import-trace: introduce InstructionImporter

The InstructionImporter does disassemble the binary and generate read
traces for every instruction byte executed.

Change-Id: I6b8697c711c009e106ed733c74c6ff8f9bbf8ac5

tools/import-trace/CMakeLists.txt

@@ -3,6 +3,32 @@ set(SRCS
   MemoryImporter.cc
 )
 
+if (BUILD_LLVM_DISASSEMBLER)
+  set(SRCS ${SRCS}
+    InstructionImporter.cc
+  )
+
+  include(FindLLVM)
+
+  # llvm-config does add -fno-exception to the command line. But this
+  # breaks some boost libraries.
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${LLVM_CXX_FLAGS} -fexceptions")
+endif(BUILD_LLVM_DISASSEMBLER)
+
+
+
 add_executable(import-trace main.cc ${SRCS})
-target_link_libraries(import-trace ${PROTOBUF_LIBRARY} -lmysqlclient fail-util fail-sal fail-comm)
+
+target_link_libraries(import-trace 
+  ${PROTOBUF_LIBRARY} 
+  -lmysqlclient 
+  fail-util 
+  fail-comm
+  fail-sal)
+
+if (BUILD_LLVM_DISASSEMBLER)
+  target_link_libraries(import-trace fail-llvmdisassembler fail-sal ${LLVM_LIBS} ${LLVM_LDFLAGS})
+endif (BUILD_LLVM_DISASSEMBLER)
+
+
 install(TARGETS import-trace RUNTIME DESTINATION bin)

tools/import-trace/Importer.hpp

@@ -78,7 +78,8 @@ public:
 								 const Trace_Event &ev) = 0;
 
 
-	void set_elf_file(fail::ElfReader *elf) { m_elf = elf; }
+	void set_elf(fail::ElfReader *elf) { m_elf = elf; }
+
 	void set_memorymap(fail::MemoryMap *mm) { m_mm = mm; }
 	void set_faultspace_rightmargin(char accesstype) { m_faultspace_rightmargin = accesstype; }
 	void set_sanitychecks(bool enabled) { m_sanitychecks = enabled; }

tools/import-trace/InstructionImporter.cc

@@ -0,0 +1,80 @@
+#ifndef __puma
+#include <sstream>
+#include <iostream>
+#include "InstructionImporter.hpp"
+#include "util/Logger.hpp"
+
+using namespace llvm;
+using namespace llvm::object;
+using namespace fail;
+
+
+static Logger LOG("InstructionImporter");
+
+bool InstructionImporter::handle_ip_event(fail::simtime_t curtime, instruction_count_t instr,
+										  const Trace_Event &ev) {
+	if (!binary) {
+		/* Disassemble the binary if necessary */
+		llvm::InitializeAllTargetInfos();
+		llvm::InitializeAllTargetMCs();
+		llvm::InitializeAllDisassemblers();
+
+		if (error_code ec = createBinary(m_elf->getFilename(), binary)) {
+			LOG << m_elf->getFilename() << "': " << ec.message() << ".\n";
+			return false;
+		}
+
+		ObjectFile *obj = dyn_cast<ObjectFile>(binary.get());
+
+		disas.reset(new LLVMDisassembler(obj));
+		disas->disassemble();
+		LLVMDisassembler::InstrMap &instr_map = disas->getInstrMap();
+		LOG << "instructions disassembled: " << instr_map.size() << " Triple: " << disas->GetTriple() <<  std::endl;
+	}
+
+	const LLVMDisassembler::InstrMap &instr_map = disas->getInstrMap();
+	const LLVMDisassembler::Instr &opcode = instr_map.at(ev.ip());
+
+	address_t from = ev.ip(), to = ev.ip() + opcode.length;
+
+	// Iterate over all accessed bytes
+	for (address_t data_address = from; data_address < to; ++data_address) {
+		// skip events outside a possibly supplied memory map
+		if (m_mm && !m_mm->isMatching(data_address)) {
+			continue;
+		}
+		margin_info_t left_margin = getOpenEC(data_address);
+		margin_info_t right_margin;
+		right_margin.time = curtime;
+		right_margin.dyninstr = instr; // !< The current instruction
+		right_margin.ip = ev.ip();
+
+		// skip zero-sized intervals: these can occur when an instruction
+		// accesses a memory location more than once (e.g., INC, CMPXCHG)
+		if (left_margin.dyninstr > right_margin.dyninstr) {
+			continue;
+		}
+
+		// we now have an interval-terminating R/W event to the memaddr
+		// we're currently looking at; the EC is defined by
+		// data_address, dynamic instruction start/end, the absolute PC at
+		// the end, and time start/end
+		access_info_t access;
+		access.access_type  = 'R'; // instruction fetch is always a read
+		access.data_address = data_address;
+		access.data_width    = 1; // exactly one byte
+		if (!add_trace_event(left_margin, right_margin, access)) {
+			LOG << "add_trace_event failed" << std::endl;
+			return false;
+		}
+
+		// next interval must start at next instruction; the aforementioned
+		// skipping mechanism wouldn't work otherwise
+		newOpenEC(data_address, curtime + 1, instr + 1, ev.ip());
+	}
+
+	return true;
+}
+
+
+#endif // !__puma

tools/import-trace/InstructionImporter.hpp

@@ -0,0 +1,27 @@
+#ifndef __INSTRUCTION_IMPORTER_H__
+#define __INSTRUCTION_IMPORTER_H__
+
+#include "Importer.hpp"
+
+#ifndef __puma
+#include "util/llvmdisassembler/LLVMDisassembler.hpp"
+#endif
+
+
+class InstructionImporter : public Importer {
+#ifndef __puma
+	llvm::OwningPtr<llvm::object::Binary> binary;
+	llvm::OwningPtr<fail::LLVMDisassembler> disas;
+#endif
+
+public:
+	virtual bool handle_ip_event(fail::simtime_t curtime, instruction_count_t instr,
+								 const Trace_Event &ev);
+	virtual bool handle_mem_event(fail::simtime_t curtime, instruction_count_t instr,
+								  const Trace_Event &ev) {
+		/* ignore on purpose */
+		return true;
+	}
+};
+
+#endif

tools/import-trace/main.cc

@@ -1,5 +1,3 @@
-#include "util/optionparser/optionparser.h"
-#include "util/optionparser/optionparser_ext.hpp"
 #include "util/CommandLine.hpp"
 #include "util/Database.hpp"
 #include "util/ElfReader.hpp"
@@ -8,9 +6,12 @@
 #include "util/Logger.hpp"
 #include <fstream>
 #include <string>
-
 #include "MemoryImporter.hpp"
 
+#ifdef BUILD_LLVM_DISASSEMBLER
+#include "InstructionImporter.hpp"
+#endif
+
 
 using namespace fail;
 using std::cerr;
@@ -118,9 +119,14 @@ int main(int argc, char *argv[]) {
 
 	if (cmd[IMPORTER].count() > 0) {
 		std::string imp(cmd[IMPORTER].first()->arg);
-		if (imp == "BasicImporter" || imp == "MemoryImporter") {
+		if (imp == "BasicImporter" || imp == "MemoryImporter" || imp == "memory" || imp == "mem") {
 			LOG << "Using MemoryImporter" << endl;
 			importer = new MemoryImporter();
+#ifdef BUILD_LLVM_DISASSEMBLER
+		} else if (imp == "InstructionImporter" || imp == "code") {
+			LOG << "Using InstructionImporter" << endl;
+			importer = new InstructionImporter();
+#endif
 		} else {
 			LOG << "Unkown import method: " << imp << endl;
 			exit(-1);
@@ -160,7 +166,7 @@ int main(int argc, char *argv[]) {
 	if (cmd[ELF_FILE].count() > 0) {
 		elf_file = new ElfReader(cmd[ELF_FILE].first()->arg);
 	}
-	importer->set_elf_file(elf_file);
+	importer->set_elf(elf_file);
 
 	if (cmd[MEMORYMAP].count() > 0) {
 		memorymap = new MemoryMap();