MDEV-19684 enable intel assembly (AESNI etc) and fastmath when compiling WolfSSL

Using different recommended speedup options for WolfSSL.

- Enable  x64 assembly code on Intel.
- in my_crypt.cc, align EVP_CIPHER_CTX buffer, since some members need
alignment of 16 (for AESNI instructions), when assembler is enabled.
- Adjust MY_AES_CTX_SIZE
- Enable fastmath in wolfssl (large integer math).

extra/wolfssl/CMakeLists.txt

@@ -1,3 +1,26 @@
+IF(MSVC)
+  PROJECT(wolfssl C ASM_MASM)
+ELSEIF(CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
+   PROJECT(wolfssl C ASM)
+ELSE()
+  PROJECT(wolfssl C)
+ENDIF()
+
+IF(MSVC AND (CMAKE_SIZEOF_VOID_P MATCHES 8))
+  SET(WOLFSSL_INTELASM ON)
+ELSEIF(CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
+  IF(CMAKE_C_COMPILER_ID MATCHES GNU  AND CMAKE_C_COMPILER_VERSION VERSION_LESS 4.9)
+    MESSAGE_ONCE(NO_INTEL_ASSEMBLY "Disable Intel assembly for WolfSSL - compiler is too old")
+  ELSE()
+    MY_CHECK_C_COMPILER_FLAG(-maes)
+    MY_CHECK_C_COMPILER_FLAG(-msse4)
+    MY_CHECK_C_COMPILER_FLAG(-mpclmul)
+  ENDIF()
+  IF(have_C__maes AND have_C__msse4 AND have_C__mpclmul)
+    SET(WOLFSSL_INTELASM ON)
+  ENDIF()
+ENDIF()
+
 SET(WOLFSSL_SRCDIR ${CMAKE_CURRENT_SOURCE_DIR}/wolfssl/src)
 ADD_DEFINITIONS(${SSL_DEFINES})
 ADD_DEFINITIONS(
@@ -32,7 +55,7 @@ ADD_DEFINITIONS(-DWOLFSSL_LIB)
 INCLUDE_DIRECTORIES(BEFORE ${CMAKE_CURRENT_SOURCE_DIR}/wolfssl)
 IF(MSVC)
   # size_t to long truncation warning
-  SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -wd4267")
+  SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -wd4267 -wd4334")
   IF(CMAKE_C_COMPILER_ID MATCHES Clang)
     # Workaround a bug with clang-cl, see https://github.com/wolfSSL/wolfssl/pull/2090
     ADD_DEFINITIONS(-DMP_16BIT)
@@ -66,7 +89,6 @@ ${WOLFCRYPT_SRCDIR}/dsa.c
 ${WOLFCRYPT_SRCDIR}/ecc.c
 ${WOLFCRYPT_SRCDIR}/error.c
 ${WOLFCRYPT_SRCDIR}/hmac.c
-${WOLFCRYPT_SRCDIR}/integer.c
 ${WOLFCRYPT_SRCDIR}/logging.c
 ${WOLFCRYPT_SRCDIR}/md4.c
 ${WOLFCRYPT_SRCDIR}/md5.c
@@ -85,5 +107,42 @@ ${WOLFCRYPT_SRCDIR}/hash.c
 ${WOLFCRYPT_SRCDIR}/wolfmath.c
 )
 
+# Use fastmath large number math library.
+IF(NOT (MSVC AND CMAKE_C_COMPILER_ID MATCHES Clang)
+   AND (CMAKE_SIZEOF_VOID_P EQUAL 8))
+  SET(WOLFSSL_FASTMATH 1)
+ENDIF()
+
+IF(WOLFSSL_FASTMATH)
+ ADD_DEFINITIONS(-DUSE_FAST_MATH)
+ # FP_MAX_BITS is set high solely to satisfy ssl_8k_key.test
+ # WolfSSL will use more stack space with it
+ ADD_DEFINITIONS(-DFP_MAX_BITS=16384)
+ SET(WOLFCRYPT_SOURCES  ${WOLFCRYPT_SOURCES} ${WOLFCRYPT_SRCDIR}/tfm.c)
+ELSE()
+ SET(WOLFCRYPT_SOURCES  ${WOLFCRYPT_SOURCES} ${WOLFCRYPT_SRCDIR}/integer.c)
+ENDIF()
+
+IF(WOLFSSL_INTELASM)
+  ADD_DEFINITIONS(-DWOLFSSL_AESNI)
+  SET(SSL_DEFINES "${SSL_DEFINES} -DWOLFSSL_AESNI" PARENT_SCOPE)
+  LIST(APPEND WOLFCRYPT_SOURCES ${WOLFCRYPT_SRCDIR}/cpuid.c)
+  IF(MSVC)
+    LIST(APPEND WOLFCRYPT_SOURCES ${WOLFCRYPT_SRCDIR}/aes_asm.asm)
+    IF(CMAKE_C_COMPILER_ID MATCHES Clang)
+      SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -maes")
+    ELSE()
+      ADD_DEFINITIONS(-DHAVE_INTEL_RDSEED -DWOLFSSL_X86_64_BUILD)
+    ENDIF()
+  ELSEIF(CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
+    ADD_DEFINITIONS(-DHAVE_INTEL_RDSEED -DUSE_INTEL_SPEEDUP)
+    LIST(APPEND WOLFCRYPT_SOURCES
+        ${WOLFCRYPT_SRCDIR}/aes_asm.S
+        ${WOLFCRYPT_SRCDIR}/sha512_asm.S
+        ${WOLFCRYPT_SRCDIR}/sha256_asm.S)
+    ADD_DEFINITIONS(-maes -msse4 -mpclmul)
+  ENDIF()
+ENDIF()
+
 ADD_CONVENIENCE_LIBRARY(wolfcrypt ${WOLFCRYPT_SOURCES})
 

include/mysql/service_my_crypt.h

@@ -45,7 +45,7 @@ extern "C" {
 /* The max key length of all supported algorithms */
 #define MY_AES_MAX_KEY_LENGTH 32
 
-#define MY_AES_CTX_SIZE 560
+#define MY_AES_CTX_SIZE 600
 
 enum my_aes_mode {
     MY_AES_ECB, MY_AES_CBC

mysys_ssl/my_crypt.cc

@@ -25,16 +25,28 @@
 
 #include <my_crypt.h>
 #include <ssl_compat.h>
+#include <cstdint>
+
+#ifdef HAVE_WOLFSSL
+#define CTX_ALIGN 16
+#else
+#define CTX_ALIGN 0
+#endif
 
 class MyCTX
 {
 public:
-  char ctx_buf[EVP_CIPHER_CTX_SIZE];
-  EVP_CIPHER_CTX *ctx;
-
+  char ctx_buf[EVP_CIPHER_CTX_SIZE + CTX_ALIGN];
+  EVP_CIPHER_CTX* ctx;
   MyCTX()
   {
-    ctx= (EVP_CIPHER_CTX *)ctx_buf;
+#if CTX_ALIGN > 0
+    uintptr_t p= ((uintptr_t)ctx_buf + (CTX_ALIGN - 1)) & ~(CTX_ALIGN - 1);
+    ctx = reinterpret_cast<EVP_CIPHER_CTX*>(p);
+#else
+    ctx = (EVP_CIPHER_CTX*)ctx_buf;
+#endif
+
     EVP_CIPHER_CTX_init(ctx);
   }
   virtual ~MyCTX()
@@ -149,8 +161,11 @@ public:
       uchar mask[MY_AES_BLOCK_SIZE];
       uint mlen;
 
-      my_aes_crypt(MY_AES_ECB, ENCRYPTION_FLAG_ENCRYPT | ENCRYPTION_FLAG_NOPAD,
+      int rc= my_aes_crypt(MY_AES_ECB, ENCRYPTION_FLAG_ENCRYPT | ENCRYPTION_FLAG_NOPAD,
                    oiv, sizeof(mask), mask, &mlen, key, klen, 0, 0);
+      DBUG_ASSERT(rc == MY_AES_OK);
+      if (rc)
+        return rc;
       DBUG_ASSERT(mlen == sizeof(mask));
 
       for (uint i=0; i < buf_len; i++)