X-Git-Url: https://bearssl.org/gitweb//home/git/?p=BearSSL;a=blobdiff_plain;f=src%2Finner.h;h=47431c00b9411cedc96fbaae05c567da0a1fcb03;hp=e5c45b357a0496a19ef6c9e55bf50129fa534a99;hb=2f88a673049d1e183d1a8f78e2030bf02362b8bf;hpb=3f00688b9d9f483a6ca97e659eea104995ea15b7 diff --git a/src/inner.h b/src/inner.h index e5c45b3..47431c0 100644 --- a/src/inner.h +++ b/src/inner.h @@ -105,6 +105,110 @@ #endif #endif +/* + * Set BR_LOMUL on platforms where it makes sense. + */ +#ifndef BR_LOMUL +#if BR_ARMEL_CORTEX_GCC +#define BR_LOMUL 1 +#endif +#endif + +/* + * Determine whether x86 AES instructions are understood by the compiler. + */ +#ifndef BR_AES_X86NI + +#if (__i386__ || __x86_64__) \ + && ((__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)) \ + || (__clang_major__ > 3 \ + || (__clang_major__ == 3 && __clang_minor__ >= 7))) +#define BR_AES_X86NI 1 +#elif (_M_IX86 || _M_X64) && (_MSC_VER >= 1700) +#define BR_AES_X86NI 1 +#endif +#endif + +/* + * If we use x86 AES instruction, determine the compiler brand. + */ +#if BR_AES_X86NI +#ifndef BR_AES_X86NI_GCC +#if __GNUC__ +#define BR_AES_X86NI_GCC 1 +#endif +#endif +#ifndef BR_AES_X86NI_MSC +#if _MSC_VER >= 1700 +#define BR_AES_X86NI_MSC 1 +#endif +#endif +#endif + +/* + * A macro to tag a function with a "target" attribute (for GCC and Clang). + */ +#if BR_AES_X86NI_GCC +#define BR_TARGET(x) __attribute__((target(x))) +#else +#define BR_TARGET(x) +#endif + +/* + * GCC versions from 4.4 to 4.8 (inclusive) must use a special #pragma + * to activate extra opcodes before including the relevant intrinsic + * headers. But these don't work with Clang (which does not need them + * either). + */ +#if BR_AES_X86NI_GCC && !defined BR_AES_X86NI_GCC_OLD +#if __GNUC__ == 4 && __GNUC_MINOR__ >= 4 && __GNUC_MINOR__ <= 8 && !__clang__ +#define BR_AES_X86NI_GCC_OLD 1 +#endif +#endif + +/* + * POWER8 crypto support. We rely on compiler macros for the + * architecture, since we do not have a reliable, simple way to detect + * the required support at runtime (we could try running an opcode, and + * trapping the exception or signal on illegal instruction, but this + * induces some non-trivial OS dependencies that we would prefer to + * avoid if possible). + */ +#ifndef BR_POWER8 +#if __GNUC__ && ((_ARCH_PWR8 || _ARCH_PPC) && __CRYPTO__) +#define BR_POWER8 1 +#endif +#endif + +/* + * Detect endinanness on POWER8. + */ +#if BR_POWER8 +#if defined BR_POWER8_LE +#undef BR_POWER8_BE +#if BR_POWER8_LE +#define BR_POWER8_BE 0 +#else +#define BR_POWER8_BE 1 +#endif +#elif defined BR_POWER8_BE +#undef BR_POWER8_LE +#if BR_POWER8_BE +#define BR_POWER8_LE 0 +#else +#define BR_POWER8_LE 1 +#endif +#else +#if __LITTLE_ENDIAN__ +#define BR_POWER8_LE 1 +#define BR_POWER8_BE 0 +#else +#define BR_POWER8_LE 0 +#define BR_POWER8_BE 1 +#endif +#endif +#endif + /* ==================================================================== */ /* * Encoding/decoding functions. @@ -1077,6 +1181,10 @@ void br_i31_mulacc(uint32_t *d, const uint32_t *a, const uint32_t *b); /* ==================================================================== */ +/* + * FIXME: document "i15" functions. + */ + static inline void br_i15_zero(uint16_t *x, uint16_t bit_len) { @@ -1102,6 +1210,9 @@ void br_i15_to_monty(uint16_t *x, const uint16_t *m); void br_i15_modpow(uint16_t *x, const unsigned char *e, size_t elen, const uint16_t *m, uint16_t m0i, uint16_t *t1, uint16_t *t2); +uint32_t br_i15_modpow_opt(uint16_t *x, const unsigned char *e, size_t elen, + const uint16_t *m, uint16_t m0i, uint16_t *tmp, size_t twlen); + void br_i15_encode(void *dst, size_t len, const uint16_t *x); uint32_t br_i15_decode_mod(uint16_t *x, @@ -1421,6 +1532,40 @@ unsigned br_aes_ct64_keysched(uint64_t *comp_skey, void br_aes_ct64_skey_expand(uint64_t *skey, unsigned num_rounds, const uint64_t *comp_skey); +/* + * Test support for AES-NI opcodes. + */ +int br_aes_x86ni_supported(void); + +/* + * AES key schedule, using x86 AES-NI instructions. This yields the + * subkeys in the encryption direction. Number of rounds is returned. + * Key size MUST be 16, 24 or 32 bytes; otherwise, 0 is returned. + */ +unsigned br_aes_x86ni_keysched_enc(unsigned char *skni, + const void *key, size_t len); + +/* + * AES key schedule, using x86 AES-NI instructions. This yields the + * subkeys in the decryption direction. Number of rounds is returned. + * Key size MUST be 16, 24 or 32 bytes; otherwise, 0 is returned. + */ +unsigned br_aes_x86ni_keysched_dec(unsigned char *skni, + const void *key, size_t len); + +/* + * Test support for AES POWER8 opcodes. + */ +int br_aes_pwr8_supported(void); + +/* + * AES key schedule, using POWER8 instructions. This yields the + * subkeys in the encryption direction. Number of rounds is returned. + * Key size MUST be 16, 24 or 32 bytes; otherwise, 0 is returned. + */ +unsigned br_aes_pwr8_keysched(unsigned char *skni, + const void *key, size_t len); + /* ==================================================================== */ /* * RSA. @@ -1698,4 +1843,85 @@ int br_ssl_choose_hash(unsigned bf); /* ==================================================================== */ +/* + * PowerPC / POWER assembly stuff. The special BR_POWER_ASM_MACROS macro + * must be defined before including this file; this is done by source + * files that use some inline assembly for PowerPC / POWER machines. + */ + +#if BR_POWER_ASM_MACROS + +#define lxvw4x(xt, ra, rb) lxvw4x_(xt, ra, rb) +#define stxvw4x(xt, ra, rb) stxvw4x_(xt, ra, rb) + +#define bdnz(foo) bdnz_(foo) +#define beq(foo) beq_(foo) + +#define li(rx, value) li_(rx, value) +#define addi(rx, ra, imm) addi_(rx, ra, imm) +#define cmpldi(rx, imm) cmpldi_(rx, imm) +#define mtctr(rx) mtctr_(rx) +#define vspltb(vrt, vrb, uim) vspltb_(vrt, vrb, uim) +#define vspltw(vrt, vrb, uim) vspltw_(vrt, vrb, uim) +#define vspltisb(vrt, imm) vspltisb_(vrt, imm) +#define vspltisw(vrt, imm) vspltisw_(vrt, imm) +#define vrlw(vrt, vra, vrb) vrlw_(vrt, vra, vrb) +#define vsbox(vrt, vra) vsbox_(vrt, vra) +#define vxor(vrt, vra, vrb) vxor_(vrt, vra, vrb) +#define vand(vrt, vra, vrb) vand_(vrt, vra, vrb) +#define vsro(vrt, vra, vrb) vsro_(vrt, vra, vrb) +#define vsl(vrt, vra, vrb) vsl_(vrt, vra, vrb) +#define vsldoi(vt, va, vb, sh) vsldoi_(vt, va, vb, sh) +#define vsr(vrt, vra, vrb) vsr_(vrt, vra, vrb) +#define vadduwm(vrt, vra, vrb) vadduwm_(vrt, vra, vrb) +#define vsububm(vrt, vra, vrb) vsububm_(vrt, vra, vrb) +#define vsubuwm(vrt, vra, vrb) vsubuwm_(vrt, vra, vrb) +#define vsrw(vrt, vra, vrb) vsrw_(vrt, vra, vrb) +#define vcipher(vt, va, vb) vcipher_(vt, va, vb) +#define vcipherlast(vt, va, vb) vcipherlast_(vt, va, vb) +#define vncipher(vt, va, vb) vncipher_(vt, va, vb) +#define vncipherlast(vt, va, vb) vncipherlast_(vt, va, vb) +#define vperm(vt, va, vb, vc) vperm_(vt, va, vb, vc) +#define vpmsumd(vt, va, vb) vpmsumd_(vt, va, vb) +#define xxpermdi(vt, va, vb, d) xxpermdi_(vt, va, vb, d) + +#define lxvw4x_(xt, ra, rb) "\tlxvw4x\t" #xt "," #ra "," #rb "\n" +#define stxvw4x_(xt, ra, rb) "\tstxvw4x\t" #xt "," #ra "," #rb "\n" + +#define label(foo) #foo "%=:\n" +#define bdnz_(foo) "\tbdnz\t" #foo "%=\n" +#define beq_(foo) "\tbeq\t" #foo "%=\n" + +#define li_(rx, value) "\tli\t" #rx "," #value "\n" +#define addi_(rx, ra, imm) "\taddi\t" #rx "," #ra "," #imm "\n" +#define cmpldi_(rx, imm) "\tcmpldi\t" #rx "," #imm "\n" +#define mtctr_(rx) "\tmtctr\t" #rx "\n" +#define vspltb_(vrt, vrb, uim) "\tvspltb\t" #vrt "," #vrb "," #uim "\n" +#define vspltw_(vrt, vrb, uim) "\tvspltw\t" #vrt "," #vrb "," #uim "\n" +#define vspltisb_(vrt, imm) "\tvspltisb\t" #vrt "," #imm "\n" +#define vspltisw_(vrt, imm) "\tvspltisw\t" #vrt "," #imm "\n" +#define vrlw_(vrt, vra, vrb) "\tvrlw\t" #vrt "," #vra "," #vrb "\n" +#define vsbox_(vrt, vra) "\tvsbox\t" #vrt "," #vra "\n" +#define vxor_(vrt, vra, vrb) "\tvxor\t" #vrt "," #vra "," #vrb "\n" +#define vand_(vrt, vra, vrb) "\tvand\t" #vrt "," #vra "," #vrb "\n" +#define vsro_(vrt, vra, vrb) "\tvsro\t" #vrt "," #vra "," #vrb "\n" +#define vsl_(vrt, vra, vrb) "\tvsl\t" #vrt "," #vra "," #vrb "\n" +#define vsldoi_(vt, va, vb, sh) "\tvsldoi\t" #vt "," #va "," #vb "," #sh "\n" +#define vsr_(vrt, vra, vrb) "\tvsr\t" #vrt "," #vra "," #vrb "\n" +#define vadduwm_(vrt, vra, vrb) "\tvadduwm\t" #vrt "," #vra "," #vrb "\n" +#define vsububm_(vrt, vra, vrb) "\tvsububm\t" #vrt "," #vra "," #vrb "\n" +#define vsubuwm_(vrt, vra, vrb) "\tvsubuwm\t" #vrt "," #vra "," #vrb "\n" +#define vsrw_(vrt, vra, vrb) "\tvsrw\t" #vrt "," #vra "," #vrb "\n" +#define vcipher_(vt, va, vb) "\tvcipher\t" #vt "," #va "," #vb "\n" +#define vcipherlast_(vt, va, vb) "\tvcipherlast\t" #vt "," #va "," #vb "\n" +#define vncipher_(vt, va, vb) "\tvncipher\t" #vt "," #va "," #vb "\n" +#define vncipherlast_(vt, va, vb) "\tvncipherlast\t" #vt "," #va "," #vb "\n" +#define vperm_(vt, va, vb, vc) "\tvperm\t" #vt "," #va "," #vb "," #vc "\n" +#define vpmsumd_(vt, va, vb) "\tvpmsumd\t" #vt "," #va "," #vb "\n" +#define xxpermdi_(vt, va, vb, d) "\txxpermdi\t" #vt "," #va "," #vb "," #d "\n" + +#endif + +/* ==================================================================== */ + #endif