1 //===-- MachODump.cpp - Object file dumping utility for llvm --------------===//
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 implements the MachO-specific dumper for llvm-objdump.
11 //
12 //===----------------------------------------------------------------------===//
14 #include "llvm-objdump.h"
15 #include "llvm/ADT/OwningPtr.h"
16 #include "llvm/ADT/STLExtras.h"
17 #include "llvm/ADT/StringExtras.h"
18 #include "llvm/ADT/Triple.h"
19 #include "llvm/DebugInfo/DIContext.h"
20 #include "llvm/MC/MCAsmInfo.h"
21 #include "llvm/MC/MCDisassembler.h"
22 #include "llvm/MC/MCInst.h"
23 #include "llvm/MC/MCInstPrinter.h"
24 #include "llvm/MC/MCInstrAnalysis.h"
25 #include "llvm/MC/MCInstrDesc.h"
26 #include "llvm/MC/MCInstrInfo.h"
27 #include "llvm/MC/MCRegisterInfo.h"
28 #include "llvm/MC/MCSubtargetInfo.h"
29 #include "llvm/Object/MachO.h"
30 #include "llvm/Support/Casting.h"
31 #include "llvm/Support/CommandLine.h"
32 #include "llvm/Support/Debug.h"
33 #include "llvm/Support/Format.h"
34 #include "llvm/Support/GraphWriter.h"
35 #include "llvm/Support/MachO.h"
36 #include "llvm/Support/MemoryBuffer.h"
37 #include "llvm/Support/TargetRegistry.h"
38 #include "llvm/Support/TargetSelect.h"
39 #include "llvm/Support/raw_ostream.h"
40 #include "llvm/Support/system_error.h"
41 #include <algorithm>
42 #include <cstring>
43 using namespace llvm;
44 using namespace object;
46 static cl::opt<bool>
47 UseDbg("g", cl::desc("Print line information from debug info if available"));
49 static cl::opt<std::string>
50 DSYMFile("dsym", cl::desc("Use .dSYM file for debug info"));
52 static const Target *GetTarget(const MachOObjectFile *MachOObj) {
53 // Figure out the target triple.
54 if (TripleName.empty()) {
55 llvm::Triple TT("unknown-unknown-unknown");
56 TT.setArch(Triple::ArchType(MachOObj->getArch()));
57 TripleName = TT.str();
58 }
60 // Get the target specific parser.
61 std::string Error;
62 const Target *TheTarget = TargetRegistry::lookupTarget(TripleName, Error);
63 if (TheTarget)
64 return TheTarget;
66 errs() << "llvm-objdump: error: unable to get target for '" << TripleName
67 << "', see --version and --triple.\n";
68 return 0;
69 }
71 struct SymbolSorter {
72 bool operator()(const SymbolRef &A, const SymbolRef &B) {
73 SymbolRef::Type AType, BType;
74 A.getType(AType);
75 B.getType(BType);
77 uint64_t AAddr, BAddr;
78 if (AType != SymbolRef::ST_Function)
79 AAddr = 0;
80 else
81 A.getAddress(AAddr);
82 if (BType != SymbolRef::ST_Function)
83 BAddr = 0;
84 else
85 B.getAddress(BAddr);
86 return AAddr < BAddr;
87 }
88 };
90 // Types for the storted data in code table that is built before disassembly
91 // and the predicate function to sort them.
92 typedef std::pair<uint64_t, DiceRef> DiceTableEntry;
93 typedef std::vector<DiceTableEntry> DiceTable;
94 typedef DiceTable::iterator dice_table_iterator;
96 static bool
97 compareDiceTableEntries(const DiceTableEntry i,
98 const DiceTableEntry j) {
99 return i.first == j.first;
100 }
102 static void DumpDataInCode(const char *bytes, uint64_t Size,
103 unsigned short Kind) {
104 uint64_t Value;
106 switch (Kind) {
107 case MachO::DICE_KIND_DATA:
108 switch (Size) {
109 case 4:
110 Value = bytes[3] << 24 |
111 bytes[2] << 16 |
112 bytes[1] << 8 |
113 bytes[0];
114 outs() << "\t.long " << Value;
115 break;
116 case 2:
117 Value = bytes[1] << 8 |
118 bytes[0];
119 outs() << "\t.short " << Value;
120 break;
121 case 1:
122 Value = bytes[0];
123 outs() << "\t.byte " << Value;
124 break;
125 }
126 outs() << "\t@ KIND_DATA\n";
127 break;
128 case MachO::DICE_KIND_JUMP_TABLE8:
129 Value = bytes[0];
130 outs() << "\t.byte " << Value << "\t@ KIND_JUMP_TABLE8";
131 break;
132 case MachO::DICE_KIND_JUMP_TABLE16:
133 Value = bytes[1] << 8 |
134 bytes[0];
135 outs() << "\t.short " << Value << "\t@ KIND_JUMP_TABLE16";
136 break;
137 case MachO::DICE_KIND_JUMP_TABLE32:
138 Value = bytes[3] << 24 |
139 bytes[2] << 16 |
140 bytes[1] << 8 |
141 bytes[0];
142 outs() << "\t.long " << Value << "\t@ KIND_JUMP_TABLE32";
143 break;
144 default:
145 outs() << "\t@ data in code kind = " << Kind << "\n";
146 break;
147 }
148 }
150 static void
151 getSectionsAndSymbols(const MachO::mach_header Header,
152 MachOObjectFile *MachOObj,
153 std::vector<SectionRef> &Sections,
154 std::vector<SymbolRef> &Symbols,
155 SmallVectorImpl<uint64_t> &FoundFns,
156 uint64_t &BaseSegmentAddress) {
157 for (symbol_iterator SI = MachOObj->symbol_begin(),
158 SE = MachOObj->symbol_end();
159 SI != SE; ++SI)
160 Symbols.push_back(*SI);
162 for (section_iterator SI = MachOObj->section_begin(),
163 SE = MachOObj->section_end();
164 SI != SE; ++SI) {
165 SectionRef SR = *SI;
166 StringRef SectName;
167 SR.getName(SectName);
168 Sections.push_back(*SI);
169 }
171 MachOObjectFile::LoadCommandInfo Command =
172 MachOObj->getFirstLoadCommandInfo();
173 bool BaseSegmentAddressSet = false;
174 for (unsigned i = 0; ; ++i) {
175 if (Command.C.cmd == MachO::LC_FUNCTION_STARTS) {
176 // We found a function starts segment, parse the addresses for later
177 // consumption.
178 MachO::linkedit_data_command LLC =
179 MachOObj->getLinkeditDataLoadCommand(Command);
181 MachOObj->ReadULEB128s(LLC.dataoff, FoundFns);
182 }
183 else if (Command.C.cmd == MachO::LC_SEGMENT) {
184 MachO::segment_command SLC =
185 MachOObj->getSegmentLoadCommand(Command);
186 StringRef SegName = SLC.segname;
187 if(!BaseSegmentAddressSet && SegName != "__PAGEZERO") {
188 BaseSegmentAddressSet = true;
189 BaseSegmentAddress = SLC.vmaddr;
190 }
191 }
193 if (i == Header.ncmds - 1)
194 break;
195 else
196 Command = MachOObj->getNextLoadCommandInfo(Command);
197 }
198 }
200 static void DisassembleInputMachO2(StringRef Filename,
201 MachOObjectFile *MachOOF);
203 void llvm::DisassembleInputMachO(StringRef Filename) {
204 OwningPtr<MemoryBuffer> Buff;
206 if (error_code ec = MemoryBuffer::getFileOrSTDIN(Filename, Buff)) {
207 errs() << "llvm-objdump: " << Filename << ": " << ec.message() << "\n";
208 return;
209 }
211 OwningPtr<MachOObjectFile> MachOOF(static_cast<MachOObjectFile *>(
212 ObjectFile::createMachOObjectFile(Buff.release()).get()));
214 DisassembleInputMachO2(Filename, MachOOF.get());
215 }
217 static void DisassembleInputMachO2(StringRef Filename,
218 MachOObjectFile *MachOOF) {
219 const Target *TheTarget = GetTarget(MachOOF);
220 if (!TheTarget) {
221 // GetTarget prints out stuff.
222 return;
223 }
224 OwningPtr<const MCInstrInfo> InstrInfo(TheTarget->createMCInstrInfo());
225 OwningPtr<MCInstrAnalysis>
226 InstrAnalysis(TheTarget->createMCInstrAnalysis(InstrInfo.get()));
228 // Set up disassembler.
229 OwningPtr<const MCRegisterInfo> MRI(TheTarget->createMCRegInfo(TripleName));
230 OwningPtr<const MCAsmInfo> AsmInfo(
231 TheTarget->createMCAsmInfo(*MRI, TripleName));
232 OwningPtr<const MCSubtargetInfo>
233 STI(TheTarget->createMCSubtargetInfo(TripleName, "", ""));
234 OwningPtr<const MCDisassembler> DisAsm(TheTarget->createMCDisassembler(*STI));
235 int AsmPrinterVariant = AsmInfo->getAssemblerDialect();
236 OwningPtr<MCInstPrinter>
237 IP(TheTarget->createMCInstPrinter(AsmPrinterVariant, *AsmInfo, *InstrInfo,
238 *MRI, *STI));
240 if (!InstrAnalysis || !AsmInfo || !STI || !DisAsm || !IP) {
241 errs() << "error: couldn't initialize disassembler for target "
242 << TripleName << '\n';
243 return;
244 }
246 outs() << '\n' << Filename << ":\n\n";
248 MachO::mach_header Header = MachOOF->getHeader();
250 // FIXME: FoundFns isn't used anymore. Using symbols/LC_FUNCTION_STARTS to
251 // determine function locations will eventually go in MCObjectDisassembler.
252 // FIXME: Using the -cfg command line option, this code used to be able to
253 // annotate relocations with the referenced symbol's name, and if this was
254 // inside a __[cf]string section, the data it points to. This is now replaced
255 // by the upcoming MCSymbolizer, which needs the appropriate setup done above.
256 std::vector<SectionRef> Sections;
257 std::vector<SymbolRef> Symbols;
258 SmallVector<uint64_t, 8> FoundFns;
259 uint64_t BaseSegmentAddress;
261 getSectionsAndSymbols(Header, MachOOF, Sections, Symbols, FoundFns,
262 BaseSegmentAddress);
264 // Sort the symbols by address, just in case they didn't come in that way.
265 std::sort(Symbols.begin(), Symbols.end(), SymbolSorter());
267 // Build a data in code table that is sorted on by the address of each entry.
268 uint64_t BaseAddress = 0;
269 if (Header.filetype == MachO::MH_OBJECT)
270 Sections[0].getAddress(BaseAddress);
271 else
272 BaseAddress = BaseSegmentAddress;
273 DiceTable Dices;
274 for (dice_iterator DI = MachOOF->begin_dices(), DE = MachOOF->end_dices();
275 DI != DE; ++DI) {
276 uint32_t Offset;
277 DI->getOffset(Offset);
278 Dices.push_back(std::make_pair(BaseAddress + Offset, *DI));
279 }
280 array_pod_sort(Dices.begin(), Dices.end());
282 #ifndef NDEBUG
283 raw_ostream &DebugOut = DebugFlag ? dbgs() : nulls();
284 #else
285 raw_ostream &DebugOut = nulls();
286 #endif
288 OwningPtr<DIContext> diContext;
289 ObjectFile *DbgObj = MachOOF;
290 // Try to find debug info and set up the DIContext for it.
291 if (UseDbg) {
292 // A separate DSym file path was specified, parse it as a macho file,
293 // get the sections and supply it to the section name parsing machinery.
294 if (!DSYMFile.empty()) {
295 OwningPtr<MemoryBuffer> Buf;
296 if (error_code ec = MemoryBuffer::getFileOrSTDIN(DSYMFile, Buf)) {
297 errs() << "llvm-objdump: " << Filename << ": " << ec.message() << '\n';
298 return;
299 }
300 DbgObj = ObjectFile::createMachOObjectFile(Buf.release()).get();
301 }
303 // Setup the DIContext
304 diContext.reset(DIContext::getDWARFContext(DbgObj));
305 }
307 for (unsigned SectIdx = 0; SectIdx != Sections.size(); SectIdx++) {
309 bool SectIsText = false;
310 Sections[SectIdx].isText(SectIsText);
311 if (SectIsText == false)
312 continue;
314 StringRef SectName;
315 if (Sections[SectIdx].getName(SectName) ||
316 SectName != "__text")
317 continue; // Skip non-text sections
319 DataRefImpl DR = Sections[SectIdx].getRawDataRefImpl();
321 StringRef SegmentName = MachOOF->getSectionFinalSegmentName(DR);
322 if (SegmentName != "__TEXT")
323 continue;
325 StringRef Bytes;
326 Sections[SectIdx].getContents(Bytes);
327 StringRefMemoryObject memoryObject(Bytes);
328 bool symbolTableWorked = false;
330 // Parse relocations.
331 std::vector<std::pair<uint64_t, SymbolRef> > Relocs;
332 for (relocation_iterator RI = Sections[SectIdx].relocation_begin(),
333 RE = Sections[SectIdx].relocation_end();
334 RI != RE; ++RI) {
335 uint64_t RelocOffset, SectionAddress;
336 RI->getOffset(RelocOffset);
337 Sections[SectIdx].getAddress(SectionAddress);
338 RelocOffset -= SectionAddress;
340 symbol_iterator RelocSym = RI->getSymbol();
342 Relocs.push_back(std::make_pair(RelocOffset, *RelocSym));
343 }
344 array_pod_sort(Relocs.begin(), Relocs.end());
346 // Disassemble symbol by symbol.
347 for (unsigned SymIdx = 0; SymIdx != Symbols.size(); SymIdx++) {
348 StringRef SymName;
349 Symbols[SymIdx].getName(SymName);
351 SymbolRef::Type ST;
352 Symbols[SymIdx].getType(ST);
353 if (ST != SymbolRef::ST_Function)
354 continue;
356 // Make sure the symbol is defined in this section.
357 bool containsSym = false;
358 Sections[SectIdx].containsSymbol(Symbols[SymIdx], containsSym);
359 if (!containsSym)
360 continue;
362 // Start at the address of the symbol relative to the section's address.
363 uint64_t SectionAddress = 0;
364 uint64_t Start = 0;
365 Sections[SectIdx].getAddress(SectionAddress);
366 Symbols[SymIdx].getAddress(Start);
367 Start -= SectionAddress;
369 // Stop disassembling either at the beginning of the next symbol or at
370 // the end of the section.
371 bool containsNextSym = false;
372 uint64_t NextSym = 0;
373 uint64_t NextSymIdx = SymIdx+1;
374 while (Symbols.size() > NextSymIdx) {
375 SymbolRef::Type NextSymType;
376 Symbols[NextSymIdx].getType(NextSymType);
377 if (NextSymType == SymbolRef::ST_Function) {
378 Sections[SectIdx].containsSymbol(Symbols[NextSymIdx],
379 containsNextSym);
380 Symbols[NextSymIdx].getAddress(NextSym);
381 NextSym -= SectionAddress;
382 break;
383 }
384 ++NextSymIdx;
385 }
387 uint64_t SectSize;
388 Sections[SectIdx].getSize(SectSize);
389 uint64_t End = containsNextSym ? NextSym : SectSize;
390 uint64_t Size;
392 symbolTableWorked = true;
394 outs() << SymName << ":\n";
395 DILineInfo lastLine;
396 for (uint64_t Index = Start; Index < End; Index += Size) {
397 MCInst Inst;
399 uint64_t SectAddress = 0;
400 Sections[SectIdx].getAddress(SectAddress);
401 outs() << format("%8" PRIx64 ":\t", SectAddress + Index);
403 // Check the data in code table here to see if this is data not an
404 // instruction to be disassembled.
405 DiceTable Dice;
406 Dice.push_back(std::make_pair(SectAddress + Index, DiceRef()));
407 dice_table_iterator DTI = std::search(Dices.begin(), Dices.end(),
408 Dice.begin(), Dice.end(),
409 compareDiceTableEntries);
410 if (DTI != Dices.end()){
411 uint16_t Length;
412 DTI->second.getLength(Length);
413 DumpBytes(StringRef(Bytes.data() + Index, Length));
414 uint16_t Kind;
415 DTI->second.getKind(Kind);
416 DumpDataInCode(Bytes.data() + Index, Length, Kind);
417 continue;
418 }
420 if (DisAsm->getInstruction(Inst, Size, memoryObject, Index,
421 DebugOut, nulls())) {
422 DumpBytes(StringRef(Bytes.data() + Index, Size));
423 IP->printInst(&Inst, outs(), "");
425 // Print debug info.
426 if (diContext) {
427 DILineInfo dli =
428 diContext->getLineInfoForAddress(SectAddress + Index);
429 // Print valid line info if it changed.
430 if (dli != lastLine && dli.getLine() != 0)
431 outs() << "\t## " << dli.getFileName() << ':'
432 << dli.getLine() << ':' << dli.getColumn();
433 lastLine = dli;
434 }
435 outs() << "\n";
436 } else {
437 errs() << "llvm-objdump: warning: invalid instruction encoding\n";
438 if (Size == 0)
439 Size = 1; // skip illegible bytes
440 }
441 }
442 }
443 if (!symbolTableWorked) {
444 // Reading the symbol table didn't work, disassemble the whole section.
445 uint64_t SectAddress;
446 Sections[SectIdx].getAddress(SectAddress);
447 uint64_t SectSize;
448 Sections[SectIdx].getSize(SectSize);
449 uint64_t InstSize;
450 for (uint64_t Index = 0; Index < SectSize; Index += InstSize) {
451 MCInst Inst;
453 if (DisAsm->getInstruction(Inst, InstSize, memoryObject, Index,
454 DebugOut, nulls())) {
455 outs() << format("%8" PRIx64 ":\t", SectAddress + Index);
456 DumpBytes(StringRef(Bytes.data() + Index, InstSize));
457 IP->printInst(&Inst, outs(), "");
458 outs() << "\n";
459 } else {
460 errs() << "llvm-objdump: warning: invalid instruction encoding\n";
461 if (InstSize == 0)
462 InstSize = 1; // skip illegible bytes
463 }
464 }
465 }
466 }
467 }