULONG *r,BN_ULONG *a); -int bn_cmp_words(BN_ULONG *a,BN_ULONG *b,int n); -void bn_mul_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2,BN_ULONG *t); +void bn_sqr_normal(BN_ULONG *r, const BN_ULONG *a, int n, BN_ULONG *tmp); +void bn_sqr_comba8(BN_ULONG *r,const BN_ULONG *a); +void bn_sqr_comba4(BN_ULONG *r,const BN_ULONG *a); +int bn_cmp_words(const BN_ULONG *a,const BN_ULONG *b,int n); +int bn_cmp_part_words(const BN_ULONG *a, const BN_ULONG *b, + int cl, int dl); +void bn_mul_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2, + int dna,int dnb,BN_ULONG *t); void bn_mul_part_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b, - int tn, int n,BN_ULONG *t); -void bn_sqr_recursive(BN_ULONG *r,BN_ULONG *a, int n2, BN_ULONG *t); + int n,int tna,int tnb,BN_ULONG *t); +void bn_sqr_recursive(BN_ULONG *r,const BN_ULONG *a, int n2, BN_ULONG *t); void bn_mul_low_normal(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b, int n); void bn_mul_low_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2, BN_ULONG *t); void bn_mul_high(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,BN_ULONG *l,int n2, BN_ULONG *t); +BN_ULONG bn_add_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, + int cl, int dl); +BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, + int cl, int dl); +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); + +BIGNUM *int_bn_mod_inverse(BIGNUM *in, + const BIGNUM *a, const BIGNUM *n, BN_CTX *ctx, int *noinv); #ifdef __cplusplus }