diff --git a/crypto/cryptlib.c b/crypto/cryptlib.c index 6e45b0d8ba1fa23d45f115a53b7939d76aac9bb6..eb24b27808300c978d041691d5bca98335a40ad1 100644 --- a/crypto/cryptlib.c +++ b/crypto/cryptlib.c @@ -42,12 +42,24 @@ void OPENSSL_cpuid_setup(void) if (!sscanf(env + off, "%lli", (long long *)&vec)) vec = strtoul(env + off, NULL, 0); # endif - if (off) - vec = OPENSSL_ia32_cpuid(OPENSSL_ia32cap_P) & ~vec; - else if (env[0] == ':') + if (off) { + IA32CAP mask = vec; + vec = OPENSSL_ia32_cpuid(OPENSSL_ia32cap_P) & ~mask; + if (mask & (1<<24)) { + /* + * User disables FXSR bit, mask even other capabilities + * that operate exclusively on XMM, so we don't have to + * double-check all the time. We mask PCLMULQDQ, AMD XOP, + * AES-NI and AVX. Formally speaking we don't have to + * do it in x86_64 case, but we can safely assume that + * x86_64 users won't actually flip this flag. + */ + vec &= ~((IA32CAP)(1<<1|1<<11|1<<25|1<<28) << 32); + } + } else if (env[0] == ':') { vec = OPENSSL_ia32_cpuid(OPENSSL_ia32cap_P); + } - OPENSSL_ia32cap_P[2] = 0; if ((env = strchr(env, ':'))) { unsigned int vecx; env++; @@ -57,9 +69,12 @@ void OPENSSL_cpuid_setup(void) OPENSSL_ia32cap_P[2] &= ~vecx; else OPENSSL_ia32cap_P[2] = vecx; + } else { + OPENSSL_ia32cap_P[2] = 0; } - } else + } else { vec = OPENSSL_ia32_cpuid(OPENSSL_ia32cap_P); + } /* * |(1<<10) sets a reserved bit to signal that variable diff --git a/crypto/modes/gcm128.c b/crypto/modes/gcm128.c index 7dead2806d1686164b42a9bd398a4eae537bdfef..c3a9259451e53af77fefbe9202d3aeb5b03daddc 100644 --- a/crypto/modes/gcm128.c +++ b/crypto/modes/gcm128.c @@ -747,8 +747,7 @@ void CRYPTO_gcm128_init(GCM128_CONTEXT *ctx, void *key, block128_f block) # endif # if defined(GHASH_ASM_X86_OR_64) # if !defined(GHASH_ASM_X86) || defined(OPENSSL_IA32_SSE2) - if (OPENSSL_ia32cap_P[0] & (1 << 24) && /* check FXSR bit */ - OPENSSL_ia32cap_P[1] & (1 << 1)) { /* check PCLMULQDQ bit */ + if (OPENSSL_ia32cap_P[1] & (1 << 1)) { /* check PCLMULQDQ bit */ if (((OPENSSL_ia32cap_P[1] >> 22) & 0x41) == 0x41) { /* AVX+MOVBE */ gcm_init_avx(ctx->Htable, ctx->H.u); ctx->gmult = gcm_gmult_avx;