Preliminary LLVM-3.6 support
authorChristoph Egger <Christoph.Egger@fau.de>
Wed, 25 Mar 2015 17:13:20 +0000 (18:13 +0100)
committerChristoph Egger <Christoph.Egger@fau.de>
Wed, 25 Mar 2015 17:13:20 +0000 (18:13 +0100)
Allows llvm-3.6 in cmake and does compile with 3.6. However it
does *not* work! It will segfault right away. One of the suspects is
different base for addresses used in llvm's getInstruction() and family.

CMakeLists.txt
cmake/Modules/FindLLVM.cmake
src/disassembler/llvm/LLVMDisassembler.cxx
src/disassembler/llvm/LLVMDisassembler.hxx
src/disassembler/llvm/include_llvm.hxx

index cbc597f0f6ed7ddc7f4bbeb7dc99cae656476d3b..c54380dd76dc17d8812553728fe7ffd5b88ac2eb 100644 (file)
@@ -96,7 +96,7 @@ qt5_use_modules(frida Widgets)
 TARGET_LINK_LIBRARIES(frida
   ${QT_LIBRARIES}
   ${LLVM_LDFLAGS}
-  -lLLVM-3.5
+  ${LLVM_DYNAMIC_LIB}
   ${LLVM_SYSTEM_LIBS}
   ${LOG4CXX_LDFLAGS}
   ${QUAZIP_LIBRARIES})
@@ -140,7 +140,16 @@ qt5_use_modules(fridatest Widgets)
 ADD_SUBDIRECTORY(/usr/src/gmock gmock)
 enable_testing()
 add_test(NAME fridatest COMMAND fridatest ${CMAKE_SOURCE_DIR}/testdata)
-target_link_libraries(fridatest gmock gtest ${QT_LIBRARIES} ${LLVM_LDFLAGS} -lLLVM-3.5 ${LLVM_SYSTEM_LIBS} ${LOG4CXX_LDFLAGS} ${GUILE_LDFLAGS} ${QUAZIP_LIBRARIES})
+target_link_libraries(fridatest
+  gmock
+  gtest
+  ${QT_LIBRARIES}
+  ${LLVM_LDFLAGS}
+  ${LLVM_DYNAMIC_LIB}
+  ${LLVM_SYSTEM_LIBS}
+  ${LOG4CXX_LDFLAGS}
+  ${GUILE_LDFLAGS}
+  ${QUAZIP_LIBRARIES})
 target_compile_options(fridatest PRIVATE -DCONFIG_TEST)
 set_property(TEST fridatest
   APPEND PROPERTY ENVIRONMENT LD_LIBRARY_PATH=${CMAKE_CURRENT_BINARY_DIR})
index ba2dd1521ba9eecfe5ac25b466b15933574cd162..6cf5e71235c3d2f5edf5de46c464b856ec49f6df 100644 (file)
@@ -1,10 +1,10 @@
-find_program(LLVM_CONFIG NAMES llvm-config-3.5 llvm-config)
+find_program(LLVM_CONFIG NAMES llvm-config-3.5 llvm-config llvm-config-3.6)
 
 if(LLVM_CONFIG)
   execute_process(COMMAND ${LLVM_CONFIG} --version
        OUTPUT_VARIABLE LLVM_VERSION)
 
-  if(((${LLVM_VERSION} VERSION_LESS 3.6.0)
+  if(((${LLVM_VERSION} VERSION_LESS 3.7.0)
                AND (${LLVM_VERSION} VERSION_GREATER 3.5))
          OR (${LLVM_VERSION} VERSION_EQUAL 3.5))
 
@@ -24,6 +24,15 @@ if(LLVM_CONFIG)
        string(REPLACE -I "-isystem" LLVM_CFLAGS ${LLVM_CFLAGS})
 
        separate_arguments(LLVM_CFLAGS)
+
+       if(${LLVM_VERSION} VERSION_LESS 3.6)
+         set(LLVM_CFLAGS ${LLVM_CFLAGS} -DLLVM_35)
+         set(LLVM_DYNAMIC_LIB -lLLVM-3.5)
+       else()
+         set(LLVM_CFLAGS ${LLVM_CFLAGS} -DLLVM_36)
+         set(LLVM_DYNAMIC_LIB -lLLVM-3.6)
+         message(WARNING "LLVM 3.6 support is highly experimental and may be broken in interesting ways")
+       endif()
   endif()
 endif()
 
index 45cd04e25566d97f830df05be7a00f42ffbe29a8..e673c371650027021455b63f64b94e422dcc7a5d 100644 (file)
@@ -36,13 +36,16 @@ Disassembler * createLLVMDisassembler(const std::string& filename, InformationMa
                LOG4CXX_ERROR(logger, ec.message());
                return NULL;
        }
-
+#if defined(LLVM_35)
        Binary * op = retval.get();
-
-       if (!op) {
-               LOG4CXX_ERROR(logger, "Could not open " << filename);
-               return NULL;
-       }
+#elif defined(LLVM_36)
+       OwningBinary<Binary> ob;
+       ob  = std::move(retval.get());
+       Binary* op = ob.getBinary();
+       auto foo = ob.takeBinary();
+       foo.first.release();
+       foo.second.release();
+#endif
 
        // ELFType<endian, maxalign, 64bit>
        if (ELF32LEObjectFile * object = dyn_cast<ELF32LEObjectFile>(op)) {
@@ -93,7 +96,15 @@ LLVMDisassembler<ELFT>::LLVMDisassembler(const std::string& filename,
                        return;
                }
 
+#if defined(LLVM_35)
                binary.reset(result.get());
+#elif defined(LLVM_36)
+               OwningBinary<Binary> ob;
+               ob  = std::move(result.get());
+               Binary* op = ob.getBinary();
+
+               binary.reset(op);
+#endif
 
                o = dyn_cast<ObjectFile>(binary.get());
        } else {
@@ -176,8 +187,8 @@ LLVMDisassembler<ELFT>::LLVMDisassembler(const std::string& filename,
        IP->setPrintImmHex(llvm::HexStyle::C);
        IP->setPrintImmHex(true);
 
-       std::unique_ptr<MCObjectDisassembler> OD(
-               new MCObjectDisassembler(*o, *DisAsm, *MIA));
+//     std::unique_ptr<MCObjectDisassembler> OD(
+//             new MCObjectDisassembler(*o, *DisAsm, *MIA));
        //Mod.reset(OD->buildModule(false));
 
        readSections();
@@ -198,9 +209,13 @@ Function* LLVMDisassembler<ELFT>::disassembleFunctionAt(uint64_t address, const
        Function * function;
        SectionRef text_section = getTextSection();
        uint64_t base_address, size;
+#if defined(LLVM_35)
        text_section.getAddress(base_address);
        text_section.getSize(size);
-
+#elif defined(LLVM_36)
+       base_address = text_section.getAddress();
+       size = text_section.getSize();
+#endif
        if (address < base_address ||
            address >= base_address + size) {
                return NULL;
@@ -235,7 +250,14 @@ void LLVMDisassembler<ELFT>::disassembleFunction(Function* function) {
        SectionRef text_section = getTextSection();
        StringRef bytes;
        text_section.getContents(bytes);
+#if defined(LLVM_35)
        StringRefMemoryObject ref(bytes);
+#elif defined(LLVM_36)
+       ArrayRef<uint8_t> bytearray(reinterpret_cast<const uint8_t *>(bytes.data()),
+                          bytes.size());
+#else
+#error LLVM != 3.5 | 3.6 not supported
+#endif
 
        LOG4CXX_DEBUG(logger, "Handling function " << function->getName());
 
@@ -245,8 +267,13 @@ void LLVMDisassembler<ELFT>::disassembleFunction(Function* function) {
        function->addBasicBlock(block);
 
        uint64_t base_address, size;
+#if defined(LLVM_35)
        text_section.getAddress(base_address);
        text_section.getSize(size);
+#elif defined(LLVM_36)
+       base_address = text_section.getAddress();
+       size = text_section.getSize();
+#endif
        LOG4CXX_DEBUG(logger, "Text section at " << std::hex << base_address << " with size " << size);
 
        while (remaining_blocks.size()) {
@@ -264,7 +291,14 @@ void LLVMDisassembler<ELFT>::disassembleFunction(Function* function) {
                        llvm::raw_string_ostream s(buf);
 
                        if(llvm::MCDisassembler::Success ==
+#if defined(LLVM_35)
                           DisAsm->getInstruction(inst, inst_size, ref, current_address, nulls(), nulls())) {
+#elif defined(LLVM_36)
+                               DisAsm->getInstruction(inst, inst_size,
+                                                      bytearray.slice(current_address),
+                                                      base_address + current_address,
+                                                      nulls(), nulls())) {
+#endif
                                uint64_t jmptarget;
 
                                if (MIA->evaluateBranch(inst, current_address, inst_size, jmptarget)) {
@@ -336,8 +370,11 @@ void LLVMDisassembler<ELFT>::disassemble() {
                bool contains;
                SymbolRef::Type symbol_type;
 
-
+#if defined(LLVM_35)
                if (text_section.containsSymbol(x->second, contains) || !contains)
+#elif defined(LLVM_36)
+               if (text_section.containsSymbol(x->second))
+#endif
                        continue;
 
                if (x->second.getType(symbol_type)
@@ -373,7 +410,11 @@ void LLVMDisassembler<ELFT>::disassemble() {
 
        if (!manager->hasFunctions()) {
                uint64_t text_entry;
+#if defined(LLVM_35)
                text_section.getAddress(text_entry);
+#elif defined(LLVM_36)
+               text_entry = text_section.getAddress();
+#endif
                LOG4CXX_INFO(logger, "No Symbols found, starting at the beginning of the text segment");
                disassembleFunctionAt(text_entry);
        }
@@ -414,7 +455,13 @@ void LLVMDisassembler<ELFT>::splitBlocks(Function* function) {
        SectionRef text_section = getTextSection();
        StringRef bytes;
        text_section.getContents(bytes);
+#if defined(LLVM_35)
        StringRefMemoryObject ref(bytes);
+#elif defined(LLVM_36)
+       ArrayRef<uint8_t> bytearray(reinterpret_cast<const uint8_t *>(bytes.data()),
+                          bytes.size());
+#endif
+
 
        LOG4CXX_DEBUG(logger, "Splitting Blocks in Function " << function->getName());
        // Split blocks where jumps are going inside the block
@@ -428,7 +475,11 @@ void LLVMDisassembler<ELFT>::splitBlocks(Function* function) {
                }
                uint64_t inst_size;
                uint64_t base_address;
+#if defined(LLVM_35)
                text_section.getAddress(base_address);
+#elif defined(LLVM_36)
+               base_address = text_section.getAddress();
+#endif
                uint64_t current_address = current_block->getStartAddress() - base_address;
                while(current_block->getEndAddress() - base_address > current_address) {
                        MCInst inst;
@@ -436,7 +487,15 @@ void LLVMDisassembler<ELFT>::splitBlocks(Function* function) {
                        llvm::raw_string_ostream s(buf);
 
                        if(llvm::MCDisassembler::Success ==
+#if defined(LLVM_35)
                           DisAsm->getInstruction(inst, inst_size, ref, current_address, nulls(), nulls())) {
+#elif defined(LLVM_36)
+                               DisAsm->getInstruction(inst, inst_size,
+                                                      bytearray.slice(current_address),
+                                                      base_address + current_address,
+                                                      nulls(), nulls())) {
+#endif
+
                                // See if some other block starts here
                                BasicBlock* other = manager->getBasicBlock(current_address
                                                                           + inst_size
@@ -541,13 +600,24 @@ std::vector<Instruction> LLVMDisassembler<ELFT>::getInstructions(const BasicBloc
        std::vector<Instruction> result;
        SectionRef text_section = getTextSection();
        uint64_t base_address;
+#if defined(LLVM_35)
        text_section.getAddress(base_address);
+#elif defined(LLVM_36)
+       base_address = text_section.getAddress();
+#endif
+
        uint64_t current_address = block->getStartAddress() - base_address;
        uint64_t end_position = block->getEndAddress() - base_address;
 
        StringRef bytes;
        text_section.getContents(bytes);
+#if defined(LLVM_35)
        StringRefMemoryObject ref(bytes);
+#elif defined(LLVM_36)
+       ArrayRef<uint8_t> bytearray(reinterpret_cast<const uint8_t *>(bytes.data()),
+                          bytes.size());
+#endif
+
 
        while (current_address < end_position) {
                uint64_t inst_size;
@@ -556,10 +626,24 @@ std::vector<Instruction> LLVMDisassembler<ELFT>::getInstructions(const BasicBloc
                llvm::raw_string_ostream s(buf);
 
                if(llvm::MCDisassembler::Success ==
-                  DisAsm->getInstruction(inst, inst_size, ref, current_address, nulls(), nulls())) {
+#if defined(LLVM_35)
+                          DisAsm->getInstruction(inst, inst_size, ref, current_address, nulls(), nulls())) {
+#elif defined(LLVM_36)
+                               DisAsm->getInstruction(inst, inst_size,
+                                                      bytearray.slice(current_address),
+                                                      base_address + current_address,
+                                                      nulls(), nulls())) {
+#endif
 
                        uint8_t bytes[inst_size+2];
+#if defined(LLVM_35)
                        ref.readBytes(current_address, inst_size, bytes);
+#elif defined(LLVM_36)
+                       size_t bytesindex(0);
+                       for (uint8_t byte : bytearray.slice(current_address, inst_size)) {
+                               bytes[bytesindex++] = byte;
+                       }
+#endif
 
                        uint64_t jmptarget;
                        std::string ref("");
@@ -579,7 +663,11 @@ std::vector<Instruction> LLVMDisassembler<ELFT>::getInstructions(const BasicBloc
                } else {
                        LOG4CXX_WARN(logger, "Invalid byte at" << std::hex << current_address + base_address);
                        uint8_t bytes[1];
+#if defined(LLVM_35)
                        ref.readBytes(current_address, 1, bytes);
+#elif defined(LLVM_36)
+                       bytes[0] = bytearray[current_address];
+#endif
                        result.push_back(Instruction(current_address + base_address, "Invalid Instruction",
                                                     std::vector<uint8_t>(bytes, bytes+1), ""));
                        inst_size = 1;
@@ -597,12 +685,23 @@ void LLVMDisassembler<ELFT>::printEachInstruction(uint64_t start, uint64_t end,
                                                                          const std::string&)> fun) {
        SectionRef text_section = getTextSection();
        uint64_t base_address;
+#if defined(LLVM_35)
        text_section.getAddress(base_address);
+#elif defined(LLVM_36)
+       base_address = text_section.getAddress();
+#endif
+
        uint64_t current_address = start - base_address;
 
        StringRef bytes;
        text_section.getContents(bytes);
+#if defined(LLVM_35)
        StringRefMemoryObject ref(bytes);
+#elif defined(LLVM_36)
+       ArrayRef<uint8_t> bytearray(reinterpret_cast<const uint8_t *>(bytes.data()),
+                          bytes.size());
+#endif
+
 
        while (current_address < end - base_address) {
                uint64_t inst_size;
@@ -611,10 +710,24 @@ void LLVMDisassembler<ELFT>::printEachInstruction(uint64_t start, uint64_t end,
                llvm::raw_string_ostream s(buf);
 
                if(llvm::MCDisassembler::Success ==
-                  DisAsm->getInstruction(inst, inst_size, ref, current_address, nulls(), nulls())) {
+#if defined(LLVM_35)
+                          DisAsm->getInstruction(inst, inst_size, ref, current_address, nulls(), nulls())) {
+#elif defined(LLVM_36)
+                               DisAsm->getInstruction(inst, inst_size,
+                                                      bytearray.slice(current_address),
+                                                      base_address + current_address,
+                                                      nulls(), nulls())) {
+#endif
 
                        uint8_t bytes[inst_size+2];
+#if defined(LLVM_35)
                        ref.readBytes(current_address, inst_size, bytes);
+#elif defined(LLVM_36)
+                       size_t bytesindex(0);
+                       for (uint8_t byte : bytearray.slice(current_address, inst_size)) {
+                               bytes[bytesindex++] = byte;
+                       }
+#endif
 
                        uint64_t jmptarget;
                        std::string ref("");
index df01c9a9f41f7f781880c4941ee12d062415fc7d..d8d4056734f458b416ae005b753c8a667ed1a35d 100644 (file)
@@ -54,7 +54,7 @@ private:
 
        std::unique_ptr<const llvm::MCRegisterInfo> MRI;
        std::unique_ptr<const llvm::MCAsmInfo> AsmInfo;
-       std::unique_ptr<llvm::MCModule> Mod;
+//     std::unique_ptr<llvm::MCModule> Mod;
        std::unique_ptr<llvm::MCInstPrinter> IP;
        std::unique_ptr<llvm::MCDisassembler> DisAsm;
        std::unique_ptr<const llvm::MCObjectFileInfo> MOFI;
index 46028f5f1ed2418f0541fcc3014e52cde4927848..2f73af47b041a37b1b18ec2e69bb37491e96cdce 100644 (file)
 #include <llvm/Object/ObjectFile.h>
 #include <llvm/Object/Archive.h>
 #include <llvm/MC/MCAsmInfo.h>
-#include <llvm/MC/MCAnalysis/MCAtom.h>
 #include <llvm/MC/MCContext.h>
 #include <llvm/MC/MCDisassembler.h>
-#include <llvm/MC/MCAnalysis/MCFunction.h>
 #include <llvm/MC/MCInst.h>
 #include <llvm/MC/MCInstPrinter.h>
 #include <llvm/MC/MCInstrAnalysis.h>
 #include <llvm/MC/MCInstrInfo.h>
-#include <llvm/MC/MCAnalysis/MCModule.h>
-#include <llvm/MC/MCObjectDisassembler.h>
 #include <llvm/MC/MCObjectFileInfo.h>
-#include <llvm/MC/MCObjectSymbolizer.h>
 #include <llvm/MC/MCRegisterInfo.h>
 #include <llvm/MC/MCRelocationInfo.h>
 #include <llvm/MC/MCSubtargetInfo.h>
 #include <llvm/Support/TargetRegistry.h>
 #include <llvm/Support/TargetSelect.h>
-#include <llvm/Support/StringRefMemoryObject.h>
 #include <llvm/Support/GraphWriter.h>
 
+#ifndef LLVM_36
+#include <llvm/MC/MCAnalysis/MCAtom.h>
+#include <llvm/MC/MCAnalysis/MCFunction.h>
+#include <llvm/MC/MCObjectSymbolizer.h>
+#include <llvm/Support/StringRefMemoryObject.h>
+#include <llvm/MC/MCAnalysis/MCModule.h>
+#include <llvm/MC/MCObjectDisassembler.h>
+#endif
+
 #endif