CMake improvements and refactor of __remill_basic_block (#447)

* Append project name to the check_git target to fix compound builds (#446)
* Change the way we get registers for __remill_basic_block so that it's the Arch class of the target that actually populates the function, and isn't based on reverse-engineering what's visible in the function as compiled by llvm
* Renamed a cmake var to reduce likelihood of conflicts, and tries to make sure remill only links against static llvm libraries
* Fix instruction dominance issues
* Minor fix on the type of PC on x86
* Added new implementation (in Arch.cpp) of the AArch64 __remill_basic_block function

Co-authored-by: Artem Dinaburg <[email protected]>
Co-authored-by: Eric Kilmer <[email protected]>

Author: 3 people

CMakeLists.txt

@@ -30,8 +30,8 @@ include(CTest)
 if (LLVM_Z3_INSTALL_DIR)
   find_package(Z3 4.7.1)
   set(need_z3 TRUE)
-elseif(DEFINED ENV{TRAILOFBITS_LIBRARIES})
-  set(LLVM_Z3_INSTALL_DIR "$ENV{TRAILOFBITS_LIBRARIES}/z3")
+elseif(DEFINED CXX_COMMON_REPOSITORY_ROOT)
+  set(LLVM_Z3_INSTALL_DIR "${CXX_COMMON_REPOSITORY_ROOT}/z3")
   set(need_z3 TRUE)
 else()
   set(need_z3 FALSE)
@@ -67,38 +67,80 @@ endif()
 # libraries
 #
 
+# LLVM
 find_package(LLVM REQUIRED CONFIG HINTS ${FINDPACKAGE_LLVM_HINTS})
 
 string(REPLACE "." ";" LLVM_VERSION_LIST ${LLVM_PACKAGE_VERSION})
 list(GET LLVM_VERSION_LIST 0 LLVM_MAJOR_VERSION)
 list(GET LLVM_VERSION_LIST 1 LLVM_MINOR_VERSION)
 
-set(LLVM_LIBRARIES
-  LLVMCore LLVMAnalysis LLVMSupport LLVMipo LLVMIRReader
-  LLVMBitReader LLVMBitWriter LLVMTransformUtils LLVMScalarOpts
-  LLVMLTO
+add_library(thirdparty_llvm INTERFACE)
+target_include_directories(thirdparty_llvm SYSTEM INTERFACE
+  ${LLVM_INCLUDE_DIRS}
+)
+target_compile_definitions(thirdparty_llvm INTERFACE
+  ${LLVM_DEFINITIONS}
+)
+
+# Go find only the static libraries of LLVM, and link against those.
+foreach(LLVM_LIB IN LISTS LLVM_AVAILABLE_LIBS)
+  get_target_property(LLVM_LIB_TYPE ${LLVM_LIB} TYPE)
+  if(LLVM_LIB_TYPE STREQUAL "STATIC_LIBRARY")
+    list(APPEND LLVM_LIBRARIES "${LLVM_LIB}")
+  endif()
+endforeach()
+
+# These are out-of-order in `LLVM_AVAILABLE_LIBS` and should always be last.
+list(REMOVE_ITEM LLVM_LIBRARIES LLVMMC LLVMCore LLVMSupport)
+list(APPEND LLVM_LIBRARIES LLVMMC LLVMCore LLVMSupport)
+message(WARNING "Libraries: ${LLVM_LIBRARIES}")
+
+
+target_link_libraries(thirdparty_llvm INTERFACE
+  ${LLVM_LIBRARIES}
 )
 
-list(APPEND PROJECT_LIBRARIES ${LLVM_LIBRARIES})
-list(APPEND PROJECT_DEFINITIONS ${LLVM_DEFINITIONS})
-list(APPEND PROJECT_INCLUDEDIRECTORIES ${LLVM_INCLUDE_DIRS})
+# Microsoft Z3
+add_library(thirdparty_z3 INTERFACE)
+if(Z3_FOUND)
+  target_include_directories(thirdparty_z3 SYSTEM INTERFACE
+    ${Z3_INCLUDE_DIR}
+  )
+  target_link_libraries(thirdparty_z3 INTERFACE
+    ${Z3_LIBRARIES}
+  )
+endif()
 
-# xed
+# Intel XED
 find_package(XED REQUIRED)
-list(APPEND PROJECT_LIBRARIES ${XED_LIBRARIES})
-list(APPEND PROJECT_INCLUDEDIRECTORIES ${XED_INCLUDE_DIRS})
+add_library(thirdparty_xed INTERFACE)
+target_include_directories(thirdparty_xed SYSTEM INTERFACE
+  ${XED_INCLUDE_DIRS}
+)
+target_link_libraries(thirdparty_xed INTERFACE
+  ${XED_LIBRARIES}
+)
 
-# google log module
+# Google glog module
 find_package(glog REQUIRED)
-list(APPEND PROJECT_LIBRARIES glog::glog)
+add_library(thirdparty_glog INTERFACE)
+target_link_libraries(thirdparty_glog INTERFACE
+  glog::glog
+)
 
-# gflags
+# Google gflags
 find_package(gflags REQUIRED)
-list(APPEND PROJECT_LIBRARIES gflags)
+add_library(thirdparty_gflags INTERFACE)
+target_link_libraries(thirdparty_gflags INTERFACE
+  gflags
+)
 
-# windows sdk
+# Windows SDK
+add_library(thirdparty_win32 INTERFACE)
 if(DEFINED WIN32)
-  list(APPEND PROJECT_LIBRARIES "Kernel32.lib")
+  target_link_libraries(thirdparty_win32 INTERFACE
+     "Kernel32.lib"
+  )
 endif()
 
 #
@@ -122,14 +164,10 @@ endif()
 set(REMILL_BUILD_SEMANTICS_DIR_X86 "${CMAKE_CURRENT_BINARY_DIR}/remill/Arch/X86/Runtime/")
 set(REMILL_BUILD_SEMANTICS_DIR_AARCH64 "${CMAKE_CURRENT_BINARY_DIR}/remill/Arch/AArch64/Runtime/")
 
-list(APPEND PROJECT_DEFINITIONS "REMILL_INSTALL_SEMANTICS_DIR=\"${REMILL_INSTALL_SEMANTICS_DIR}/\"")
-list(APPEND PROJECT_DEFINITIONS "REMILL_BUILD_SEMANTICS_DIR_X86=\"${REMILL_BUILD_SEMANTICS_DIR_X86}\"")
-list(APPEND PROJECT_DEFINITIONS "REMILL_BUILD_SEMANTICS_DIR_AARCH64=\"${REMILL_BUILD_SEMANTICS_DIR_AARCH64}\"")
-
 # verion data
 add_subdirectory(remill/Version)
 
-add_library(${PROJECT_NAME} STATIC
+add_library(remill STATIC
   remill/Arch/AArch64/Arch.cpp
   remill/Arch/AArch64/Decode.cpp
   remill/Arch/AArch64/Extract.cpp
@@ -151,14 +189,102 @@ add_library(${PROJECT_NAME} STATIC
   remill/OS/OS.cpp
 )
 
-set_property(TARGET ${PROJECT_NAME} PROPERTY POSITION_INDEPENDENT_CODE ON)
+set_property(TARGET remill PROPERTY POSITION_INDEPENDENT_CODE ON)
+set(THIRDPARTY_LIBRARY_LIST thirdparty_z3 thirdparty_llvm thirdparty_xed thirdparty_glog thirdparty_gflags)
 
 # add everything as public.
-target_link_libraries(${PROJECT_NAME} PUBLIC ${PROJECT_LIBRARIES} RemillVersion)
-target_include_directories(${PROJECT_NAME} PUBLIC ${CMAKE_SOURCE_DIR})
-target_include_directories(${PROJECT_NAME} SYSTEM PUBLIC ${PROJECT_INCLUDEDIRECTORIES})
-target_compile_definitions(${PROJECT_NAME} PUBLIC ${PROJECT_DEFINITIONS} ${GLOBAL_DEFINITIONS})
-target_compile_options(${PROJECT_NAME} PUBLIC ${GLOBAL_CXXFLAGS})
+
+if(UNIX AND NOT APPLE)
+  #For Linux builds, group LLVM libraries into a single group
+  # that avoids frustrating library ordering issues
+  set(LINKER_START_GROUP "-Wl,--start-group")
+  set(LINKER_END_GROUP "-Wl,--end-group")
+else()
+  set(LINKER_START_GROUP "")
+  set(LINKER_END_GROUP "")
+endif()
+
+
+target_link_libraries(remill LINK_PUBLIC
+  ${LINKER_START_GROUP} ${THIRDPARTY_LIBRARY_LIST} ${LINKER_END_GROUP}
+  RemillVersion
+)
+
+target_include_directories(remill PUBLIC ${CMAKE_SOURCE_DIR})
+
+if(WIN32)
+  # warnings and compiler settings
+  target_compile_options(remill PUBLIC
+    /MD /nologo /W3 /EHsc /wd4141 /wd4146 /wd4180 /wd4244
+    /wd4258 /wd4267 /wd4291 /wd4345 /wd4351 /wd4355 /wd4456
+    /wd4457 /wd4458 /wd4459 /wd4503 /wd4624 /wd4722 /wd4800
+    /wd4100 /wd4127 /wd4512 /wd4505 /wd4610 /wd4510 /wd4702
+    /wd4245 /wd4706 /wd4310 /wd4701 /wd4703 /wd4389 /wd4611
+    /wd4805 /wd4204 /wd4577 /wd4091 /wd4592 /wd4324
+  )
+
+  target_compile_definitions(remill PUBLIC
+    _CRT_SECURE_NO_DEPRECATE
+    _CRT_SECURE_NO_WARNINGS
+    _CRT_NONSTDC_NO_DEPRECATE
+    _CRT_NONSTDC_NO_WARNINGS
+    _SCL_SECURE_NO_DEPRECATE
+    _SCL_SECURE_NO_WARNINGS
+    GOOGLE_PROTOBUF_NO_RTTI
+  )
+
+else()
+  # warnings and compiler settings
+  target_compile_options(remill PUBLIC
+    -Wall -Wextra -Wno-unused-parameter -Wno-c++98-compat
+    -Wno-unreachable-code-return -Wno-nested-anon-types
+    -Wno-extended-offsetof
+    -Wno-variadic-macros -Wno-return-type-c-linkage
+    -Wno-c99-extensions -Wno-ignored-attributes -Wno-unused-local-typedef
+    -Wno-unknown-pragmas -Wno-unknown-warning-option -fPIC
+    -fno-omit-frame-pointer -fvisibility-inlines-hidden
+    -fno-asynchronous-unwind-tables
+  )
+
+  # Clang-specific warnings/error options
+  if("${CMAKE_C_COMPILER_ID}" STREQUAL "Clang" OR "${CMAKE_C_COMPILER_ID}" STREQUAL "AppleClang")
+    target_compile_options(remill PUBLIC
+      -Wgnu-alignof-expression -Wno-gnu-anonymous-struct -Wno-gnu-designator
+      -Wno-gnu-zero-variadic-macro-arguments -Wno-gnu-statement-expression
+      -fno-aligned-allocation
+    )
+  endif()
+
+  # debug symbols
+  if(CMAKE_BUILD_TYPE STREQUAL "Debug" OR CMAKE_BUILD_TYPE STREQUAL "RelWithDebInfo")
+    target_compile_options(remill PUBLIC
+      -gdwarf-2 -g3
+    )
+  endif()
+
+  # optimization flags and definitions
+  if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+    target_compile_options(remill PUBLIC
+      -O0
+    )
+    target_compile_definitions(remill PUBLIC
+      "DEBUG"
+    )
+  else()
+    target_compile_options(remill PUBLIC
+      -O2
+    )
+    target_compile_definitions(remill PUBLIC
+      "NDEBUG"
+    )
+  endif()
+endif()
+
+target_compile_definitions(remill PUBLIC
+  "REMILL_INSTALL_SEMANTICS_DIR=\"${REMILL_INSTALL_SEMANTICS_DIR}/\""
+  "REMILL_BUILD_SEMANTICS_DIR_X86=\"${REMILL_BUILD_SEMANTICS_DIR_X86}\""
+  "REMILL_BUILD_SEMANTICS_DIR_AARCH64=\"${REMILL_BUILD_SEMANTICS_DIR_AARCH64}\""
+)
 
 #
 # Also install clang, libllvm and llvm-link
@@ -234,7 +360,7 @@ endfunction()
 set(INSTALLED_CLANG_NAME "remill-clang-${REMILL_LLVM_VERSION}${executable_extension}")
 set(INSTALLED_LLVMLINK_NAME "remill-llvm-link-${REMILL_LLVM_VERSION}${executable_extension}")
 
-if("${LIBRARY_REPOSITORY_ROOT}" STREQUAL "" OR NOT EXISTS "${LIBRARY_REPOSITORY_ROOT}/llvm")
+if("${CXX_COMMON_REPOSITORY_ROOT}" STREQUAL "" OR NOT EXISTS "${CXX_COMMON_REPOSITORY_ROOT}/llvm")
   set(INSTALLED_LIBLLVM_NAME "${dynamic_lib_prefix}LLVM-${REMILL_LLVM_VERSION}.${dynamic_lib_extension}")
 
   # system binaries are not built statically, so we need to fix the rpath
@@ -268,10 +394,10 @@ if("${LIBRARY_REPOSITORY_ROOT}" STREQUAL "" OR NOT EXISTS "${LIBRARY_REPOSITORY_
 else()
   # The executable in our binary repository are statically built, meaning that we don't need
   # to change the rpath
-  InstallExternalTarget("ext_clang" "${LIBRARY_REPOSITORY_ROOT}/llvm/bin/clang${executable_extension}"
+  InstallExternalTarget("ext_clang" "${CXX_COMMON_REPOSITORY_ROOT}/llvm/bin/clang${executable_extension}"
     "${install_folder}/bin" "${INSTALLED_CLANG_NAME}")
 
-  InstallExternalTarget("ext_llvmlink" "${LIBRARY_REPOSITORY_ROOT}/llvm/bin/llvm-link${executable_extension}"
+  InstallExternalTarget("ext_llvmlink" "${CXX_COMMON_REPOSITORY_ROOT}/llvm/bin/llvm-link${executable_extension}"
     "${install_folder}/bin" "${INSTALLED_LLVMLINK_NAME}")
 endif()
 
@@ -282,9 +408,42 @@ install(TARGETS "${PROJECT_NAME}"
 
 set(REMILL_LIBRARY_LOCATION "${install_folder}/lib/libremill.a")
 set(REMILL_INCLUDE_LOCATION "${install_folder}/include")
+get_target_property(REMILL_COMPILE_OPTIONS remill COMPILE_OPTIONS)
+get_target_property(REMILL_COMPILE_DEFINITIONS remill COMPILE_DEFINITIONS)
+
+GetTargetTree(THIRDPARTY_LIBRARIES ${THIRDPARTY_LIBRARY_LIST})
+GetPublicIncludeFolders(THIRDPARTY_INCLUDE_DIRECTORIES ${THIRDPARTY_LIBRARIES})
+foreach(THIRDPARTY_LIB IN LISTS THIRDPARTY_LIBRARIES)
+  string(SUBSTRING "${THIRDPARTY_LIB}" 0 1 THIRDPARTY_LIB_PREFIX)
+  if(TARGET ${THIRDPARTY_LIB})
+    get_target_property(THIRDPARTY_LIB_TYPE ${THIRDPARTY_LIB} TYPE)
+    if(THIRDPARTY_LIB_TYPE STREQUAL "STATIC_LIBRARY" OR THIRDPARTY_LIB_TYPE STREQUAL "SHARED_LIBRARY")
+      list(APPEND THIRDPARTY_LIBRARY_FILES "$${}<TARGET_FILE:${THIRDPARTY_LIB}>")
+    endif()
+  elseif("${THIRDPARTY_LIB_PREFIX}" STREQUAL "$${}")
+    # E.g. $<LINK_ONLY:...>
+  else()
+    list(APPEND THIRDPARTY_LIBRARY_FILES "${THIRDPARTY_LIB}")
+  endif()
+endforeach()
+
+list(REMOVE_DUPLICATES THIRDPARTY_LIBRARY_FILES)
+if(NOT "x{$LINKER_START_GROUP}x" STREQUAL "xx")
+  list(INSERT THIRDPARTY_LIBRARY_FILES 0 "${LINKER_START_GROUP}")
+  list(APPEND THIRDPARTY_LIBRARY_FILES "${LINKER_END_GROUP}")
+endif()
+
+# First do the basic substitutions.
 configure_file(
   "${CMAKE_CURRENT_SOURCE_DIR}/cmake/remillConfig.cmake.in"
-  "${CMAKE_CURRENT_BINARY_DIR}/remillConfig.cmake"
+  "${CMAKE_CURRENT_BINARY_DIR}/remillConfig.cmake.pregen"
+  @ONLY
+)
+
+# Then expand the generator expressions added to `THIRDPARTY_LIBRARY_FILES`.
+file(GENERATE
+  OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/remillConfig.cmake"
+  INPUT "${CMAKE_CURRENT_BINARY_DIR}/remillConfig.cmake.pregen"
 )
 
 install(FILES "${CMAKE_CURRENT_BINARY_DIR}/remillConfig.cmake"
@@ -316,6 +475,7 @@ install(FILES
   "${CMAKE_CURRENT_SOURCE_DIR}/remill/BC/Compat/Attributes.h"
   "${CMAKE_CURRENT_SOURCE_DIR}/remill/BC/Compat/BitcodeReaderWriter.h"
   "${CMAKE_CURRENT_SOURCE_DIR}/remill/BC/Compat/CallingConvention.h"
+  "${CMAKE_CURRENT_SOURCE_DIR}/remill/BC/Compat/CTypes.h"
   "${CMAKE_CURRENT_SOURCE_DIR}/remill/BC/Compat/DataLayout.h"
   "${CMAKE_CURRENT_SOURCE_DIR}/remill/BC/Compat/DebugInfo.h"
   "${CMAKE_CURRENT_SOURCE_DIR}/remill/BC/Compat/Error.h"

cmake/BCCompiler.cmake

@@ -17,6 +17,7 @@ set(DEFAULT_BC_COMPILER_FLAGS
   -ffreestanding -fno-common -fno-builtin -fno-exceptions -fno-rtti
   -fno-asynchronous-unwind-tables -Wno-unneeded-internal-declaration
   -Wno-unused-function -Wgnu-inline-cpp-without-extern -std=c++14
+  -Wno-pass-failed=transform-warning
   ${EXTRA_BC_SYSROOT}
 )
 

cmake/git_watcher.cmake

@@ -158,17 +158,18 @@ function(GetGitState _working_dir)
 
     RunGitCommand(show -s "--format=%s" ${object})
     if(exit_code EQUAL 0)
+        string(REPLACE "\"" "\\\"" output "${output}")
         set(ENV{GIT_COMMIT_SUBJECT} "${output}")
     endif()
 
     RunGitCommand(show -s "--format=%b" ${object})
     if(exit_code EQUAL 0)
         if(output)
             # Escape line breaks in the commit message.
-            string(REPLACE "\r\n" "\\r\\n\\\r\n" safe ${output})
+            string(REPLACE "\r\n" "\\r\\n\\\r\n" safe "${output}")
             if(safe STREQUAL output)
                 # Didn't have windows lines - try unix lines.
-                string(REPLACE "\n" "\\n\\\n" safe ${output})
+                string(REPLACE "\n" "\\n\\\n" safe "${output}")
             endif()
         else()
             # There was no commit body - set the safe string to empty.

cmake/remillConfig.cmake.in

@@ -16,20 +16,28 @@ cmake_minimum_required(VERSION 3.2)
 
 if(NOT TARGET remill)
   if(WIN32)
-    set(REMILL_LIBRARY_LOCATION "${CMAKE_INSTALL_PREFIX}/remill/lib/remill.lib")
-    set(REMILL_INCLUDE_LOCATION "${CMAKE_INSTALL_PREFIX}/remill/include")
+    set(REMILL_LIBRARY_LOCATION "@CMAKE_INSTALL_PREFIX@/remill/lib/remill.lib")
+    set(REMILL_INCLUDE_LOCATION "@CMAKE_INSTALL_PREFIX@/remill/include")
   else()
-    set(REMILL_LIBRARY_LOCATION "${CMAKE_INSTALL_PREFIX}/lib/libremill.a")
-    set(REMILL_INCLUDE_LOCATION "${CMAKE_INSTALL_PREFIX}/include")
+    set(REMILL_LIBRARY_LOCATION "@CMAKE_INSTALL_PREFIX@/lib/libremill.a")
+    set(REMILL_INCLUDE_LOCATION "@CMAKE_INSTALL_PREFIX@/include")
   endif()
 
-  add_library(remill STATIC IMPORTED)
-  set_property(TARGET remill PROPERTY IMPORTED_LOCATION "${REMILL_LIBRARY_LOCATION}")
-  target_include_directories(remill INTERFACE "${REMILL_INCLUDE_LOCATION}")
+  if(NOT "x@CXX_COMMON_REPOSITORY_ROOT@x" STREQUAL "xx")
+    set(CXX_COMMON_REPOSITORY_ROOT "@CXX_COMMON_REPOSITORY_ROOT@")
+  endif()
 
-  target_compile_definitions(remill INTERFACE "REMILL_INSTALL_SEMANTICS_DIR=\"${REMILL_INSTALL_SEMANTICS_DIR}\"")
-  target_compile_definitions(remill INTERFACE "REMILL_BUILD_SEMANTICS_DIR_X86=\"${REMILL_INSTALL_SEMANTICS_DIR}\"")
-  target_compile_definitions(remill INTERFACE "REMILL_BUILD_SEMANTICS_DIR_AARCH64=\"${REMILL_INSTALL_SEMANTICS_DIR}\"")
+  set(LLVM_MAJOR_VERSION @LLVM_MAJOR_VERSION@)
+  set(LLVM_MINOR_VERSION @LLVM_MINOR_VERSION@)
+  set(REMILL_LLVM_VERSION "@LLVM_MAJOR_VERSION@.@LLVM_MINOR_VERSION@")
+
+  add_library(remill STATIC IMPORTED)
+  set_property(TARGET remill PROPERTY IMPORTED_LOCATION "@REMILL_LIBRARY_LOCATION@")
+  target_include_directories(remill INTERFACE @THIRDPARTY_INCLUDE_DIRECTORIES@)
+  target_include_directories(remill INTERFACE @REMILL_INCLUDE_LOCATION@)
+  target_compile_options(remill INTERFACE @REMILL_COMPILE_OPTIONS@)
+  target_compile_definitions(remill INTERFACE @REMILL_COMPILE_DEFINITIONS@)
+  target_link_libraries(remill INTERFACE @THIRDPARTY_LIBRARY_FILES@)
 
   # Add a dummy 'semantics' target to satisfy the protobuf generator
   add_custom_target(semantics)