X-Git-Url: https://git.openssl.org/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fppccap.c;h=9601674041020e2cb65d725ecb771a7f1a53ff3a;hp=85901598ddef0f2fd0ed08ca9899fde9eadf93fe;hb=a28e4890eed847e6122a1c4d50653566e0813f45;hpb=d741cf2267eca194817d300912b35da02806ca3e diff --git a/crypto/ppccap.c b/crypto/ppccap.c index 85901598dd..9601674041 100644 --- a/crypto/ppccap.c +++ b/crypto/ppccap.c @@ -1,91 +1,417 @@ +/* + * Copyright 2009-2018 The OpenSSL Project Authors. All Rights Reserved. + * + * Licensed under the Apache License 2.0 (the "License"). You may not use + * this file except in compliance with the License. You can obtain a copy + * in the file LICENSE in the source distribution or at + * https://www.openssl.org/source/license.html + */ + #include #include #include #include #include +#include +#if defined(__linux) || defined(_AIX) +# include +#endif +#if defined(_AIX53) /* defined even on post-5.3 */ +# include +# if !defined(__power_set) +# define __power_set(a) (_system_configuration.implementation & (a)) +# endif +#endif +#if defined(__APPLE__) && defined(__MACH__) +# include +# include +#endif +#include #include +#include +#include +#include "bn/bn_lcl.h" -#define PPC_FPU64 (1<<0) +#include "ppc_arch.h" -static int OPENSSL_ppccap_P = 0; +unsigned int OPENSSL_ppccap_P = 0; static sigset_t all_masked; #ifdef OPENSSL_BN_ASM_MONT -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) - { - int bn_mul_mont_fpu64(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np, const BN_ULONG *n0, int num); - int bn_mul_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np, const BN_ULONG *n0, int num); - - if (sizeof(size_t)==4) - { -#if (defined(__APPLE__) && defined(__MACH__)) - if ((OPENSSL_ppccap_P&PPC_FPU64)) - return bn_mul_mont_fpu64(rp,ap,bp,np,n0,num); -#else - /* boundary of 32 was experimentally determined on - Linux 2.6.22, might have to be adjusted on AIX... */ - if ((num>=32) && (OPENSSL_ppccap_P&PPC_FPU64)) - { - sigset_t oset; - int ret; - - sigprocmask(SIG_SETMASK,&all_masked,&oset); - ret=bn_mul_mont_fpu64(rp,ap,bp,np,n0,num); - sigprocmask(SIG_SETMASK,&oset,NULL); - - return ret; - } +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) +{ + int bn_mul_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, + const BN_ULONG *np, const BN_ULONG *n0, int num); + int bn_mul4x_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, + const BN_ULONG *np, const BN_ULONG *n0, int num); + + if (num < 4) + return 0; + + if ((num & 3) == 0) + return bn_mul4x_mont_int(rp, ap, bp, np, n0, num); + + /* + * There used to be [optional] call to bn_mul_mont_fpu64 here, + * but above subroutine is faster on contemporary processors. + * Formulation means that there might be old processors where + * FPU code path would be faster, POWER6 perhaps, but there was + * no opportunity to figure it out... + */ + + return bn_mul_mont_int(rp, ap, bp, np, n0, num); +} #endif - } - else if ((OPENSSL_ppccap_P&PPC_FPU64)) - /* this is a "must" on Power 6, but run-time detection - * is not implemented yet... */ - return bn_mul_mont_fpu64(rp,ap,bp,np,n0,num); - - return bn_mul_mont_int(rp,ap,bp,np,n0,num); - } + +void sha256_block_p8(void *ctx, const void *inp, size_t len); +void sha256_block_ppc(void *ctx, const void *inp, size_t len); +void sha256_block_data_order(void *ctx, const void *inp, size_t len); +void sha256_block_data_order(void *ctx, const void *inp, size_t len) +{ + OPENSSL_ppccap_P & PPC_CRYPTO207 ? sha256_block_p8(ctx, inp, len) : + sha256_block_ppc(ctx, inp, len); +} + +void sha512_block_p8(void *ctx, const void *inp, size_t len); +void sha512_block_ppc(void *ctx, const void *inp, size_t len); +void sha512_block_data_order(void *ctx, const void *inp, size_t len); +void sha512_block_data_order(void *ctx, const void *inp, size_t len) +{ + OPENSSL_ppccap_P & PPC_CRYPTO207 ? sha512_block_p8(ctx, inp, len) : + sha512_block_ppc(ctx, inp, len); +} + +#ifndef OPENSSL_NO_CHACHA +void ChaCha20_ctr32_int(unsigned char *out, const unsigned char *inp, + size_t len, const unsigned int key[8], + const unsigned int counter[4]); +void ChaCha20_ctr32_vmx(unsigned char *out, const unsigned char *inp, + size_t len, const unsigned int key[8], + const unsigned int counter[4]); +void ChaCha20_ctr32_vsx(unsigned char *out, const unsigned char *inp, + size_t len, const unsigned int key[8], + const unsigned int counter[4]); +void ChaCha20_ctr32(unsigned char *out, const unsigned char *inp, + size_t len, const unsigned int key[8], + const unsigned int counter[4]) +{ + OPENSSL_ppccap_P & PPC_CRYPTO207 + ? ChaCha20_ctr32_vsx(out, inp, len, key, counter) + : OPENSSL_ppccap_P & PPC_ALTIVEC + ? ChaCha20_ctr32_vmx(out, inp, len, key, counter) + : ChaCha20_ctr32_int(out, inp, len, key, counter); +} +#endif + +#ifndef OPENSSL_NO_POLY1305 +void poly1305_init_int(void *ctx, const unsigned char key[16]); +void poly1305_blocks(void *ctx, const unsigned char *inp, size_t len, + unsigned int padbit); +void poly1305_emit(void *ctx, unsigned char mac[16], + const unsigned int nonce[4]); +void poly1305_init_fpu(void *ctx, const unsigned char key[16]); +void poly1305_blocks_fpu(void *ctx, const unsigned char *inp, size_t len, + unsigned int padbit); +void poly1305_emit_fpu(void *ctx, unsigned char mac[16], + const unsigned int nonce[4]); +void poly1305_init_vsx(void *ctx, const unsigned char key[16]); +void poly1305_blocks_vsx(void *ctx, const unsigned char *inp, size_t len, + unsigned int padbit); +void poly1305_emit_vsx(void *ctx, unsigned char mac[16], + const unsigned int nonce[4]); +int poly1305_init(void *ctx, const unsigned char key[16], void *func[2]); +int poly1305_init(void *ctx, const unsigned char key[16], void *func[2]) +{ + if (OPENSSL_ppccap_P & PPC_CRYPTO207) { + poly1305_init_int(ctx, key); + func[0] = (void*)(uintptr_t)poly1305_blocks_vsx; + func[1] = (void*)(uintptr_t)poly1305_emit; + } else if (sizeof(size_t) == 4 && (OPENSSL_ppccap_P & PPC_FPU)) { + poly1305_init_fpu(ctx, key); + func[0] = (void*)(uintptr_t)poly1305_blocks_fpu; + func[1] = (void*)(uintptr_t)poly1305_emit_fpu; + } else { + poly1305_init_int(ctx, key); + func[0] = (void*)(uintptr_t)poly1305_blocks; + func[1] = (void*)(uintptr_t)poly1305_emit; + } + return 1; +} +#endif + +#ifdef ECP_NISTZ256_ASM +void ecp_nistz256_mul_mont(unsigned long res[4], const unsigned long a[4], + const unsigned long b[4]); + +void ecp_nistz256_to_mont(unsigned long res[4], const unsigned long in[4]); +void ecp_nistz256_to_mont(unsigned long res[4], const unsigned long in[4]) +{ + static const unsigned long RR[] = { 0x0000000000000003U, + 0xfffffffbffffffffU, + 0xfffffffffffffffeU, + 0x00000004fffffffdU }; + + ecp_nistz256_mul_mont(res, in, RR); +} + +void ecp_nistz256_from_mont(unsigned long res[4], const unsigned long in[4]); +void ecp_nistz256_from_mont(unsigned long res[4], const unsigned long in[4]) +{ + static const unsigned long one[] = { 1, 0, 0, 0 }; + + ecp_nistz256_mul_mont(res, in, one); +} #endif static sigjmp_buf ill_jmp; -static void ill_handler (int sig) { siglongjmp(ill_jmp,sig); } +static void ill_handler(int sig) +{ + siglongjmp(ill_jmp, sig); +} + +void OPENSSL_fpu_probe(void); +void OPENSSL_ppc64_probe(void); +void OPENSSL_altivec_probe(void); +void OPENSSL_crypto207_probe(void); +void OPENSSL_madd300_probe(void); + +long OPENSSL_rdtsc_mftb(void); +long OPENSSL_rdtsc_mfspr268(void); + +uint32_t OPENSSL_rdtsc(void) +{ + if (OPENSSL_ppccap_P & PPC_MFTB) + return OPENSSL_rdtsc_mftb(); + else if (OPENSSL_ppccap_P & PPC_MFSPR268) + return OPENSSL_rdtsc_mfspr268(); + else + return 0; +} + +size_t OPENSSL_instrument_bus_mftb(unsigned int *, size_t); +size_t OPENSSL_instrument_bus_mfspr268(unsigned int *, size_t); + +size_t OPENSSL_instrument_bus(unsigned int *out, size_t cnt) +{ + if (OPENSSL_ppccap_P & PPC_MFTB) + return OPENSSL_instrument_bus_mftb(out, cnt); + else if (OPENSSL_ppccap_P & PPC_MFSPR268) + return OPENSSL_instrument_bus_mfspr268(out, cnt); + else + return 0; +} + +size_t OPENSSL_instrument_bus2_mftb(unsigned int *, size_t, size_t); +size_t OPENSSL_instrument_bus2_mfspr268(unsigned int *, size_t, size_t); +size_t OPENSSL_instrument_bus2(unsigned int *out, size_t cnt, size_t max) +{ + if (OPENSSL_ppccap_P & PPC_MFTB) + return OPENSSL_instrument_bus2_mftb(out, cnt, max); + else if (OPENSSL_ppccap_P & PPC_MFSPR268) + return OPENSSL_instrument_bus2_mfspr268(out, cnt, max); + else + return 0; +} + +#if defined(__GLIBC__) && defined(__GLIBC_PREREQ) +# if __GLIBC_PREREQ(2, 16) +# include +# define OSSL_IMPLEMENT_GETAUXVAL +# endif +#endif + +/* I wish was universally available */ +#define HWCAP 16 /* AT_HWCAP */ +#define HWCAP_PPC64 (1U << 30) +#define HWCAP_ALTIVEC (1U << 28) +#define HWCAP_FPU (1U << 27) +#define HWCAP_POWER6_EXT (1U << 9) +#define HWCAP_VSX (1U << 7) + +#define HWCAP2 26 /* AT_HWCAP2 */ +#define HWCAP_VEC_CRYPTO (1U << 25) +#define HWCAP_ARCH_3_00 (1U << 23) + +# if defined(__GNUC__) && __GNUC__>=2 +__attribute__ ((constructor)) +# endif void OPENSSL_cpuid_setup(void) - { - char *e; - - sigfillset(&all_masked); - sigdelset(&all_masked,SIGSEGV); - sigdelset(&all_masked,SIGILL); - sigdelset(&all_masked,SIGBUS); - sigdelset(&all_masked,SIGFPE); - - if ((e=getenv("OPENSSL_ppccap"))) - { - OPENSSL_ppccap_P=strtoul(e,NULL,0); - return; - } - - if (sizeof(size_t)==4) - { - struct sigaction ill_oact,ill_act; - sigset_t oset; - - memset(&ill_act,0,sizeof(ill_act)); - ill_act.sa_handler = ill_handler; - ill_act.sa_mask = all_masked; - sigprocmask(SIG_SETMASK,&ill_act.sa_mask,&oset); - sigaction (SIGILL,&ill_act,&ill_oact); - if (sigsetjmp(ill_jmp,0) == 0) - { - OPENSSL_ppc64_probe(); - OPENSSL_ppccap_P |= PPC_FPU64; - } - else - { - OPENSSL_ppccap_P &= ~PPC_FPU64; - } - sigaction (SIGILL,&ill_oact,NULL); - sigprocmask(SIG_SETMASK,&oset,NULL); - } - } +{ + char *e; + struct sigaction ill_oact, ill_act; + sigset_t oset; + static int trigger = 0; + + if (trigger) + return; + trigger = 1; + + if ((e = getenv("OPENSSL_ppccap"))) { + OPENSSL_ppccap_P = strtoul(e, NULL, 0); + return; + } + + OPENSSL_ppccap_P = 0; + +#if defined(_AIX) + OPENSSL_ppccap_P |= PPC_FPU; + + if (sizeof(size_t) == 4) { + struct utsname uts; +# if defined(_SC_AIX_KERNEL_BITMODE) + if (sysconf(_SC_AIX_KERNEL_BITMODE) != 64) + return; +# endif + if (uname(&uts) != 0 || atoi(uts.version) < 6) + return; + } + +# if defined(__power_set) + /* + * Value used in __power_set is a single-bit 1<