-//===-- RuntimeDyldMachO.cpp - Run-time dynamic linker for MC-JIT ------*- C++ -*-===//
+//===-- RuntimeDyldMachO.cpp - Run-time dynamic linker for MC-JIT -*- C++ -*-=//
//
// The LLVM Compiler Infrastructure
//
//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "dyld"
-#include "llvm/ADT/OwningPtr.h"
-#include "llvm/ADT/StringRef.h"
+#include "RuntimeDyldMachO.h"
#include "llvm/ADT/STLExtras.h"
-#include "RuntimeDyldImpl.h"
+#include "llvm/ADT/StringRef.h"
+
+#include "Targets/RuntimeDyldMachOARM.h"
+#include "Targets/RuntimeDyldMachOAArch64.h"
+#include "Targets/RuntimeDyldMachOI386.h"
+#include "Targets/RuntimeDyldMachOX86_64.h"
+
using namespace llvm;
using namespace llvm::object;
-namespace llvm {
+#define DEBUG_TYPE "dyld"
-bool RuntimeDyldMachO::
-resolveRelocation(uint8_t *Address, uint8_t *Value, bool isPCRel,
- unsigned Type, unsigned Size) {
- // This just dispatches to the proper target specific routine.
- switch (CPUType) {
- default: assert(0 && "Unsupported CPU type!");
- case mach::CTM_x86_64:
- return resolveX86_64Relocation((uintptr_t)Address, (uintptr_t)Value,
- isPCRel, Type, Size);
- case mach::CTM_ARM:
- return resolveARMRelocation((uintptr_t)Address, (uintptr_t)Value,
- isPCRel, Type, Size);
- }
- llvm_unreachable("");
-}
+namespace {
-bool RuntimeDyldMachO::
-resolveX86_64Relocation(uintptr_t Address, uintptr_t Value,
- bool isPCRel, unsigned Type,
- unsigned Size) {
- // If the relocation is PC-relative, the value to be encoded is the
- // pointer difference.
- if (isPCRel)
- // FIXME: It seems this value needs to be adjusted by 4 for an effective PC
- // address. Is that expected? Only for branches, perhaps?
- Value -= Address + 4;
-
- switch(Type) {
- default:
- llvm_unreachable("Invalid relocation type!");
- case macho::RIT_X86_64_Unsigned:
- case macho::RIT_X86_64_Branch: {
- // Mask in the target value a byte at a time (we don't have an alignment
- // guarantee for the target address, so this is safest).
- uint8_t *p = (uint8_t*)Address;
- for (unsigned i = 0; i < Size; ++i) {
- *p++ = (uint8_t)Value;
- Value >>= 8;
- }
- return false;
- }
- case macho::RIT_X86_64_Signed:
- case macho::RIT_X86_64_GOTLoad:
- case macho::RIT_X86_64_GOT:
- case macho::RIT_X86_64_Subtractor:
- case macho::RIT_X86_64_Signed1:
- case macho::RIT_X86_64_Signed2:
- case macho::RIT_X86_64_Signed4:
- case macho::RIT_X86_64_TLV:
- return Error("Relocation type not implemented yet!");
- }
- return false;
-}
+class LoadedMachOObjectInfo : public RuntimeDyld::LoadedObjectInfo {
+public:
+ LoadedMachOObjectInfo(RuntimeDyldImpl &RTDyld, unsigned BeginIdx,
+ unsigned EndIdx)
+ : RuntimeDyld::LoadedObjectInfo(RTDyld, BeginIdx, EndIdx) {}
-bool RuntimeDyldMachO::resolveARMRelocation(uintptr_t Address, uintptr_t Value,
- bool isPCRel, unsigned Type,
- unsigned Size) {
- // If the relocation is PC-relative, the value to be encoded is the
- // pointer difference.
- if (isPCRel) {
- Value -= Address;
- // ARM PCRel relocations have an effective-PC offset of two instructions
- // (four bytes in Thumb mode, 8 bytes in ARM mode).
- // FIXME: For now, assume ARM mode.
- Value -= 8;
+ OwningBinary<ObjectFile>
+ getObjectForDebug(const ObjectFile &Obj) const override {
+ return OwningBinary<ObjectFile>();
}
+};
- switch(Type) {
- default:
- llvm_unreachable("Invalid relocation type!");
- case macho::RIT_Vanilla: {
- llvm_unreachable("Invalid relocation type!");
- // Mask in the target value a byte at a time (we don't have an alignment
- // guarantee for the target address, so this is safest).
- uint8_t *p = (uint8_t*)Address;
- for (unsigned i = 0; i < Size; ++i) {
- *p++ = (uint8_t)Value;
- Value >>= 8;
- }
- break;
- }
- case macho::RIT_ARM_Branch24Bit: {
- // Mask the value into the target address. We know instructions are
- // 32-bit aligned, so we can do it all at once.
- uint32_t *p = (uint32_t*)Address;
- // The low two bits of the value are not encoded.
- Value >>= 2;
- // Mask the value to 24 bits.
- Value &= 0xffffff;
- // FIXME: If the destination is a Thumb function (and the instruction
- // is a non-predicated BL instruction), we need to change it to a BLX
- // instruction instead.
-
- // Insert the value into the instruction.
- *p = (*p & ~0xffffff) | Value;
- break;
- }
- case macho::RIT_ARM_ThumbBranch22Bit:
- case macho::RIT_ARM_ThumbBranch32Bit:
- case macho::RIT_ARM_Half:
- case macho::RIT_ARM_HalfDifference:
- case macho::RIT_Pair:
- case macho::RIT_Difference:
- case macho::RIT_ARM_LocalDifference:
- case macho::RIT_ARM_PreboundLazyPointer:
- return Error("Relocation type not implemented yet!");
- }
- return false;
}
-bool RuntimeDyldMachO::
-loadSegment32(const MachOObject *Obj,
- const MachOObject::LoadCommandInfo *SegmentLCI,
- const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) {
- InMemoryStruct<macho::SegmentLoadCommand> SegmentLC;
- Obj->ReadSegmentLoadCommand(*SegmentLCI, SegmentLC);
- if (!SegmentLC)
- return Error("unable to load segment load command");
-
- for (unsigned SectNum = 0; SectNum != SegmentLC->NumSections; ++SectNum) {
- InMemoryStruct<macho::Section> Sect;
- Obj->ReadSection(*SegmentLCI, SectNum, Sect);
- if (!Sect)
- return Error("unable to load section: '" + Twine(SectNum) + "'");
-
- // FIXME: For the time being, we're only loading text segments.
- if (Sect->Flags != 0x80000400)
- continue;
+namespace llvm {
- // Address and names of symbols in the section.
- typedef std::pair<uint64_t, StringRef> SymbolEntry;
- SmallVector<SymbolEntry, 64> Symbols;
- // Index of all the names, in this section or not. Used when we're
- // dealing with relocation entries.
- SmallVector<StringRef, 64> SymbolNames;
- for (unsigned i = 0; i != SymtabLC->NumSymbolTableEntries; ++i) {
- InMemoryStruct<macho::SymbolTableEntry> STE;
- Obj->ReadSymbolTableEntry(SymtabLC->SymbolTableOffset, i, STE);
- if (!STE)
- return Error("unable to read symbol: '" + Twine(i) + "'");
- if (STE->SectionIndex > SegmentLC->NumSections)
- return Error("invalid section index for symbol: '" + Twine(i) + "'");
- // Get the symbol name.
- StringRef Name = Obj->getStringAtIndex(STE->StringIndex);
- SymbolNames.push_back(Name);
-
- // Just skip symbols not defined in this section.
- if ((unsigned)STE->SectionIndex - 1 != SectNum)
- continue;
-
- // FIXME: Check the symbol type and flags.
- if (STE->Type != 0xF) // external, defined in this section.
- continue;
- // Flags == 0x8 marks a thumb function for ARM, which is fine as it
- // doesn't require any special handling here.
- // Flags in the upper nibble we don't care about.
- if ((STE->Flags & 0xf) != 0x0 && STE->Flags != 0x8)
- continue;
-
- // Remember the symbol.
- Symbols.push_back(SymbolEntry(STE->Value, Name));
-
- DEBUG(dbgs() << "Function sym: '" << Name << "' @ " <<
- (Sect->Address + STE->Value) << "\n");
- }
- // Sort the symbols by address, just in case they didn't come in that way.
- array_pod_sort(Symbols.begin(), Symbols.end());
+int64_t RuntimeDyldMachO::memcpyAddend(const RelocationEntry &RE) const {
+ unsigned NumBytes = 1 << RE.Size;
+ uint8_t *Src = Sections[RE.SectionID].Address + RE.Offset;
- // If there weren't any functions (odd, but just in case...)
- if (!Symbols.size())
- continue;
+ return static_cast<int64_t>(readBytesUnaligned(Src, NumBytes));
+}
- // Extract the function data.
- uint8_t *Base = (uint8_t*)Obj->getData(SegmentLC->FileOffset,
- SegmentLC->FileSize).data();
- for (unsigned i = 0, e = Symbols.size() - 1; i != e; ++i) {
- uint64_t StartOffset = Sect->Address + Symbols[i].first;
- uint64_t EndOffset = Symbols[i + 1].first - 1;
- DEBUG(dbgs() << "Extracting function: " << Symbols[i].second
- << " from [" << StartOffset << ", " << EndOffset << "]\n");
- extractFunction(Symbols[i].second, Base + StartOffset, Base + EndOffset);
- }
- // The last symbol we do after since the end address is calculated
- // differently because there is no next symbol to reference.
- uint64_t StartOffset = Symbols[Symbols.size() - 1].first;
- uint64_t EndOffset = Sect->Size - 1;
- DEBUG(dbgs() << "Extracting function: " << Symbols[Symbols.size()-1].second
- << " from [" << StartOffset << ", " << EndOffset << "]\n");
- extractFunction(Symbols[Symbols.size()-1].second,
- Base + StartOffset, Base + EndOffset);
-
- // Now extract the relocation information for each function and process it.
- for (unsigned j = 0; j != Sect->NumRelocationTableEntries; ++j) {
- InMemoryStruct<macho::RelocationEntry> RE;
- Obj->ReadRelocationEntry(Sect->RelocationTableOffset, j, RE);
- if (RE->Word0 & macho::RF_Scattered)
- return Error("NOT YET IMPLEMENTED: scattered relocations.");
- // Word0 of the relocation is the offset into the section where the
- // relocation should be applied. We need to translate that into an
- // offset into a function since that's our atom.
- uint32_t Offset = RE->Word0;
- // Look for the function containing the address. This is used for JIT
- // code, so the number of functions in section is almost always going
- // to be very small (usually just one), so until we have use cases
- // where that's not true, just use a trivial linear search.
- unsigned SymbolNum;
- unsigned NumSymbols = Symbols.size();
- assert(NumSymbols > 0 && Symbols[0].first <= Offset &&
- "No symbol containing relocation!");
- for (SymbolNum = 0; SymbolNum < NumSymbols - 1; ++SymbolNum)
- if (Symbols[SymbolNum + 1].first > Offset)
- break;
- // Adjust the offset to be relative to the symbol.
- Offset -= Symbols[SymbolNum].first;
- // Get the name of the symbol containing the relocation.
- StringRef TargetName = SymbolNames[SymbolNum];
-
- bool isExtern = (RE->Word1 >> 27) & 1;
- // Figure out the source symbol of the relocation. If isExtern is true,
- // this relocation references the symbol table, otherwise it references
- // a section in the same object, numbered from 1 through NumSections
- // (SectionBases is [0, NumSections-1]).
- // FIXME: Some targets (ARM) use internal relocations even for
- // externally visible symbols, if the definition is in the same
- // file as the reference. We need to convert those back to by-name
- // references. We can resolve the address based on the section
- // offset and see if we have a symbol at that address. If we do,
- // use that; otherwise, puke.
- if (!isExtern)
- return Error("Internal relocations not supported.");
- uint32_t SourceNum = RE->Word1 & 0xffffff; // 24-bit value
- StringRef SourceName = SymbolNames[SourceNum];
-
- // FIXME: Get the relocation addend from the target address.
-
- // Now store the relocation information. Associate it with the source
- // symbol.
- Relocations[SourceName].push_back(RelocationEntry(TargetName,
- Offset,
- RE->Word1,
- 0 /*Addend*/));
- DEBUG(dbgs() << "Relocation at '" << TargetName << "' + " << Offset
- << " from '" << SourceName << "(Word1: "
- << format("0x%x", RE->Word1) << ")\n");
+RelocationValueRef RuntimeDyldMachO::getRelocationValueRef(
+ const ObjectFile &BaseTObj, const relocation_iterator &RI,
+ const RelocationEntry &RE, ObjSectionToIDMap &ObjSectionToID) {
+
+ const MachOObjectFile &Obj =
+ static_cast<const MachOObjectFile &>(BaseTObj);
+ MachO::any_relocation_info RelInfo =
+ Obj.getRelocation(RI->getRawDataRefImpl());
+ RelocationValueRef Value;
+
+ bool IsExternal = Obj.getPlainRelocationExternal(RelInfo);
+ if (IsExternal) {
+ symbol_iterator Symbol = RI->getSymbol();
+ StringRef TargetName;
+ Symbol->getName(TargetName);
+ SymbolTableMap::const_iterator SI =
+ GlobalSymbolTable.find(TargetName.data());
+ if (SI != GlobalSymbolTable.end()) {
+ Value.SectionID = SI->second.first;
+ Value.Offset = SI->second.second + RE.Addend;
+ } else {
+ Value.SymbolName = TargetName.data();
+ Value.Offset = RE.Addend;
}
+ } else {
+ SectionRef Sec = Obj.getRelocationSection(RelInfo);
+ bool IsCode = Sec.isText();
+ Value.SectionID = findOrEmitSection(Obj, Sec, IsCode, ObjSectionToID);
+ uint64_t Addr = Sec.getAddress();
+ Value.Offset = RE.Addend - Addr;
}
- return false;
-}
+ return Value;
+}
-bool RuntimeDyldMachO::
-loadSegment64(const MachOObject *Obj,
- const MachOObject::LoadCommandInfo *SegmentLCI,
- const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) {
- InMemoryStruct<macho::Segment64LoadCommand> Segment64LC;
- Obj->ReadSegment64LoadCommand(*SegmentLCI, Segment64LC);
- if (!Segment64LC)
- return Error("unable to load segment load command");
+void RuntimeDyldMachO::makeValueAddendPCRel(RelocationValueRef &Value,
+ const ObjectFile &BaseTObj,
+ const relocation_iterator &RI,
+ unsigned OffsetToNextPC) {
+ const MachOObjectFile &Obj =
+ static_cast<const MachOObjectFile &>(BaseTObj);
+ MachO::any_relocation_info RelInfo =
+ Obj.getRelocation(RI->getRawDataRefImpl());
+
+ bool IsPCRel = Obj.getAnyRelocationPCRel(RelInfo);
+ if (IsPCRel) {
+ uint64_t RelocAddr = 0;
+ RI->getAddress(RelocAddr);
+ Value.Offset += RelocAddr + OffsetToNextPC;
+ }
+}
- for (unsigned SectNum = 0; SectNum != Segment64LC->NumSections; ++SectNum) {
- InMemoryStruct<macho::Section64> Sect;
- Obj->ReadSection64(*SegmentLCI, SectNum, Sect);
- if (!Sect)
- return Error("unable to load section: '" + Twine(SectNum) + "'");
+void RuntimeDyldMachO::dumpRelocationToResolve(const RelocationEntry &RE,
+ uint64_t Value) const {
+ const SectionEntry &Section = Sections[RE.SectionID];
+ uint8_t *LocalAddress = Section.Address + RE.Offset;
+ uint64_t FinalAddress = Section.LoadAddress + RE.Offset;
+
+ dbgs() << "resolveRelocation Section: " << RE.SectionID
+ << " LocalAddress: " << format("%p", LocalAddress)
+ << " FinalAddress: " << format("0x%016" PRIx64, FinalAddress)
+ << " Value: " << format("0x%016" PRIx64, Value) << " Addend: " << RE.Addend
+ << " isPCRel: " << RE.IsPCRel << " MachoType: " << RE.RelType
+ << " Size: " << (1 << RE.Size) << "\n";
+}
- // FIXME: For the time being, we're only loading text segments.
- if (Sect->Flags != 0x80000400)
- continue;
+section_iterator
+RuntimeDyldMachO::getSectionByAddress(const MachOObjectFile &Obj,
+ uint64_t Addr) {
+ section_iterator SI = Obj.section_begin();
+ section_iterator SE = Obj.section_end();
+
+ for (; SI != SE; ++SI) {
+ uint64_t SAddr = SI->getAddress();
+ uint64_t SSize = SI->getSize();
+ if ((Addr >= SAddr) && (Addr < SAddr + SSize))
+ return SI;
+ }
- // Address and names of symbols in the section.
- typedef std::pair<uint64_t, StringRef> SymbolEntry;
- SmallVector<SymbolEntry, 64> Symbols;
- // Index of all the names, in this section or not. Used when we're
- // dealing with relocation entries.
- SmallVector<StringRef, 64> SymbolNames;
- for (unsigned i = 0; i != SymtabLC->NumSymbolTableEntries; ++i) {
- InMemoryStruct<macho::Symbol64TableEntry> STE;
- Obj->ReadSymbol64TableEntry(SymtabLC->SymbolTableOffset, i, STE);
- if (!STE)
- return Error("unable to read symbol: '" + Twine(i) + "'");
- if (STE->SectionIndex > Segment64LC->NumSections)
- return Error("invalid section index for symbol: '" + Twine(i) + "'");
- // Get the symbol name.
- StringRef Name = Obj->getStringAtIndex(STE->StringIndex);
- SymbolNames.push_back(Name);
-
- // Just skip symbols not defined in this section.
- if ((unsigned)STE->SectionIndex - 1 != SectNum)
- continue;
-
- // FIXME: Check the symbol type and flags.
- if (STE->Type != 0xF) // external, defined in this section.
- continue;
- // Flags in the upper nibble we don't care about.
- if ((STE->Flags & 0xf) != 0x0)
- continue;
-
- // Remember the symbol.
- Symbols.push_back(SymbolEntry(STE->Value, Name));
-
- DEBUG(dbgs() << "Function sym: '" << Name << "' @ " <<
- (Sect->Address + STE->Value) << "\n");
- }
- // Sort the symbols by address, just in case they didn't come in that way.
- array_pod_sort(Symbols.begin(), Symbols.end());
+ return SE;
+}
- // If there weren't any functions (odd, but just in case...)
- if (!Symbols.size())
- continue;
- // Extract the function data.
- uint8_t *Base = (uint8_t*)Obj->getData(Segment64LC->FileOffset,
- Segment64LC->FileSize).data();
- for (unsigned i = 0, e = Symbols.size() - 1; i != e; ++i) {
- uint64_t StartOffset = Sect->Address + Symbols[i].first;
- uint64_t EndOffset = Symbols[i + 1].first - 1;
- DEBUG(dbgs() << "Extracting function: " << Symbols[i].second
- << " from [" << StartOffset << ", " << EndOffset << "]\n");
- extractFunction(Symbols[i].second, Base + StartOffset, Base + EndOffset);
- }
- // The last symbol we do after since the end address is calculated
- // differently because there is no next symbol to reference.
- uint64_t StartOffset = Symbols[Symbols.size() - 1].first;
- uint64_t EndOffset = Sect->Size - 1;
- DEBUG(dbgs() << "Extracting function: " << Symbols[Symbols.size()-1].second
- << " from [" << StartOffset << ", " << EndOffset << "]\n");
- extractFunction(Symbols[Symbols.size()-1].second,
- Base + StartOffset, Base + EndOffset);
-
- // Now extract the relocation information for each function and process it.
- for (unsigned j = 0; j != Sect->NumRelocationTableEntries; ++j) {
- InMemoryStruct<macho::RelocationEntry> RE;
- Obj->ReadRelocationEntry(Sect->RelocationTableOffset, j, RE);
- if (RE->Word0 & macho::RF_Scattered)
- return Error("NOT YET IMPLEMENTED: scattered relocations.");
- // Word0 of the relocation is the offset into the section where the
- // relocation should be applied. We need to translate that into an
- // offset into a function since that's our atom.
- uint32_t Offset = RE->Word0;
- // Look for the function containing the address. This is used for JIT
- // code, so the number of functions in section is almost always going
- // to be very small (usually just one), so until we have use cases
- // where that's not true, just use a trivial linear search.
- unsigned SymbolNum;
- unsigned NumSymbols = Symbols.size();
- assert(NumSymbols > 0 && Symbols[0].first <= Offset &&
- "No symbol containing relocation!");
- for (SymbolNum = 0; SymbolNum < NumSymbols - 1; ++SymbolNum)
- if (Symbols[SymbolNum + 1].first > Offset)
- break;
- // Adjust the offset to be relative to the symbol.
- Offset -= Symbols[SymbolNum].first;
- // Get the name of the symbol containing the relocation.
- StringRef TargetName = SymbolNames[SymbolNum];
-
- bool isExtern = (RE->Word1 >> 27) & 1;
- // Figure out the source symbol of the relocation. If isExtern is true,
- // this relocation references the symbol table, otherwise it references
- // a section in the same object, numbered from 1 through NumSections
- // (SectionBases is [0, NumSections-1]).
- if (!isExtern)
- return Error("Internal relocations not supported.");
- uint32_t SourceNum = RE->Word1 & 0xffffff; // 24-bit value
- StringRef SourceName = SymbolNames[SourceNum];
-
- // FIXME: Get the relocation addend from the target address.
-
- // Now store the relocation information. Associate it with the source
- // symbol.
- Relocations[SourceName].push_back(RelocationEntry(TargetName,
- Offset,
- RE->Word1,
- 0 /*Addend*/));
- DEBUG(dbgs() << "Relocation at '" << TargetName << "' + " << Offset
- << " from '" << SourceName << "(Word1: "
- << format("0x%x", RE->Word1) << ")\n");
- }
+// Populate __pointers section.
+void RuntimeDyldMachO::populateIndirectSymbolPointersSection(
+ const MachOObjectFile &Obj,
+ const SectionRef &PTSection,
+ unsigned PTSectionID) {
+ assert(!Obj.is64Bit() &&
+ "Pointer table section not supported in 64-bit MachO.");
+
+ MachO::dysymtab_command DySymTabCmd = Obj.getDysymtabLoadCommand();
+ MachO::section Sec32 = Obj.getSection(PTSection.getRawDataRefImpl());
+ uint32_t PTSectionSize = Sec32.size;
+ unsigned FirstIndirectSymbol = Sec32.reserved1;
+ const unsigned PTEntrySize = 4;
+ unsigned NumPTEntries = PTSectionSize / PTEntrySize;
+ unsigned PTEntryOffset = 0;
+
+ assert((PTSectionSize % PTEntrySize) == 0 &&
+ "Pointers section does not contain a whole number of stubs?");
+
+ DEBUG(dbgs() << "Populating pointer table section "
+ << Sections[PTSectionID].Name
+ << ", Section ID " << PTSectionID << ", "
+ << NumPTEntries << " entries, " << PTEntrySize
+ << " bytes each:\n");
+
+ for (unsigned i = 0; i < NumPTEntries; ++i) {
+ unsigned SymbolIndex =
+ Obj.getIndirectSymbolTableEntry(DySymTabCmd, FirstIndirectSymbol + i);
+ symbol_iterator SI = Obj.getSymbolByIndex(SymbolIndex);
+ StringRef IndirectSymbolName;
+ SI->getName(IndirectSymbolName);
+ DEBUG(dbgs() << " " << IndirectSymbolName << ": index " << SymbolIndex
+ << ", PT offset: " << PTEntryOffset << "\n");
+ RelocationEntry RE(PTSectionID, PTEntryOffset,
+ MachO::GENERIC_RELOC_VANILLA, 0, false, 2);
+ addRelocationForSymbol(RE, IndirectSymbolName);
+ PTEntryOffset += PTEntrySize;
}
- return false;
}
-bool RuntimeDyldMachO::loadObject(MemoryBuffer *InputBuffer) {
- // If the linker is in an error state, don't do anything.
- if (hasError())
- return true;
- // Load the Mach-O wrapper object.
- std::string ErrorStr;
- OwningPtr<MachOObject> Obj(
- MachOObject::LoadFromBuffer(InputBuffer, &ErrorStr));
- if (!Obj)
- return Error("unable to load object: '" + ErrorStr + "'");
-
- // Get the CPU type information from the header.
- const macho::Header &Header = Obj->getHeader();
-
- // FIXME: Error checking that the loaded object is compatible with
- // the system we're running on.
- CPUType = Header.CPUType;
- CPUSubtype = Header.CPUSubtype;
-
- // Validate that the load commands match what we expect.
- const MachOObject::LoadCommandInfo *SegmentLCI = 0, *SymtabLCI = 0,
- *DysymtabLCI = 0;
- for (unsigned i = 0; i != Header.NumLoadCommands; ++i) {
- const MachOObject::LoadCommandInfo &LCI = Obj->getLoadCommandInfo(i);
- switch (LCI.Command.Type) {
- case macho::LCT_Segment:
- case macho::LCT_Segment64:
- if (SegmentLCI)
- return Error("unexpected input object (multiple segments)");
- SegmentLCI = &LCI;
- break;
- case macho::LCT_Symtab:
- if (SymtabLCI)
- return Error("unexpected input object (multiple symbol tables)");
- SymtabLCI = &LCI;
- break;
- case macho::LCT_Dysymtab:
- if (DysymtabLCI)
- return Error("unexpected input object (multiple symbol tables)");
- DysymtabLCI = &LCI;
- break;
- default:
- return Error("unexpected input object (unexpected load command");
- }
- }
+bool RuntimeDyldMachO::isCompatibleFile(const object::ObjectFile &Obj) const {
+ return Obj.isMachO();
+}
- if (!SymtabLCI)
- return Error("no symbol table found in object");
- if (!SegmentLCI)
- return Error("no symbol table found in object");
-
- // Read and register the symbol table data.
- InMemoryStruct<macho::SymtabLoadCommand> SymtabLC;
- Obj->ReadSymtabLoadCommand(*SymtabLCI, SymtabLC);
- if (!SymtabLC)
- return Error("unable to load symbol table load command");
- Obj->RegisterStringTable(*SymtabLC);
-
- // Read the dynamic link-edit information, if present (not present in static
- // objects).
- if (DysymtabLCI) {
- InMemoryStruct<macho::DysymtabLoadCommand> DysymtabLC;
- Obj->ReadDysymtabLoadCommand(*DysymtabLCI, DysymtabLC);
- if (!DysymtabLC)
- return Error("unable to load dynamic link-exit load command");
-
- // FIXME: We don't support anything interesting yet.
-// if (DysymtabLC->LocalSymbolsIndex != 0)
-// return Error("NOT YET IMPLEMENTED: local symbol entries");
-// if (DysymtabLC->ExternalSymbolsIndex != 0)
-// return Error("NOT YET IMPLEMENTED: non-external symbol entries");
-// if (DysymtabLC->UndefinedSymbolsIndex != SymtabLC->NumSymbolTableEntries)
-// return Error("NOT YET IMPLEMENTED: undefined symbol entries");
+template <typename Impl>
+void RuntimeDyldMachOCRTPBase<Impl>::finalizeLoad(const ObjectFile &ObjImg,
+ ObjSectionToIDMap &SectionMap) {
+ unsigned EHFrameSID = RTDYLD_INVALID_SECTION_ID;
+ unsigned TextSID = RTDYLD_INVALID_SECTION_ID;
+ unsigned ExceptTabSID = RTDYLD_INVALID_SECTION_ID;
+ ObjSectionToIDMap::iterator i, e;
+
+ for (i = SectionMap.begin(), e = SectionMap.end(); i != e; ++i) {
+ const SectionRef &Section = i->first;
+ StringRef Name;
+ Section.getName(Name);
+ if (Name == "__eh_frame")
+ EHFrameSID = i->second;
+ else if (Name == "__text")
+ TextSID = i->second;
+ else if (Name == "__gcc_except_tab")
+ ExceptTabSID = i->second;
+ else
+ impl().finalizeSection(ObjImg, i->second, Section);
}
+ UnregisteredEHFrameSections.push_back(
+ EHFrameRelatedSections(EHFrameSID, TextSID, ExceptTabSID));
+}
- // Load the segment load command.
- if (SegmentLCI->Command.Type == macho::LCT_Segment) {
- if (loadSegment32(Obj.get(), SegmentLCI, SymtabLC))
- return true;
- } else {
- if (loadSegment64(Obj.get(), SegmentLCI, SymtabLC))
- return true;
+template <typename Impl>
+unsigned char *RuntimeDyldMachOCRTPBase<Impl>::processFDE(unsigned char *P,
+ int64_t DeltaForText,
+ int64_t DeltaForEH) {
+ typedef typename Impl::TargetPtrT TargetPtrT;
+
+ DEBUG(dbgs() << "Processing FDE: Delta for text: " << DeltaForText
+ << ", Delta for EH: " << DeltaForEH << "\n");
+ uint32_t Length = readBytesUnaligned(P, 4);
+ P += 4;
+ unsigned char *Ret = P + Length;
+ uint32_t Offset = readBytesUnaligned(P, 4);
+ if (Offset == 0) // is a CIE
+ return Ret;
+
+ P += 4;
+ TargetPtrT FDELocation = readBytesUnaligned(P, sizeof(TargetPtrT));
+ TargetPtrT NewLocation = FDELocation - DeltaForText;
+ writeBytesUnaligned(NewLocation, P, sizeof(TargetPtrT));
+
+ P += sizeof(TargetPtrT);
+
+ // Skip the FDE address range
+ P += sizeof(TargetPtrT);
+
+ uint8_t Augmentationsize = *P;
+ P += 1;
+ if (Augmentationsize != 0) {
+ TargetPtrT LSDA = readBytesUnaligned(P, sizeof(TargetPtrT));
+ TargetPtrT NewLSDA = LSDA - DeltaForEH;
+ writeBytesUnaligned(NewLSDA, P, sizeof(TargetPtrT));
}
- return false;
+ return Ret;
}
-// Assign an address to a symbol name and resolve all the relocations
-// associated with it.
-void RuntimeDyldMachO::reassignSymbolAddress(StringRef Name, uint8_t *Addr) {
- // Assign the address in our symbol table.
- SymbolTable[Name] = Addr;
-
- RelocationList &Relocs = Relocations[Name];
- for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
- RelocationEntry &RE = Relocs[i];
- uint8_t *Target = SymbolTable[RE.Target] + RE.Offset;
- bool isPCRel = (RE.Data >> 24) & 1;
- unsigned Type = (RE.Data >> 28) & 0xf;
- unsigned Size = 1 << ((RE.Data >> 25) & 3);
-
- DEBUG(dbgs() << "Resolving relocation at '" << RE.Target
- << "' + " << RE.Offset << " (" << format("%p", Target) << ")"
- << " from '" << Name << " (" << format("%p", Addr) << ")"
- << "(" << (isPCRel ? "pcrel" : "absolute")
- << ", type: " << Type << ", Size: " << Size << ").\n");
-
- resolveRelocation(Target, Addr, isPCRel, Type, Size);
- RE.isResolved = true;
+static int64_t computeDelta(SectionEntry *A, SectionEntry *B) {
+ int64_t ObjDistance = A->ObjAddress - B->ObjAddress;
+ int64_t MemDistance = A->LoadAddress - B->LoadAddress;
+ return ObjDistance - MemDistance;
+}
+
+template <typename Impl>
+void RuntimeDyldMachOCRTPBase<Impl>::registerEHFrames() {
+
+ if (!MemMgr)
+ return;
+ for (int i = 0, e = UnregisteredEHFrameSections.size(); i != e; ++i) {
+ EHFrameRelatedSections &SectionInfo = UnregisteredEHFrameSections[i];
+ if (SectionInfo.EHFrameSID == RTDYLD_INVALID_SECTION_ID ||
+ SectionInfo.TextSID == RTDYLD_INVALID_SECTION_ID)
+ continue;
+ SectionEntry *Text = &Sections[SectionInfo.TextSID];
+ SectionEntry *EHFrame = &Sections[SectionInfo.EHFrameSID];
+ SectionEntry *ExceptTab = nullptr;
+ if (SectionInfo.ExceptTabSID != RTDYLD_INVALID_SECTION_ID)
+ ExceptTab = &Sections[SectionInfo.ExceptTabSID];
+
+ int64_t DeltaForText = computeDelta(Text, EHFrame);
+ int64_t DeltaForEH = 0;
+ if (ExceptTab)
+ DeltaForEH = computeDelta(ExceptTab, EHFrame);
+
+ unsigned char *P = EHFrame->Address;
+ unsigned char *End = P + EHFrame->Size;
+ do {
+ P = processFDE(P, DeltaForText, DeltaForEH);
+ } while (P != End);
+
+ MemMgr->registerEHFrames(EHFrame->Address, EHFrame->LoadAddress,
+ EHFrame->Size);
+ }
+ UnregisteredEHFrameSections.clear();
+}
+
+std::unique_ptr<RuntimeDyldMachO>
+RuntimeDyldMachO::create(Triple::ArchType Arch, RTDyldMemoryManager *MM) {
+ switch (Arch) {
+ default:
+ llvm_unreachable("Unsupported target for RuntimeDyldMachO.");
+ break;
+ case Triple::arm: return make_unique<RuntimeDyldMachOARM>(MM);
+ case Triple::aarch64: return make_unique<RuntimeDyldMachOAArch64>(MM);
+ case Triple::x86: return make_unique<RuntimeDyldMachOI386>(MM);
+ case Triple::x86_64: return make_unique<RuntimeDyldMachOX86_64>(MM);
}
}
-bool RuntimeDyldMachO::isKnownFormat(const MemoryBuffer *InputBuffer) {
- StringRef Magic = InputBuffer->getBuffer().slice(0, 4);
- if (Magic == "\xFE\xED\xFA\xCE") return true;
- if (Magic == "\xCE\xFA\xED\xFE") return true;
- if (Magic == "\xFE\xED\xFA\xCF") return true;
- if (Magic == "\xCF\xFA\xED\xFE") return true;
- return false;
+std::unique_ptr<RuntimeDyld::LoadedObjectInfo>
+RuntimeDyldMachO::loadObject(const object::ObjectFile &O) {
+ unsigned SectionStartIdx, SectionEndIdx;
+ std::tie(SectionStartIdx, SectionEndIdx) = loadObjectImpl(O);
+ return llvm::make_unique<LoadedMachOObjectInfo>(*this, SectionStartIdx,
+ SectionEndIdx);
}
} // end namespace llvm