MCObjectDisassembler.h revision 263508 
1//===-- llvm/MC/MCObjectDisassembler.h --------------------------*- C++ -*-===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file contains the declaration of the MCObjectDisassembler class, which
11// can be used to construct an MCModule and an MC CFG from an ObjectFile.
12//
13//===----------------------------------------------------------------------===//
14
15#ifndef LLVM_MC_MCOBJECTDISASSEMBLER_H
16#define LLVM_MC_MCOBJECTDISASSEMBLER_H
17
18#include "llvm/ADT/ArrayRef.h"
19#include "llvm/ADT/OwningPtr.h"
20#include "llvm/ADT/StringRef.h"
21#include "llvm/Support/DataTypes.h"
22#include "llvm/Support/MemoryObject.h"
23#include <vector>
24
25namespace llvm {
26
27namespace object {
28  class ObjectFile;
29  class MachOObjectFile;
30}
31
32class MCBasicBlock;
33class MCDisassembler;
34class MCFunction;
35class MCInstrAnalysis;
36class MCModule;
37class MCObjectSymbolizer;
38
39/// \brief Disassemble an ObjectFile to an MCModule and MCFunctions.
40/// This class builds on MCDisassembler to disassemble whole sections, creating
41/// MCAtom (MCTextAtom for disassembled sections and MCDataAtom for raw data).
42/// It can also be used to create a control flow graph consisting of MCFunctions
43/// and MCBasicBlocks.
44class MCObjectDisassembler {
45public:
46  MCObjectDisassembler(const object::ObjectFile &Obj,
47                       const MCDisassembler &Dis,
48                       const MCInstrAnalysis &MIA);
49  virtual ~MCObjectDisassembler() {}
50
51  /// \brief Build an MCModule, creating atoms and optionally functions.
52  /// \param withCFG Also build a CFG by adding MCFunctions to the Module.
53  /// If withCFG is false, the MCModule built only contains atoms, representing
54  /// what was found in the object file. If withCFG is true, MCFunctions are
55  /// created, containing MCBasicBlocks. All text atoms are split to form basic
56  /// block atoms, which then each back an MCBasicBlock.
57  MCModule *buildModule(bool withCFG = false);
58
59  MCModule *buildEmptyModule();
60
61  typedef std::vector<uint64_t> AddressSetTy;
62  /// \name Create a new MCFunction.
63  MCFunction *createFunction(MCModule *Module, uint64_t BeginAddr,
64                             AddressSetTy &CallTargets,
65                             AddressSetTy &TailCallTargets);
66
67  /// \brief Set the region on which to fallback if disassembly was requested
68  /// somewhere not accessible in the object file.
69  /// This is used for dynamic disassembly (see RawMemoryObject).
70  void setFallbackRegion(OwningPtr<MemoryObject> &Region) {
71    FallbackRegion.reset(Region.take());
72  }
73
74  /// \brief Set the symbolizer to use to get information on external functions.
75  /// Note that this isn't used to do instruction-level symbolization (that is,
76  /// plugged into MCDisassembler), but to symbolize function call targets.
77  void setSymbolizer(MCObjectSymbolizer *ObjectSymbolizer) {
78    MOS = ObjectSymbolizer;
79  }
80
81  /// \brief Get the effective address of the entrypoint, or 0 if there is none.
82  virtual uint64_t getEntrypoint();
83
84  /// \name Get the addresses of static constructors/destructors in the object.
85  /// The caller is expected to know how to interpret the addresses;
86  /// for example, Mach-O init functions expect 5 arguments, not for ELF.
87  /// The addresses are original object file load addresses, not effective.
88  /// @{
89  virtual ArrayRef<uint64_t> getStaticInitFunctions();
90  virtual ArrayRef<uint64_t> getStaticExitFunctions();
91  /// @}
92
93  /// \name Translation between effective and objectfile load address.
94  /// @{
95  /// \brief Compute the effective load address, from an objectfile virtual
96  /// address. This is implemented in a format-specific way, to take into
97  /// account things like PIE/ASLR when doing dynamic disassembly.
98  /// For example, on Mach-O this would be done by adding the VM addr slide,
99  /// on glibc ELF by keeping a map between segment load addresses, filled
100  /// using dl_iterate_phdr, etc..
101  /// In most static situations and in the default impl., this returns \p Addr.
102  virtual uint64_t getEffectiveLoadAddr(uint64_t Addr);
103
104  /// \brief Compute the original load address, as specified in the objectfile.
105  /// This is the inverse of getEffectiveLoadAddr.
106  virtual uint64_t getOriginalLoadAddr(uint64_t EffectiveAddr);
107  /// @}
108
109protected:
110  const object::ObjectFile &Obj;
111  const MCDisassembler &Dis;
112  const MCInstrAnalysis &MIA;
113  MCObjectSymbolizer *MOS;
114
115  /// \brief The fallback memory region, outside the object file.
116  OwningPtr<MemoryObject> FallbackRegion;
117
118  /// \brief Return a memory region suitable for reading starting at \p Addr.
119  /// In most cases, this returns a StringRefMemoryObject backed by the
120  /// containing section. When no section was found, this returns the
121  /// FallbackRegion, if it is suitable.
122  /// If it is not, or if there is no fallback region, this returns 0.
123  MemoryObject *getRegionFor(uint64_t Addr);
124
125private:
126  /// \brief Fill \p Module by creating an atom for each section.
127  /// This could be made much smarter, using information like symbols, but also
128  /// format-specific features, like mach-o function_start or data_in_code LCs.
129  void buildSectionAtoms(MCModule *Module);
130
131  /// \brief Enrich \p Module with a CFG consisting of MCFunctions.
132  /// \param Module An MCModule returned by buildModule, with no CFG.
133  /// NOTE: Each MCBasicBlock in a MCFunction is backed by a single MCTextAtom.
134  /// When the CFG is built, contiguous instructions that were previously in a
135  /// single MCTextAtom will be split in multiple basic block atoms.
136  void buildCFG(MCModule *Module);
137
138  MCBasicBlock *getBBAt(MCModule *Module, MCFunction *MCFN, uint64_t BeginAddr,
139                        AddressSetTy &CallTargets,
140                        AddressSetTy &TailCallTargets);
141};
142
143class MCMachOObjectDisassembler : public MCObjectDisassembler {
144  const object::MachOObjectFile &MOOF;
145
146  uint64_t VMAddrSlide;
147  uint64_t HeaderLoadAddress;
148
149  // __DATA;__mod_init_func support.
150  llvm::StringRef ModInitContents;
151  // __DATA;__mod_exit_func support.
152  llvm::StringRef ModExitContents;
153
154public:
155  /// \brief Construct a Mach-O specific object disassembler.
156  /// \param VMAddrSlide The virtual address slide applied by dyld.
157  /// \param HeaderLoadAddress The load address of the mach_header for this
158  /// object.
159  MCMachOObjectDisassembler(const object::MachOObjectFile &MOOF,
160                            const MCDisassembler &Dis,
161                            const MCInstrAnalysis &MIA, uint64_t VMAddrSlide,
162                            uint64_t HeaderLoadAddress);
163
164protected:
165  uint64_t getEffectiveLoadAddr(uint64_t Addr) LLVM_OVERRIDE;
166  uint64_t getOriginalLoadAddr(uint64_t EffectiveAddr) LLVM_OVERRIDE;
167  uint64_t getEntrypoint() LLVM_OVERRIDE;
168
169  ArrayRef<uint64_t> getStaticInitFunctions() LLVM_OVERRIDE;
170  ArrayRef<uint64_t> getStaticExitFunctions() LLVM_OVERRIDE;
171};
172
173}
174
175#endif
176