MC: Disassembled CFG reconstruction.
authorAhmed Bougacha <ahmed.bougacha@gmail.com>
Fri, 24 May 2013 01:07:04 +0000 (01:07 +0000)
committerAhmed Bougacha <ahmed.bougacha@gmail.com>
Fri, 24 May 2013 01:07:04 +0000 (01:07 +0000)
commitef99356dfebb96f6f90efb912c2877214bad060e
tree76250a4be7eff5e9bec963f6ff0daef8cb8d84bf
parent2c94d0faa0e1c268893d5e04dc77e8a35889db00
MC: Disassembled CFG reconstruction.

This patch builds on some existing code to do CFG reconstruction from
a disassembled binary:
- MCModule represents the binary, and has a list of MCAtoms.
- MCAtom represents either disassembled instructions (MCTextAtom), or
  contiguous data (MCDataAtom), and covers a specific range of addresses.
- MCBasicBlock and MCFunction form the reconstructed CFG. An MCBB is
  backed by an MCTextAtom, and has the usual successors/predecessors.
- MCObjectDisassembler creates a module from an ObjectFile using a
  disassembler. It first builds an atom for each section. It can also
  construct the CFG, and this splits the text atoms into basic blocks.

MCModule and MCAtom were only sketched out; MCFunction and MCBB were
implemented under the experimental "-cfg" llvm-objdump -macho option.
This cleans them up for further use; llvm-objdump -d -cfg now generates
graphviz files for each function found in the binary.

In the future, MCObjectDisassembler may be the right place to do
"intelligent" disassembly: for example, handling constant islands is just
a matter of splitting the atom, using information that may be available
in the ObjectFile. Also, better initial atom formation than just using
sections is possible using symbols (and things like Mach-O's
function_starts load command).

This brings two minor regressions in llvm-objdump -macho -cfg:
- The printing of a relocation's referenced symbol.
- An annotation on loop BBs, i.e., which are their own successor.

Relocation printing is replaced by the MCSymbolizer; the basic CFG
annotation will be superseded by more related functionality.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@182628 91177308-0d34-0410-b5e6-96231b3b80d8
22 files changed:
include/llvm/MC/MCAtom.h
include/llvm/MC/MCFunction.h [new file with mode: 0644]
include/llvm/MC/MCInstrAnalysis.h
include/llvm/MC/MCModule.h
include/llvm/MC/MCObjectDisassembler.h [new file with mode: 0644]
include/llvm/Support/StringRefMemoryObject.h [new file with mode: 0644]
lib/MC/CMakeLists.txt
lib/MC/MCAtom.cpp
lib/MC/MCFunction.cpp [new file with mode: 0644]
lib/MC/MCInstrAnalysis.cpp
lib/MC/MCModule.cpp
lib/MC/MCObjectDisassembler.cpp [new file with mode: 0644]
lib/Support/CMakeLists.txt
lib/Support/StringRefMemoryObject.cpp [new file with mode: 0644]
lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.cpp
lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.cpp
tools/llvm-objdump/CMakeLists.txt
tools/llvm-objdump/MCFunction.cpp [deleted file]
tools/llvm-objdump/MCFunction.h [deleted file]
tools/llvm-objdump/MachODump.cpp
tools/llvm-objdump/llvm-objdump.cpp
tools/llvm-objdump/llvm-objdump.h