#include "llvm/Support/StringRefMemoryObject.h"
#include "llvm/Support/raw_ostream.h"
#include <map>
-#include <vector>
using namespace llvm;
using namespace object;
: Obj(Obj), Dis(Dis), MIA(MIA), MOS(0) {}
uint64_t MCObjectDisassembler::getEntrypoint() {
- error_code ec;
- for (symbol_iterator SI = Obj.begin_symbols(), SE = Obj.end_symbols();
- SI != SE; SI.increment(ec)) {
- if (ec)
- break;
+ for (const SymbolRef &Symbol : Obj.symbols()) {
StringRef Name;
- SI->getName(Name);
+ Symbol.getName(Name);
if (Name == "main" || Name == "_main") {
uint64_t Entrypoint;
- SI->getAddress(Entrypoint);
+ Symbol.getAddress(Entrypoint);
return getEffectiveLoadAddr(Entrypoint);
}
}
return ArrayRef<uint64_t>();
}
+MemoryObject *MCObjectDisassembler::getRegionFor(uint64_t Addr) {
+ // FIXME: Keep track of object sections.
+ return FallbackRegion.get();
+}
+
uint64_t MCObjectDisassembler::getEffectiveLoadAddr(uint64_t Addr) {
return Addr;
}
}
void MCObjectDisassembler::buildSectionAtoms(MCModule *Module) {
- error_code ec;
- for (section_iterator SI = Obj.begin_sections(),
- SE = Obj.end_sections();
- SI != SE;
- SI.increment(ec)) {
- if (ec) break;
-
- bool isText; SI->isText(isText);
- bool isData; SI->isData(isData);
+ for (const SectionRef &Section : Obj.sections()) {
+ bool isText;
+ Section.isText(isText);
+ bool isData;
+ Section.isData(isData);
if (!isData && !isText)
continue;
- uint64_t StartAddr; SI->getAddress(StartAddr);
- uint64_t SecSize; SI->getSize(SecSize);
+ uint64_t StartAddr;
+ Section.getAddress(StartAddr);
+ uint64_t SecSize;
+ Section.getSize(SecSize);
if (StartAddr == UnknownAddressOrSize || SecSize == UnknownAddressOrSize)
continue;
StartAddr = getEffectiveLoadAddr(StartAddr);
- StringRef Contents; SI->getContents(Contents);
+ StringRef Contents;
+ Section.getContents(Contents);
StringRefMemoryObject memoryObject(Contents, StartAddr);
// We don't care about things like non-file-backed sections yet.
continue;
uint64_t EndAddr = StartAddr + SecSize - 1;
- StringRef SecName; SI->getName(SecName);
+ StringRef SecName;
+ Section.getName(SecName);
if (isText) {
MCTextAtom *Text = 0;
Text->addInst(Inst, InstSize);
InvalidData = 0;
} else {
+ assert(InstSize && "getInstruction() consumed no bytes");
if (!InvalidData) {
Text = 0;
- InvalidData = Module->createDataAtom(CurAddr, EndAddr);
+ InvalidData = Module->createDataAtom(CurAddr, CurAddr+InstSize - 1);
}
- InvalidData->addData(Contents[Index]);
+ for (uint64_t I = 0; I < InstSize; ++I)
+ InvalidData->addData(Contents[Index+I]);
}
}
} else {
MCBasicBlock *BB;
BBInfoSetTy Succs;
BBInfoSetTy Preds;
+ MCObjectDisassembler::AddressSetTy SuccAddrs;
BBInfo() : Atom(0), BB(0) {}
};
}
+static void RemoveDupsFromAddressVector(MCObjectDisassembler::AddressSetTy &V) {
+ std::sort(V.begin(), V.end());
+ V.erase(std::unique(V.begin(), V.end()), V.end());
+}
+
void MCObjectDisassembler::buildCFG(MCModule *Module) {
typedef std::map<uint64_t, BBInfo> BBInfoByAddrTy;
BBInfoByAddrTy BBInfos;
- typedef std::vector<uint64_t> AddressSetTy;
AddressSetTy Splits;
AddressSetTy Calls;
- error_code ec;
- for (symbol_iterator SI = Obj.begin_symbols(), SE = Obj.end_symbols();
- SI != SE; SI.increment(ec)) {
- if (ec)
- break;
+ for (const SymbolRef &Symbol : Obj.symbols()) {
SymbolRef::Type SymType;
- SI->getType(SymType);
+ Symbol.getType(SymType);
if (SymType == SymbolRef::ST_Function) {
uint64_t SymAddr;
- SI->getAddress(SymAddr);
+ Symbol.getAddress(SymAddr);
SymAddr = getEffectiveLoadAddr(SymAddr);
Calls.push_back(SymAddr);
Splits.push_back(SymAddr);
}
}
- std::sort(Splits.begin(), Splits.end());
- Splits.erase(std::unique(Splits.begin(), Splits.end()), Splits.end());
-
- std::sort(Calls.begin(), Calls.end());
- Calls.erase(std::unique(Calls.begin(), Calls.end()), Calls.end());
+ RemoveDupsFromAddressVector(Splits);
+ RemoveDupsFromAddressVector(Calls);
// Split text atoms into basic block atoms.
for (AddressSetTy::const_iterator SI = Splits.begin(), SE = Splits.end();
}
}
+// Basic idea of the disassembly + discovery:
+//
+// start with the wanted address, insert it in the worklist
+// while worklist not empty, take next address in the worklist:
+// - check if atom exists there
+// - if middle of atom:
+// - split basic blocks referencing the atom
+// - look for an already encountered BBInfo (using a map<atom, bbinfo>)
+// - if there is, split it (new one, fallthrough, move succs, etc..)
+// - if start of atom: nothing else to do
+// - if no atom: create new atom and new bbinfo
+// - look at the last instruction in the atom, add succs to worklist
+// for all elements in the worklist:
+// - create basic block, update preds/succs, etc..
+//
+MCBasicBlock *MCObjectDisassembler::getBBAt(MCModule *Module, MCFunction *MCFN,
+ uint64_t BBBeginAddr,
+ AddressSetTy &CallTargets,
+ AddressSetTy &TailCallTargets) {
+ typedef std::map<uint64_t, BBInfo> BBInfoByAddrTy;
+ typedef SmallSetVector<uint64_t, 16> AddrWorklistTy;
+ BBInfoByAddrTy BBInfos;
+ AddrWorklistTy Worklist;
+
+ Worklist.insert(BBBeginAddr);
+ for (size_t wi = 0; wi < Worklist.size(); ++wi) {
+ const uint64_t BeginAddr = Worklist[wi];
+ BBInfo *BBI = &BBInfos[BeginAddr];
+
+ MCTextAtom *&TA = BBI->Atom;
+ assert(!TA && "Discovered basic block already has an associated atom!");
+
+ // Look for an atom at BeginAddr.
+ if (MCAtom *A = Module->findAtomContaining(BeginAddr)) {
+ // FIXME: We don't care about mixed atoms, see above.
+ TA = cast<MCTextAtom>(A);
+
+ // The found atom doesn't begin at BeginAddr, we have to split it.
+ if (TA->getBeginAddr() != BeginAddr) {
+ // FIXME: Handle overlapping atoms: middle-starting instructions, etc..
+ MCTextAtom *NewTA = TA->split(BeginAddr);
+
+ // Look for an already encountered basic block that needs splitting
+ BBInfoByAddrTy::iterator It = BBInfos.find(TA->getBeginAddr());
+ if (It != BBInfos.end() && It->second.Atom) {
+ BBI->SuccAddrs = It->second.SuccAddrs;
+ It->second.SuccAddrs.clear();
+ It->second.SuccAddrs.push_back(BeginAddr);
+ }
+ TA = NewTA;
+ }
+ BBI->Atom = TA;
+ } else {
+ // If we didn't find an atom, then we have to disassemble to create one!
+
+ MemoryObject *Region = getRegionFor(BeginAddr);
+ if (!Region)
+ llvm_unreachable(("Couldn't find suitable region for disassembly at " +
+ utostr(BeginAddr)).c_str());
+
+ uint64_t InstSize;
+ uint64_t EndAddr = Region->getBase() + Region->getExtent();
+
+ // We want to stop before the next atom and have a fallthrough to it.
+ if (MCTextAtom *NextAtom =
+ cast_or_null<MCTextAtom>(Module->findFirstAtomAfter(BeginAddr)))
+ EndAddr = std::min(EndAddr, NextAtom->getBeginAddr());
+
+ for (uint64_t Addr = BeginAddr; Addr < EndAddr; Addr += InstSize) {
+ MCInst Inst;
+ if (Dis.getInstruction(Inst, InstSize, *Region, Addr, nulls(),
+ nulls())) {
+ if (!TA)
+ TA = Module->createTextAtom(Addr, Addr);
+ TA->addInst(Inst, InstSize);
+ } else {
+ // We don't care about splitting mixed atoms either.
+ llvm_unreachable("Couldn't disassemble instruction in atom.");
+ }
+
+ uint64_t BranchTarget;
+ if (MIA.evaluateBranch(Inst, Addr, InstSize, BranchTarget)) {
+ if (MIA.isCall(Inst))
+ CallTargets.push_back(BranchTarget);
+ }
+
+ if (MIA.isTerminator(Inst))
+ break;
+ }
+ BBI->Atom = TA;
+ }
+
+ assert(TA && "Couldn't disassemble atom, none was created!");
+ assert(TA->begin() != TA->end() && "Empty atom!");
+
+ MemoryObject *Region = getRegionFor(TA->getBeginAddr());
+ assert(Region && "Couldn't find region for already disassembled code!");
+ uint64_t EndRegion = Region->getBase() + Region->getExtent();
+
+ // Now we have a basic block atom, add successors.
+ // Add the fallthrough block.
+ if ((MIA.isConditionalBranch(TA->back().Inst) ||
+ !MIA.isTerminator(TA->back().Inst)) &&
+ (TA->getEndAddr() + 1 < EndRegion)) {
+ BBI->SuccAddrs.push_back(TA->getEndAddr() + 1);
+ Worklist.insert(TA->getEndAddr() + 1);
+ }
+
+ // If the terminator is a branch, add the target block.
+ if (MIA.isBranch(TA->back().Inst)) {
+ uint64_t BranchTarget;
+ if (MIA.evaluateBranch(TA->back().Inst, TA->back().Address,
+ TA->back().Size, BranchTarget)) {
+ StringRef ExtFnName;
+ if (MOS)
+ ExtFnName =
+ MOS->findExternalFunctionAt(getOriginalLoadAddr(BranchTarget));
+ if (!ExtFnName.empty()) {
+ TailCallTargets.push_back(BranchTarget);
+ CallTargets.push_back(BranchTarget);
+ } else {
+ BBI->SuccAddrs.push_back(BranchTarget);
+ Worklist.insert(BranchTarget);
+ }
+ }
+ }
+ }
+
+ for (size_t wi = 0, we = Worklist.size(); wi != we; ++wi) {
+ const uint64_t BeginAddr = Worklist[wi];
+ BBInfo *BBI = &BBInfos[BeginAddr];
+
+ assert(BBI->Atom && "Found a basic block without an associated atom!");
+
+ // Look for a basic block at BeginAddr.
+ BBI->BB = MCFN->find(BeginAddr);
+ if (BBI->BB) {
+ // FIXME: check that the succs/preds are the same
+ continue;
+ }
+ // If there was none, we have to create one from the atom.
+ BBI->BB = &MCFN->createBlock(*BBI->Atom);
+ }
+
+ for (size_t wi = 0, we = Worklist.size(); wi != we; ++wi) {
+ const uint64_t BeginAddr = Worklist[wi];
+ BBInfo *BBI = &BBInfos[BeginAddr];
+ MCBasicBlock *BB = BBI->BB;
+
+ RemoveDupsFromAddressVector(BBI->SuccAddrs);
+ for (AddressSetTy::const_iterator SI = BBI->SuccAddrs.begin(),
+ SE = BBI->SuccAddrs.end();
+ SE != SE; ++SI) {
+ MCBasicBlock *Succ = BBInfos[*SI].BB;
+ BB->addSuccessor(Succ);
+ Succ->addPredecessor(BB);
+ }
+ }
+
+ assert(BBInfos[Worklist[0]].BB &&
+ "No basic block created at requested address?");
+
+ return BBInfos[Worklist[0]].BB;
+}
+
+MCFunction *
+MCObjectDisassembler::createFunction(MCModule *Module, uint64_t BeginAddr,
+ AddressSetTy &CallTargets,
+ AddressSetTy &TailCallTargets) {
+ // First, check if this is an external function.
+ StringRef ExtFnName;
+ if (MOS)
+ ExtFnName = MOS->findExternalFunctionAt(getOriginalLoadAddr(BeginAddr));
+ if (!ExtFnName.empty())
+ return Module->createFunction(ExtFnName);
+
+ // If it's not, look for an existing function.
+ for (MCModule::func_iterator FI = Module->func_begin(),
+ FE = Module->func_end();
+ FI != FE; ++FI) {
+ if ((*FI)->empty())
+ continue;
+ // FIXME: MCModule should provide a findFunctionByAddr()
+ if ((*FI)->getEntryBlock()->getInsts()->getBeginAddr() == BeginAddr)
+ return *FI;
+ }
+
+ // Finally, just create a new one.
+ MCFunction *MCFN = Module->createFunction("");
+ getBBAt(Module, MCFN, BeginAddr, CallTargets, TailCallTargets);
+ return MCFN;
+}
+
// MachO MCObjectDisassembler implementation.
MCMachOObjectDisassembler::MCMachOObjectDisassembler(
: MCObjectDisassembler(MOOF, Dis, MIA), MOOF(MOOF),
VMAddrSlide(VMAddrSlide), HeaderLoadAddress(HeaderLoadAddress) {
- error_code ec;
- for (section_iterator SI = MOOF.begin_sections(), SE = MOOF.end_sections();
- SI != SE; SI.increment(ec)) {
- if (ec)
- break;
+ for (const SectionRef &Section : MOOF.sections()) {
StringRef Name;
- SI->getName(Name);
+ Section.getName(Name);
// FIXME: We should use the S_ section type instead of the name.
if (Name == "__mod_init_func") {
DEBUG(dbgs() << "Found __mod_init_func section!\n");
- SI->getContents(ModInitContents);
+ Section.getContents(ModInitContents);
} else if (Name == "__mod_exit_func") {
DEBUG(dbgs() << "Found __mod_exit_func section!\n");
- SI->getContents(ModExitContents);
+ Section.getContents(ModExitContents);
}
}
}
// Look for LC_MAIN.
{
- uint32_t LoadCommandCount = MOOF.getHeader().NumLoadCommands;
+ uint32_t LoadCommandCount = MOOF.getHeader().ncmds;
MachOObjectFile::LoadCommandInfo Load = MOOF.getFirstLoadCommandInfo();
for (unsigned I = 0;; ++I) {
- if (Load.C.Type == MachO::LoadCommandMain) {
+ if (Load.C.cmd == MachO::LC_MAIN) {
EntryFileOffset =
((const MachO::entry_point_command *)Load.Ptr)->entryoff;
break;