From a894c8e34453493a9d3fb2ffbbc21151c3965b63 Mon Sep 17 00:00:00 2001 From: Benjamin Kramer Date: Tue, 20 Sep 2011 17:53:01 +0000 Subject: [PATCH] llvm-objdump: factor code better, add comments. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140153 91177308-0d34-0410-b5e6-96231b3b80d8 --- tools/llvm-objdump/MCFunction.cpp | 34 ++-- tools/llvm-objdump/MCFunction.h | 6 +- tools/llvm-objdump/MachODump.cpp | 302 ++++++++++++++++-------------- 3 files changed, 183 insertions(+), 159 deletions(-) diff --git a/tools/llvm-objdump/MCFunction.cpp b/tools/llvm-objdump/MCFunction.cpp index 03390c2b586..5c67f1b70a9 100644 --- a/tools/llvm-objdump/MCFunction.cpp +++ b/tools/llvm-objdump/MCFunction.cpp @@ -47,44 +47,39 @@ MCFunction::createFunctionFromMC(StringRef Name, const MCDisassembler *DisAsm, while (!WorkList.empty()) { uint64_t Index = WorkList.pop_back_val(); if (VisitedInsts.find(Index) != VisitedInsts.end()) - continue; + continue; // Already visited this location. for (;Index < End; Index += Size) { - MCInst Inst; + VisitedInsts.insert(Index); + MCInst Inst; if (DisAsm->getInstruction(Inst, Size, Region, Index, DebugOut, nulls())){ + Instructions.push_back(MCDecodedInst(Index, Size, Inst)); if (Ana->isBranch(Inst)) { uint64_t targ = Ana->evaluateBranch(Inst, Index, Size); - if (targ != -1ULL && targ == Index+Size) { - Instructions.push_back(MCDecodedInst(Index, Size, Inst)); - VisitedInsts.insert(Index); - continue; - } + if (targ != -1ULL && targ == Index+Size) + continue; // Skip nop jumps. + + // If we could determine the branch target, make a note to start a + // new basic block there and add the target to the worklist. if (targ != -1ULL) { Splits.insert(targ); WorkList.push_back(targ); WorkList.push_back(Index+Size); } Splits.insert(Index+Size); - Instructions.push_back(MCDecodedInst(Index, Size, Inst)); - VisitedInsts.insert(Index); break; } else if (Ana->isReturn(Inst)) { + // Return instruction. This basic block ends here. Splits.insert(Index+Size); - Instructions.push_back(MCDecodedInst(Index, Size, Inst)); - VisitedInsts.insert(Index); break; } else if (Ana->isCall(Inst)) { uint64_t targ = Ana->evaluateBranch(Inst, Index, Size); - if (targ != -1ULL && targ != Index+Size) { + // Add the call to the call list if the destination is known. + if (targ != -1ULL && targ != Index+Size) Calls.push_back(targ); - } } - - Instructions.push_back(MCDecodedInst(Index, Size, Inst)); - VisitedInsts.insert(Index); } else { - VisitedInsts.insert(Index); errs().write_hex(Index) << ": warning: invalid instruction encoding\n"; if (Size == 0) Size = 1; // skip illegible bytes @@ -93,9 +88,10 @@ MCFunction::createFunctionFromMC(StringRef Name, const MCDisassembler *DisAsm, } } + // Make sure the instruction list is sorted. std::sort(Instructions.begin(), Instructions.end()); - // Create basic blocks. + // Create basic blocks. unsigned ii = 0, ie = Instructions.size(); for (std::set::iterator spi = Splits.begin(), spe = llvm::prior(Splits.end()); spi != spe; ++spi) { @@ -115,7 +111,7 @@ MCFunction::createFunctionFromMC(StringRef Name, const MCDisassembler *DisAsm, // Calculate successors of each block. for (MCFunction::iterator i = f.begin(), e = f.end(); i != e; ++i) { - MCBasicBlock &BB = i->second; + MCBasicBlock &BB = const_cast(i->second); if (BB.getInsts().empty()) continue; const MCDecodedInst &Inst = BB.getInsts().back(); diff --git a/tools/llvm-objdump/MCFunction.h b/tools/llvm-objdump/MCFunction.h index 4677d91bb77..6d3a548d48e 100644 --- a/tools/llvm-objdump/MCFunction.h +++ b/tools/llvm-objdump/MCFunction.h @@ -83,9 +83,9 @@ public: const MCInstrAnalysis *Ana, raw_ostream &DebugOut, SmallVectorImpl &Calls); - typedef MapTy::iterator iterator; - iterator begin() { return Blocks.begin(); } - iterator end() { return Blocks.end(); } + typedef MapTy::const_iterator iterator; + iterator begin() const { return Blocks.begin(); } + iterator end() const { return Blocks.end(); } StringRef getName() const { return Name; } diff --git a/tools/llvm-objdump/MachODump.cpp b/tools/llvm-objdump/MachODump.cpp index afb765d4afd..a0fff215ea8 100644 --- a/tools/llvm-objdump/MachODump.cpp +++ b/tools/llvm-objdump/MachODump.cpp @@ -94,6 +94,29 @@ struct Symbol { bool operator<(const Symbol &RHS) const { return Value < RHS.Value; } }; + +template +static Section copySection(const T &Sect) { + Section S; + memcpy(S.Name, Sect->Name, 16); + S.Address = Sect->Address; + S.Size = Sect->Size; + S.Offset = Sect->Offset; + S.NumRelocs = Sect->NumRelocationTableEntries; + S.RelocTableOffset = Sect->RelocationTableOffset; + return S; +} + +template +static Symbol copySymbol(const T &STE) { + Symbol S; + S.StringIndex = STE->StringIndex; + S.SectionIndex = STE->SectionIndex; + S.Value = STE->Value; + return S; +} + +// Print addtitional information about an address, if available. static void DumpAddress(uint64_t Address, ArrayRef
Sections, MachOObject *MachOObj, raw_ostream &OS) { for (unsigned i = 0; i != Sections.size(); ++i) { @@ -102,14 +125,86 @@ static void DumpAddress(uint64_t Address, ArrayRef
Sections, Sections[i].Address + Sections[i].Size > Address) { StringRef bytes = MachOObj->getData(Sections[i].Offset, Sections[i].Size); + // Print constant strings. if (!strcmp(Sections[i].Name, "__cstring")) OS << '"' << bytes.substr(addr, bytes.find('\0', addr)) << '"'; + // Print constant CFStrings. if (!strcmp(Sections[i].Name, "__cfstring")) OS << "@\"" << bytes.substr(addr, bytes.find('\0', addr)) << '"'; } } } +typedef std::map FunctionMapTy; +typedef SmallVector FunctionListTy; +static void createMCFunctionAndSaveCalls(StringRef Name, + const MCDisassembler *DisAsm, + MemoryObject &Object, uint64_t Start, + uint64_t End, + MCInstrAnalysis *InstrAnalysis, + uint64_t Address, + raw_ostream &DebugOut, + FunctionMapTy &FunctionMap, + FunctionListTy &Functions) { + SmallVector Calls; + MCFunction f = + MCFunction::createFunctionFromMC(Name, DisAsm, Object, Start, End, + InstrAnalysis, DebugOut, Calls); + Functions.push_back(f); + FunctionMap[Address] = &Functions.back(); + + // Add the gathered callees to the map. + for (unsigned i = 0, e = Calls.size(); i != e; ++i) + FunctionMap.insert(std::make_pair(Calls[i], (MCFunction*)0)); +} + +// Write a graphviz file for the CFG inside an MCFunction. +static void emitDOTFile(const char *FileName, const MCFunction &f, + MCInstPrinter *IP) { + // Start a new dot file. + std::string Error; + raw_fd_ostream Out(FileName, Error); + if (!Error.empty()) { + errs() << "llvm-objdump: warning: " << Error << '\n'; + return; + } + + Out << "digraph " << f.getName() << " {\n"; + Out << "graph [ rankdir = \"LR\" ];\n"; + for (MCFunction::iterator i = f.begin(), e = f.end(); i != e; ++i) { + bool hasPreds = false; + // Only print blocks that have predecessors. + // FIXME: Slow. + for (MCFunction::iterator pi = f.begin(), pe = f.end(); pi != pe; + ++pi) + if (pi->second.contains(i->first)) { + hasPreds = true; + break; + } + + if (!hasPreds && i != f.begin()) + continue; + + Out << '"' << i->first << "\" [ label=\""; + // Print instructions. + for (unsigned ii = 0, ie = i->second.getInsts().size(); ii != ie; + ++ii) { + // Escape special chars and print the instruction in mnemonic form. + std::string Str; + raw_string_ostream OS(Str); + IP->printInst(&i->second.getInsts()[ii].Inst, OS, ""); + Out << DOT::EscapeString(OS.str()) << '|'; + } + Out << "\" shape=\"record\" ];\n"; + + // Add edges. + for (MCBasicBlock::succ_iterator si = i->second.succ_begin(), + se = i->second.succ_end(); si != se; ++si) + Out << i->first << ":o -> " << *si <<":a\n"; + } + Out << "}\n"; +} + void llvm::DisassembleInputMachO(StringRef Filename) { OwningPtr Buff; @@ -131,44 +226,28 @@ void llvm::DisassembleInputMachO(StringRef Filename) { // Set up disassembler. OwningPtr AsmInfo(TheTarget->createMCAsmInfo(TripleName)); - - if (!AsmInfo) { - errs() << "error: no assembly info for target " << TripleName << "\n"; - return; - } - OwningPtr STI(TheTarget->createMCSubtargetInfo(TripleName, "", "")); - - if (!STI) { - errs() << "error: no subtarget info for target " << TripleName << "\n"; - return; - } - OwningPtr DisAsm(TheTarget->createMCDisassembler(*STI)); - if (!DisAsm) { - errs() << "error: no disassembler for target " << TripleName << "\n"; - return; - } - int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); OwningPtr IP(TheTarget->createMCInstPrinter( - AsmPrinterVariant, *AsmInfo, *STI)); - if (!IP) { - errs() << "error: no instruction printer for target " << TripleName << '\n'; + AsmPrinterVariant, *AsmInfo, *STI)); + + if (!InstrAnalysis || !AsmInfo || !STI || !DisAsm || !IP) { + errs() << "error: couldn't initialize disassmbler for target " + << TripleName << '\n'; return; } - outs() << '\n'; - outs() << Filename << ":\n\n"; + outs() << '\n' << Filename << ":\n\n"; const macho::Header &Header = MachOObj->getHeader(); const MachOObject::LoadCommandInfo *SymtabLCI = 0; + // First, find the symbol table segment. for (unsigned i = 0; i != Header.NumLoadCommands; ++i) { const MachOObject::LoadCommandInfo &LCI = MachOObj->getLoadCommandInfo(i); - switch (LCI.Command.Type) { - case macho::LCT_Symtab: + if (LCI.Command.Type == macho::LCT_Symtab) { SymtabLCI = &LCI; break; } @@ -184,34 +263,24 @@ void llvm::DisassembleInputMachO(StringRef Filename) { std::vector UnsortedSymbols; // FIXME: duplication SmallVector FoundFns; + // Make a list of all symbols in the object file. for (unsigned i = 0; i != Header.NumLoadCommands; ++i) { const MachOObject::LoadCommandInfo &LCI = MachOObj->getLoadCommandInfo(i); if (LCI.Command.Type == macho::LCT_Segment) { InMemoryStruct SegmentLC; MachOObj->ReadSegmentLoadCommand(LCI, SegmentLC); + // Store the sections in this segment. for (unsigned SectNum = 0; SectNum != SegmentLC->NumSections; ++SectNum) { InMemoryStruct Sect; MachOObj->ReadSection(LCI, SectNum, Sect); + Sections.push_back(copySection(Sect)); - Section S; - memcpy(S.Name, Sect->Name, 16); - S.Address = Sect->Address; - S.Size = Sect->Size; - S.Offset = Sect->Offset; - S.NumRelocs = Sect->NumRelocationTableEntries; - S.RelocTableOffset = Sect->RelocationTableOffset; - Sections.push_back(S); - + // Store the symbols in this section. for (unsigned i = 0; i != SymtabLC->NumSymbolTableEntries; ++i) { InMemoryStruct STE; MachOObj->ReadSymbolTableEntry(SymtabLC->SymbolTableOffset, i, STE); - - Symbol S; - S.StringIndex = STE->StringIndex; - S.SectionIndex = STE->SectionIndex; - S.Value = STE->Value; - Symbols.push_back(S); + Symbols.push_back(copySymbol(STE)); UnsortedSymbols.push_back(Symbols.back()); } } @@ -219,32 +288,24 @@ void llvm::DisassembleInputMachO(StringRef Filename) { InMemoryStruct Segment64LC; MachOObj->ReadSegment64LoadCommand(LCI, Segment64LC); - for (unsigned SectNum = 0; SectNum != Segment64LC->NumSections; ++SectNum) { + // Store the sections in this segment. + for (unsigned SectNum = 0; SectNum != Segment64LC->NumSections; + ++SectNum) { InMemoryStruct Sect64; MachOObj->ReadSection64(LCI, SectNum, Sect64); + Sections.push_back(copySection(Sect64)); - Section S; - memcpy(S.Name, Sect64->Name, 16); - S.Address = Sect64->Address; - S.Size = Sect64->Size; - S.Offset = Sect64->Offset; - S.NumRelocs = Sect64->NumRelocationTableEntries; - S.RelocTableOffset = Sect64->RelocationTableOffset; - Sections.push_back(S); - + // Store the symbols in this section. for (unsigned i = 0; i != SymtabLC->NumSymbolTableEntries; ++i) { InMemoryStruct STE; MachOObj->ReadSymbol64TableEntry(SymtabLC->SymbolTableOffset, i, STE); - - Symbol S; - S.StringIndex = STE->StringIndex; - S.SectionIndex = STE->SectionIndex; - S.Value = STE->Value; - Symbols.push_back(S); + Symbols.push_back(copySymbol(STE)); UnsortedSymbols.push_back(Symbols.back()); } } } else if (LCI.Command.Type == macho::LCT_FunctionStarts) { + // We found a function starts segment, parse the addresses for later + // consumption. InMemoryStruct LLC; MachOObj->ReadLinkeditDataLoadCommand(LCI, LLC); @@ -252,7 +313,6 @@ void llvm::DisassembleInputMachO(StringRef Filename) { } } - std::map FunctionMap; // Sort the symbols by address, just in case they didn't come in that way. array_pod_sort(Symbols.begin(), Symbols.end()); @@ -263,12 +323,14 @@ void llvm::DisassembleInputMachO(StringRef Filename) { raw_ostream &DebugOut = nulls(); #endif - SmallVector Functions; + FunctionMapTy FunctionMap; + FunctionListTy Functions; for (unsigned SectIdx = 0; SectIdx != Sections.size(); SectIdx++) { if (strcmp(Sections[SectIdx].Name, "__text")) - continue; + continue; // Skip non-text sections + // Insert the functions from the function starts segment into our map. uint64_t VMAddr = Sections[SectIdx].Address - Sections[SectIdx].Offset; for (unsigned i = 0, e = FoundFns.size(); i != e; ++i) FunctionMap.insert(std::make_pair(FoundFns[i]+VMAddr, (MCFunction*)0)); @@ -278,6 +340,7 @@ void llvm::DisassembleInputMachO(StringRef Filename) { StringRefMemoryObject memoryObject(Bytes); bool symbolTableWorked = false; + // Parse relocations. std::vector > Relocs; for (unsigned j = 0; j != Sections[SectIdx].NumRelocs; ++j) { InMemoryStruct RE; @@ -286,11 +349,16 @@ void llvm::DisassembleInputMachO(StringRef Filename) { } array_pod_sort(Relocs.begin(), Relocs.end()); + // Disassemble symbol by symbol. for (unsigned SymIdx = 0; SymIdx != Symbols.size(); SymIdx++) { + // Make sure the symbol is defined in this section. if ((unsigned)Symbols[SymIdx].SectionIndex - 1 != SectIdx) continue; + // Start at the address of the symbol relative to the section's address. uint64_t Start = Symbols[SymIdx].Value - Sections[SectIdx].Address; + // Stop disassembling either at the beginning of the next symbol or at + // the end of the section. uint64_t End = (SymIdx+1 == Symbols.size() || Symbols[SymIdx].SectionIndex != Symbols[SymIdx+1].SectionIndex) ? Sections[SectIdx].Size : @@ -303,6 +371,7 @@ void llvm::DisassembleInputMachO(StringRef Filename) { symbolTableWorked = true; if (!CFG) { + // Normal disassembly, print addresses, bytes and mnemonic form. outs() << MachOObj->getStringAtIndex(Symbols[SymIdx].StringIndex) << ":\n"; for (uint64_t Index = Start; Index < End; Index += Size) { @@ -322,42 +391,27 @@ void llvm::DisassembleInputMachO(StringRef Filename) { } } else { // Create CFG and use it for disassembly. - SmallVector Calls; - MCFunction f = - MCFunction::createFunctionFromMC( - MachOObj->getStringAtIndex(Symbols[SymIdx].StringIndex), - DisAsm.get(), - memoryObject, Start, End, - InstrAnalysis.get(), DebugOut, - Calls); - - Functions.push_back(f); - FunctionMap[Start] = &Functions.back(); - - for (unsigned i = 0, e = Calls.size(); i != e; ++i) - FunctionMap.insert(std::make_pair(Calls[i], (MCFunction*)0)); + createMCFunctionAndSaveCalls( + MachOObj->getStringAtIndex(Symbols[SymIdx].StringIndex), + DisAsm.get(), memoryObject, Start, End, InstrAnalysis.get(), + Start, DebugOut, FunctionMap, Functions); } } if (CFG) { if (!symbolTableWorked) { - // Create CFG and use it for disassembly. - SmallVector Calls; - MCFunction f = - MCFunction::createFunctionFromMC("__TEXT", DisAsm.get(), - memoryObject, 0, Sections[SectIdx].Size, - InstrAnalysis.get(), DebugOut, - Calls); - - Functions.push_back(f); - FunctionMap[Sections[SectIdx].Offset] = &Functions.back(); - - for (unsigned i = 0, e = Calls.size(); i != e; ++i) - FunctionMap.insert(std::make_pair(Calls[i], (MCFunction*)0)); + // Reading the symbol table didn't work, create a big __TEXT symbol. + createMCFunctionAndSaveCalls("__TEXT", DisAsm.get(), memoryObject, + 0, Sections[SectIdx].Size, + InstrAnalysis.get(), + Sections[SectIdx].Offset, DebugOut, + FunctionMap, Functions); } for (std::map::iterator mi = FunctionMap.begin(), me = FunctionMap.end(); mi != me; ++mi) if (mi->second == 0) { + // Create functions for the remaining callees we have gathered, + // but we didn't find a name for them. SmallVector Calls; MCFunction f = MCFunction::createFunctionFromMC("unknown", DisAsm.get(), @@ -367,10 +421,11 @@ void llvm::DisassembleInputMachO(StringRef Filename) { Calls); Functions.push_back(f); mi->second = &Functions.back(); - for (unsigned i = 0, e = Calls.size(); i != e; ++i) - if (FunctionMap.insert(std::make_pair(Calls[i], (MCFunction*)0)) - .second) + for (unsigned i = 0, e = Calls.size(); i != e; ++i) { + std::pair p(Calls[i], (MCFunction*)0); + if (FunctionMap.insert(p).second) mi = FunctionMap.begin(); + } } DenseSet PrintedBlocks; @@ -378,10 +433,13 @@ void llvm::DisassembleInputMachO(StringRef Filename) { MCFunction &f = Functions[ffi]; for (MCFunction::iterator fi = f.begin(), fe = f.end(); fi != fe; ++fi){ if (!PrintedBlocks.insert(fi->first).second) - continue; + continue; // We already printed this block. + + // We assume a block has predecessors when it's the first block after + // a symbol. bool hasPreds = FunctionMap.find(fi->first) != FunctionMap.end(); - // Only print blocks that have predecessors. + // See if this block has predecessors. // FIXME: Slow. for (MCFunction::iterator pi = f.begin(), pe = f.end(); pi != pe; ++pi) @@ -390,8 +448,8 @@ void llvm::DisassembleInputMachO(StringRef Filename) { break; } - // Data block. - if (!hasPreds && fi != f.begin()) { + // No predecessors, this is a data block. Print as .byte directives. + if (!hasPreds) { uint64_t End = llvm::next(fi) == fe ? Sections[SectIdx].Size : llvm::next(fi)->first; outs() << "# " << End-fi->first << " bytes of data:\n"; @@ -403,23 +461,31 @@ void llvm::DisassembleInputMachO(StringRef Filename) { continue; } - if (fi->second.contains(fi->first)) + if (fi->second.contains(fi->first)) // Print a header for simple loops outs() << "# Loop begin:\n"; + // Walk over the instructions and print them. for (unsigned ii = 0, ie = fi->second.getInsts().size(); ii != ie; ++ii) { const MCDecodedInst &Inst = fi->second.getInsts()[ii]; + + // If there's a symbol at this address, print its name. if (FunctionMap.find(Sections[SectIdx].Address + Inst.Address) != FunctionMap.end()) outs() << FunctionMap[Sections[SectIdx].Address + Inst.Address]-> getName() << ":\n"; + outs() << format("%8llx:\t", Sections[SectIdx].Address + Inst.Address); DumpBytes(StringRef(Bytes.data() + Inst.Address, Inst.Size)); - // Simple loops. - if (fi->second.contains(fi->first)) + + if (fi->second.contains(fi->first)) // Indent simple loops. outs() << '\t'; + IP->printInst(&Inst.Inst, outs(), ""); + + // Look for relocations inside this instructions, if there is one + // print its target and additional information if availbable. for (unsigned j = 0; j != Relocs.size(); ++j) if (Relocs[j].first >= Sections[SectIdx].Address + Inst.Address && Relocs[j].first < Sections[SectIdx].Address + Inst.Address + @@ -431,6 +497,9 @@ void llvm::DisassembleInputMachO(StringRef Filename) { DumpAddress(UnsortedSymbols[Relocs[j].second].Value, Sections, MachOObj.get(), outs()); } + + // If this instructions contains an address, see if we can evaluate + // it and print additional information. uint64_t targ = InstrAnalysis->evaluateBranch(Inst.Inst, Inst.Address, Inst.Size); @@ -441,48 +510,7 @@ void llvm::DisassembleInputMachO(StringRef Filename) { } } - // Start a new dot file. - std::string Error; - raw_fd_ostream Out((f.getName().str() + ".dot").c_str(), Error); - if (!Error.empty()) { - errs() << "llvm-objdump: warning: " << Error << '\n'; - continue; - } - - Out << "digraph " << f.getName() << " {\n"; - Out << "graph [ rankdir = \"LR\" ];\n"; - for (MCFunction::iterator i = f.begin(), e = f.end(); i != e; ++i) { - bool hasPreds = false; - // Only print blocks that have predecessors. - // FIXME: Slow. - for (MCFunction::iterator pi = f.begin(), pe = f.end(); pi != pe; - ++pi) - if (pi->second.contains(i->first)) { - hasPreds = true; - break; - } - - if (!hasPreds && i != f.begin()) - continue; - - Out << '"' << i->first << "\" [ label=\""; - // Print instructions. - for (unsigned ii = 0, ie = i->second.getInsts().size(); ii != ie; - ++ii) { - // Escape special chars and print the instruction in mnemonic form. - std::string Str; - raw_string_ostream OS(Str); - IP->printInst(&i->second.getInsts()[ii].Inst, OS, ""); - Out << DOT::EscapeString(OS.str()) << '|'; - } - Out << "\" shape=\"record\" ];\n"; - - // Add edges. - for (MCBasicBlock::succ_iterator si = i->second.succ_begin(), - se = i->second.succ_end(); si != se; ++si) - Out << i->first << ":o -> " << *si <<":a\n"; - } - Out << "}\n"; + emitDOTFile((f.getName().str() + ".dot").c_str(), f, IP.get()); } } } -- 2.34.1