//
//===----------------------------------------------------------------------===//
-#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 "ObjectImageCommon.h"
+#include "JITRegistrar.h"
using namespace llvm;
using namespace llvm::object;
+#define DEBUG_TYPE "dyld"
+
namespace llvm {
-bool RuntimeDyldMachO::
-resolveRelocation(uint8_t *Address, uint64_t Value, bool isPCRel,
- unsigned Type, unsigned Size, int64_t Addend) {
+class MachOObjectImage : public ObjectImageCommon {
+private:
+ typedef SmallVector<uint64_t, 1> SectionAddrList;
+ SectionAddrList OldSectionAddrList;
+
+protected:
+ bool is64;
+ bool Registered;
+
+private:
+ void initOldAddress() {
+ MachOObjectFile *objf = static_cast<MachOObjectFile *>(ObjFile.get());
+ // Unfortunately we need to do this, since there's information encoded
+ // in the original addr of the section that we could not otherwise
+ // recover. The reason for this is that symbols do not actually store
+ // their file offset, but only their vmaddr. This means that in order
+ // to locate the symbol correctly in the object file, we need to know
+ // where the original start of the section was (including any padding,
+ // etc).
+ for (section_iterator i = objf->section_begin(), e = objf->section_end();
+ i != e; ++i) {
+ uint64_t Addr;
+ i->getAddress(Addr);
+ OldSectionAddrList[i->getRawDataRefImpl().d.a] = Addr;
+ }
+ }
+
+public:
+ MachOObjectImage(ObjectBuffer *Input, bool is64)
+ : ObjectImageCommon(Input),
+ OldSectionAddrList(ObjFile->section_end()->getRawDataRefImpl().d.a, 0),
+ is64(is64), Registered(false) {
+ initOldAddress();
+ }
+
+ MachOObjectImage(std::unique_ptr<object::ObjectFile> Input, bool is64)
+ : ObjectImageCommon(std::move(Input)),
+ OldSectionAddrList(ObjFile->section_end()->getRawDataRefImpl().d.a, 0),
+ is64(is64), Registered(false) {
+ initOldAddress();
+ }
+
+ virtual ~MachOObjectImage() {
+ if (Registered)
+ deregisterWithDebugger();
+ }
+
+ // Subclasses can override these methods to update the image with loaded
+ // addresses for sections and common symbols
+ virtual void updateSectionAddress(const SectionRef &Sec, uint64_t Addr) {
+ MachOObjectFile *objf = static_cast<MachOObjectFile *>(ObjFile.get());
+ char *data =
+ const_cast<char *>(objf->getSectionPointer(Sec.getRawDataRefImpl()));
+
+ uint64_t oldAddr = OldSectionAddrList[Sec.getRawDataRefImpl().d.a];
+
+ if (is64) {
+ ((MachO::section_64 *)data)->addr = Addr;
+ } else {
+ ((MachO::section *)data)->addr = Addr;
+ }
+
+ for (symbol_iterator i = objf->symbol_begin(), e = objf->symbol_end();
+ i != e; ++i) {
+ section_iterator symSec(objf->section_end());
+ (*i).getSection(symSec);
+ if (*symSec == Sec) {
+ uint64_t symAddr;
+ (*i).getAddress(symAddr);
+ updateSymbolAddress(*i, symAddr + Addr - oldAddr);
+ }
+ }
+ }
+
+ uint64_t getOldSectionAddr(const SectionRef &Sec) const {
+ return OldSectionAddrList[Sec.getRawDataRefImpl().d.a];
+ }
+
+ virtual void updateSymbolAddress(const SymbolRef &Sym, uint64_t Addr) {
+ char *data = const_cast<char *>(
+ reinterpret_cast<const char *>(Sym.getRawDataRefImpl().p));
+ if (is64)
+ ((MachO::nlist_64 *)data)->n_value = Addr;
+ else
+ ((MachO::nlist *)data)->n_value = Addr;
+ }
+
+ virtual void registerWithDebugger() {
+ JITRegistrar::getGDBRegistrar().registerObject(*Buffer);
+ Registered = true;
+ }
+
+ virtual void deregisterWithDebugger() {
+ JITRegistrar::getGDBRegistrar().deregisterObject(*Buffer);
+ }
+};
+
+static uint32_t readMachOMagic(const char *InputBuffer, unsigned BufferSize) {
+ if (BufferSize < 4)
+ return 0;
+ StringRef Magic(InputBuffer, 4);
+ if (Magic == "\xFE\xED\xFA\xCE" || Magic == "\xCE\xFA\xED\xFE")
+ return 0xFEEDFACE;
+ else if (Magic == "\xFE\xED\xFA\xCF" || Magic == "\xCF\xFA\xED\xFE")
+ return 0xFEEDFACF;
+ // else
+ return 0;
+}
+
+ObjectImage *RuntimeDyldMachO::createObjectImage(ObjectBuffer *Buffer) {
+ uint32_t magic = readMachOMagic(Buffer->getBufferStart(),
+ Buffer->getBufferSize());
+ bool is64 = (magic == MachO::MH_MAGIC_64);
+ assert((magic == MachO::MH_MAGIC_64 || magic == MachO::MH_MAGIC) &&
+ "Unrecognized Macho Magic");
+ return new MachOObjectImage(Buffer, is64);
+}
+
+ObjectImage *RuntimeDyldMachO::createObjectImageFromFile(
+ std::unique_ptr<object::ObjectFile> ObjFile) {
+ if (!ObjFile)
+ return nullptr;
+
+ MemoryBuffer *Buffer =
+ MemoryBuffer::getMemBuffer(ObjFile->getData(), "", false);
+
+ uint32_t magic = readMachOMagic(Buffer->getBufferStart(),
+ Buffer->getBufferSize());
+ bool is64 = (magic == MachO::MH_MAGIC_64);
+ assert((magic == MachO::MH_MAGIC_64 || magic == MachO::MH_MAGIC) &&
+ "Unrecognized Macho Magic");
+ return new MachOObjectImage(std::move(ObjFile), is64);
+}
+
+static unsigned char *processFDE(unsigned char *P, intptr_t DeltaForText,
+ intptr_t DeltaForEH) {
+ DEBUG(dbgs() << "Processing FDE: Delta for text: " << DeltaForText
+ << ", Delta for EH: " << DeltaForEH << "\n");
+ uint32_t Length = *((uint32_t *)P);
+ P += 4;
+ unsigned char *Ret = P + Length;
+ uint32_t Offset = *((uint32_t *)P);
+ if (Offset == 0) // is a CIE
+ return Ret;
+
+ P += 4;
+ intptr_t FDELocation = *((intptr_t *)P);
+ intptr_t NewLocation = FDELocation - DeltaForText;
+ *((intptr_t *)P) = NewLocation;
+ P += sizeof(intptr_t);
+
+ // Skip the FDE address range
+ P += sizeof(intptr_t);
+
+ uint8_t Augmentationsize = *P;
+ P += 1;
+ if (Augmentationsize != 0) {
+ intptr_t LSDA = *((intptr_t *)P);
+ intptr_t NewLSDA = LSDA - DeltaForEH;
+ *((intptr_t *)P) = NewLSDA;
+ }
+
+ return Ret;
+}
+
+static intptr_t computeDelta(SectionEntry *A, SectionEntry *B) {
+ intptr_t ObjDistance = A->ObjAddress - B->ObjAddress;
+ intptr_t MemDistance = A->LoadAddress - B->LoadAddress;
+ return ObjDistance - MemDistance;
+}
+
+void RuntimeDyldMachO::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];
+
+ intptr_t DeltaForText = computeDelta(Text, EHFrame);
+ intptr_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();
+}
+
+void RuntimeDyldMachO::finalizeLoad(ObjectImage &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 if (Name == "__jump_table")
+ populateJumpTable(cast<MachOObjectFile>(*ObjImg.getObjectFile()),
+ Section, i->second);
+ else if (Name == "__pointers")
+ populatePointersSection(cast<MachOObjectFile>(*ObjImg.getObjectFile()),
+ Section, i->second);
+ }
+ UnregisteredEHFrameSections.push_back(
+ EHFrameRelatedSections(EHFrameSID, TextSID, ExceptTabSID));
+}
+
+// The target location for the relocation is described by RE.SectionID and
+// RE.Offset. RE.SectionID can be used to find the SectionEntry. Each
+// SectionEntry has three members describing its location.
+// SectionEntry::Address is the address at which the section has been loaded
+// into memory in the current (host) process. SectionEntry::LoadAddress is the
+// address that the section will have in the target process.
+// SectionEntry::ObjAddress is the address of the bits for this section in the
+// original emitted object image (also in the current address space).
+//
+// Relocations will be applied as if the section were loaded at
+// SectionEntry::LoadAddress, but they will be applied at an address based
+// on SectionEntry::Address. SectionEntry::ObjAddress will be used to refer to
+// Target memory contents if they are required for value calculations.
+//
+// The Value parameter here is the load address of the symbol for the
+// relocation to be applied. For relocations which refer to symbols in the
+// current object Value will be the LoadAddress of the section in which
+// the symbol resides (RE.Addend provides additional information about the
+// symbol location). For external symbols, Value will be the address of the
+// symbol in the target address space.
+void RuntimeDyldMachO::resolveRelocation(const RelocationEntry &RE,
+ uint64_t Value) {
+ DEBUG (
+ 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("%p", FinalAddress)
+ << " Value: " << format("%p", Value)
+ << " Addend: " << RE.Addend
+ << " isPCRel: " << RE.IsPCRel
+ << " MachoType: " << RE.RelType
+ << " Size: " << (1 << RE.Size) << "\n";
+ );
+
// 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, Addend);
- case mach::CTM_ARM:
- return resolveARMRelocation((uintptr_t)Address, (uintptr_t)Value,
- isPCRel, Type, Size, Addend);
- }
- llvm_unreachable("");
+ switch (Arch) {
+ default:
+ llvm_unreachable("Unsupported CPU type!");
+ case Triple::x86_64:
+ resolveX86_64Relocation(RE, Value);
+ break;
+ case Triple::x86:
+ resolveI386Relocation(RE, Value);
+ break;
+ case Triple::arm: // Fall through.
+ case Triple::thumb:
+ resolveARMRelocation(RE, Value);
+ break;
+ case Triple::aarch64:
+ case Triple::arm64:
+ resolveAArch64Relocation(RE, Value);
+ break;
+ }
}
-bool RuntimeDyldMachO::
-resolveX86_64Relocation(uintptr_t Address, uintptr_t Value, bool isPCRel,
- unsigned Type, unsigned Size, int64_t Addend) {
+bool RuntimeDyldMachO::resolveI386Relocation(const RelocationEntry &RE,
+ uint64_t Value) {
+ const SectionEntry &Section = Sections[RE.SectionID];
+ uint8_t* LocalAddress = Section.Address + RE.Offset;
+
+ if (RE.IsPCRel) {
+ uint64_t FinalAddress = Section.LoadAddress + RE.Offset;
+ Value -= FinalAddress + 4; // see MachOX86_64::resolveRelocation.
+ }
+
+ switch (RE.RelType) {
+ default:
+ llvm_unreachable("Invalid relocation type!");
+ case MachO::GENERIC_RELOC_VANILLA:
+ return applyRelocationValue(LocalAddress, Value + RE.Addend,
+ 1 << RE.Size);
+ case MachO::GENERIC_RELOC_SECTDIFF:
+ case MachO::GENERIC_RELOC_LOCAL_SECTDIFF: {
+ uint64_t SectionABase = Sections[RE.Sections.SectionA].LoadAddress;
+ uint64_t SectionBBase = Sections[RE.Sections.SectionB].LoadAddress;
+ assert((Value == SectionABase || Value == SectionBBase) &&
+ "Unexpected SECTDIFF relocation value.");
+ Value = SectionABase - SectionBBase + RE.Addend;
+ return applyRelocationValue(LocalAddress, Value, 1 << RE.Size);
+ }
+ case MachO::GENERIC_RELOC_PB_LA_PTR:
+ return Error("Relocation type not implemented yet!");
+ }
+}
+
+bool RuntimeDyldMachO::resolveX86_64Relocation(const RelocationEntry &RE,
+ uint64_t Value) {
+ const SectionEntry &Section = Sections[RE.SectionID];
+ uint8_t* LocalAddress = Section.Address + RE.Offset;
+
// If the relocation is PC-relative, the value to be encoded is the
// pointer difference.
- if (isPCRel)
+ if (RE.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;
+ uint64_t FinalAddress = Section.LoadAddress + RE.Offset;
+ Value -= FinalAddress + 4; // see MachOX86_64::resolveRelocation.
+ }
- switch(Type) {
+ switch (RE.RelType) {
default:
llvm_unreachable("Invalid relocation type!");
- case macho::RIT_X86_64_Signed1:
- case macho::RIT_X86_64_Signed2:
- case macho::RIT_X86_64_Signed4:
- case macho::RIT_X86_64_Signed:
- case macho::RIT_X86_64_Unsigned:
- case macho::RIT_X86_64_Branch: {
- Value += Addend;
- // 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_GOTLoad:
- case macho::RIT_X86_64_GOT:
- case macho::RIT_X86_64_Subtractor:
- case macho::RIT_X86_64_TLV:
+ case MachO::X86_64_RELOC_SIGNED_1:
+ case MachO::X86_64_RELOC_SIGNED_2:
+ case MachO::X86_64_RELOC_SIGNED_4:
+ case MachO::X86_64_RELOC_SIGNED:
+ case MachO::X86_64_RELOC_UNSIGNED:
+ case MachO::X86_64_RELOC_BRANCH:
+ return applyRelocationValue(LocalAddress, Value + RE.Addend, 1 << RE.Size);
+ case MachO::X86_64_RELOC_GOT_LOAD:
+ case MachO::X86_64_RELOC_GOT:
+ case MachO::X86_64_RELOC_SUBTRACTOR:
+ case MachO::X86_64_RELOC_TLV:
return Error("Relocation type not implemented yet!");
}
}
-bool RuntimeDyldMachO::
-resolveARMRelocation(uintptr_t Address, uintptr_t Value, bool isPCRel,
- unsigned Type, unsigned Size, int64_t Addend) {
+bool RuntimeDyldMachO::resolveARMRelocation(const RelocationEntry &RE,
+ uint64_t Value) {
+ const SectionEntry &Section = Sections[RE.SectionID];
+ uint8_t* LocalAddress = Section.Address + RE.Offset;
+
// If the relocation is PC-relative, the value to be encoded is the
// pointer difference.
- if (isPCRel) {
- Value -= Address;
+ if (RE.IsPCRel) {
+ uint64_t FinalAddress = Section.LoadAddress + RE.Offset;
+ Value -= FinalAddress;
// 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;
}
- switch(Type) {
+ switch (RE.RelType) {
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: {
+ case MachO::ARM_RELOC_VANILLA:
+ return applyRelocationValue(LocalAddress, Value, 1 << RE.Size);
+ case MachO::ARM_RELOC_BR24: {
// 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;
+ uint32_t *p = (uint32_t *)LocalAddress;
// The low two bits of the value are not encoded.
Value >>= 2;
// Mask the value to 24 bits.
- Value &= 0xffffff;
+ uint64_t FinalValue = Value & 0xffffff;
+ // Check for overflow.
+ if (Value != FinalValue)
+ return Error("ARM BR24 relocation out of range.");
// 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;
+ *p = (*p & ~0xffffff) | FinalValue;
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:
+ case MachO::ARM_THUMB_RELOC_BR22:
+ case MachO::ARM_THUMB_32BIT_BRANCH:
+ case MachO::ARM_RELOC_HALF:
+ case MachO::ARM_RELOC_HALF_SECTDIFF:
+ case MachO::ARM_RELOC_PAIR:
+ case MachO::ARM_RELOC_SECTDIFF:
+ case MachO::ARM_RELOC_LOCAL_SECTDIFF:
+ case MachO::ARM_RELOC_PB_LA_PTR:
return Error("Relocation type not implemented yet!");
}
return false;
}
-bool RuntimeDyldMachO::
-loadSegment32(const MachOObject *Obj,
- const MachOObject::LoadCommandInfo *SegmentLCI,
- const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) {
- // FIXME: This should really be combined w/ loadSegment64. Templatized
- // function on the 32/64 datatypes maybe?
- InMemoryStruct<macho::SegmentLoadCommand> SegmentLC;
- Obj->ReadSegmentLoadCommand(*SegmentLCI, SegmentLC);
- if (!SegmentLC)
- return Error("unable to load segment load command");
-
-
- SmallVector<unsigned, 16> SectionMap;
- 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) + "'");
-
- // Allocate memory via the MM for the section.
- uint8_t *Buffer;
- uint32_t SectionID = Sections.size();
- if (Sect->Flags != 0x80000400)
- Buffer = MemMgr->allocateCodeSection(Sect->Size, Sect->Align, SectionID);
- else
- Buffer = MemMgr->allocateDataSection(Sect->Size, Sect->Align, SectionID);
-
- DEBUG(dbgs() << "Loading "
- << ((Sect->Flags == 0x80000400) ? "text" : "data")
- << " (ID #" << SectionID << ")"
- << " '" << Sect->SegmentName << ","
- << Sect->Name << "' of size " << Sect->Size
- << " to address " << Buffer << ".\n");
-
- // Copy the payload from the object file into the allocated buffer.
- uint8_t *Base = (uint8_t*)Obj->getData(SegmentLC->FileOffset,
- SegmentLC->FileSize).data();
- memcpy(Buffer, Base + Sect->Address, Sect->Size);
-
- // Remember what got allocated for this SectionID.
- Sections.push_back(sys::MemoryBlock(Buffer, Sect->Size));
- SectionLocalMemToID[Buffer] = SectionID;
-
- // By default, the load address of a section is its memory buffer.
- SectionLoadAddress.push_back((uint64_t)Buffer);
-
- // Keep a map of object file section numbers to corresponding SectionIDs
- // while processing the file.
- SectionMap.push_back(SectionID);
- }
-
- // Process the symbol table.
- SmallVector<StringRef, 64> SymbolNames;
- processSymbols32(Obj, SectionMap, SymbolNames, SymtabLC);
-
- // Process the relocations for each section we're loading.
- Relocations.grow(Relocations.size() + SegmentLC->NumSections);
- 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) + "'");
- 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;
- bool isExtern = (RE->Word1 >> 27) & 1;
-
- // FIXME: Get the relocation addend from the target address.
- // FIXME: VERY imporant for internal relocations.
-
- // 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]).
- uint32_t SourceNum = RE->Word1 & 0xffffff; // 24-bit value
- if (!isExtern) {
- assert(SourceNum > 0 && "Invalid relocation section number!");
- unsigned SectionID = SectionMap[SourceNum - 1];
- unsigned TargetID = SectionMap[SectNum];
- DEBUG(dbgs() << "Internal relocation at Section #"
- << TargetID << " + " << Offset
- << " from Section #"
- << SectionID << " (Word1: "
- << format("0x%x", RE->Word1) << ")\n");
-
- // Store the relocation information. It will get resolved when
- // the section addresses are assigned.
- Relocations[SectionID].push_back(RelocationEntry(TargetID,
- Offset,
- RE->Word1,
- 0 /*Addend*/));
- } else {
- StringRef SourceName = SymbolNames[SourceNum];
-
- // Now store the relocation information. Associate it with the source
- // symbol. Just add it to the unresolved list and let the general
- // path post-load resolve it if we know where the symbol is.
- UnresolvedRelocations[SourceName].push_back(RelocationEntry(SectNum,
- Offset,
- RE->Word1,
- 0 /*Addend*/));
- DEBUG(dbgs() << "Relocation at Section #" << SectNum << " + " << Offset
- << " from '" << SourceName << "(Word1: "
- << format("0x%x", RE->Word1) << ")\n");
- }
- }
- }
-
- // Resolve the addresses of any symbols that were defined in this segment.
- for (int i = 0, e = SymbolNames.size(); i != e; ++i)
- resolveSymbol(SymbolNames[i]);
-
- return false;
-}
+bool RuntimeDyldMachO::resolveAArch64Relocation(const RelocationEntry &RE,
+ uint64_t Value) {
+ const SectionEntry &Section = Sections[RE.SectionID];
+ uint8_t* LocalAddress = Section.Address + RE.Offset;
+ switch (RE.RelType) {
+ default:
+ llvm_unreachable("Invalid relocation type!");
+ case MachO::ARM64_RELOC_UNSIGNED: {
+ assert(!RE.IsPCRel && "PCRel and ARM64_RELOC_UNSIGNED not supported");
+ // 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).
+ if (RE.Size < 2)
+ llvm_unreachable("Invalid size for ARM64_RELOC_UNSIGNED");
-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");
-
-
- SmallVector<unsigned, 16> SectionMap;
- 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) + "'");
-
- // Allocate memory via the MM for the section.
- uint8_t *Buffer;
- uint32_t SectionID = Sections.size();
- if (Sect->Flags == 0x80000400)
- Buffer = MemMgr->allocateCodeSection(Sect->Size, Sect->Align, SectionID);
- else
- Buffer = MemMgr->allocateDataSection(Sect->Size, Sect->Align, SectionID);
-
- DEBUG(dbgs() << "Loading "
- << ((Sect->Flags == 0x80000400) ? "text" : "data")
- << " (ID #" << SectionID << ")"
- << " '" << Sect->SegmentName << ","
- << Sect->Name << "' of size " << Sect->Size
- << " to address " << Buffer << ".\n");
-
- // Copy the payload from the object file into the allocated buffer.
- uint8_t *Base = (uint8_t*)Obj->getData(Segment64LC->FileOffset,
- Segment64LC->FileSize).data();
- memcpy(Buffer, Base + Sect->Address, Sect->Size);
-
- // Remember what got allocated for this SectionID.
- Sections.push_back(sys::MemoryBlock(Buffer, Sect->Size));
- SectionLocalMemToID[Buffer] = SectionID;
-
- // By default, the load address of a section is its memory buffer.
- SectionLoadAddress.push_back((uint64_t)Buffer);
-
- // Keep a map of object file section numbers to corresponding SectionIDs
- // while processing the file.
- SectionMap.push_back(SectionID);
- }
-
- // Process the symbol table.
- SmallVector<StringRef, 64> SymbolNames;
- processSymbols64(Obj, SectionMap, SymbolNames, SymtabLC);
-
- // Process the relocations for each section we're loading.
- Relocations.grow(Relocations.size() + Segment64LC->NumSections);
- 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) + "'");
- 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;
- bool isExtern = (RE->Word1 >> 27) & 1;
-
- // FIXME: Get the relocation addend from the target address.
- // FIXME: VERY imporant for internal relocations.
-
- // 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]).
- uint32_t SourceNum = RE->Word1 & 0xffffff; // 24-bit value
- if (!isExtern) {
- assert(SourceNum > 0 && "Invalid relocation section number!");
- unsigned SectionID = SectionMap[SourceNum - 1];
- unsigned TargetID = SectionMap[SectNum];
- DEBUG(dbgs() << "Internal relocation at Section #"
- << TargetID << " + " << Offset
- << " from Section #"
- << SectionID << " (Word1: "
- << format("0x%x", RE->Word1) << ")\n");
-
- // Store the relocation information. It will get resolved when
- // the section addresses are assigned.
- Relocations[SectionID].push_back(RelocationEntry(TargetID,
- Offset,
- RE->Word1,
- 0 /*Addend*/));
- } else {
- StringRef SourceName = SymbolNames[SourceNum];
-
- // Now store the relocation information. Associate it with the source
- // symbol. Just add it to the unresolved list and let the general
- // path post-load resolve it if we know where the symbol is.
- UnresolvedRelocations[SourceName].push_back(RelocationEntry(SectNum,
- Offset,
- RE->Word1,
- 0 /*Addend*/));
- DEBUG(dbgs() << "Relocation at Section #" << SectNum << " + " << Offset
- << " from '" << SourceName << "(Word1: "
- << format("0x%x", RE->Word1) << ")\n");
+ applyRelocationValue(LocalAddress, Value + RE.Addend, 1 << RE.Size);
+ break;
+ }
+ case MachO::ARM64_RELOC_BRANCH26: {
+ assert(RE.IsPCRel && "not PCRel and ARM64_RELOC_BRANCH26 not supported");
+ // 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*)LocalAddress;
+ // Check if the addend is encoded in the instruction.
+ uint32_t EncodedAddend = *p & 0x03FFFFFF;
+ if (EncodedAddend != 0 ) {
+ if (RE.Addend == 0)
+ llvm_unreachable("branch26 instruction has embedded addend.");
+ else
+ llvm_unreachable("branch26 instruction has embedded addend and" \
+ "ARM64_RELOC_ADDEND.");
+ }
+ // Check if branch is in range.
+ uint64_t FinalAddress = Section.LoadAddress + RE.Offset;
+ uint64_t PCRelVal = Value - FinalAddress + RE.Addend;
+ assert(isInt<26>(PCRelVal) && "Branch target out of range!");
+ // Insert the value into the instruction.
+ *p = (*p & 0xFC000000) | ((uint32_t)(PCRelVal >> 2) & 0x03FFFFFF);
+ break;
+ }
+ case MachO::ARM64_RELOC_GOT_LOAD_PAGE21:
+ case MachO::ARM64_RELOC_PAGE21: {
+ assert(RE.IsPCRel && "not PCRel and ARM64_RELOC_PAGE21 not supported");
+ // 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*)LocalAddress;
+ // Check if the addend is encoded in the instruction.
+ uint32_t EncodedAddend = ((*p & 0x60000000) >> 29) |
+ ((*p & 0x01FFFFE0) >> 3);
+ if (EncodedAddend != 0) {
+ if (RE.Addend == 0)
+ llvm_unreachable("adrp instruction has embedded addend.");
+ else
+ llvm_unreachable("adrp instruction has embedded addend and" \
+ "ARM64_RELOC_ADDEND.");
+ }
+ // Adjust for PC-relative relocation and offset.
+ uint64_t FinalAddress = Section.LoadAddress + RE.Offset;
+ uint64_t PCRelVal = ((Value + RE.Addend) & (-4096)) -
+ (FinalAddress & (-4096));
+ // Check that the value fits into 21 bits (+ 12 lower bits).
+ assert(isInt<33>(PCRelVal) && "Invalid page reloc value!");
+ // Insert the value into the instruction.
+ uint32_t ImmLoValue = (uint32_t)(PCRelVal << 17) & 0x60000000;
+ uint32_t ImmHiValue = (uint32_t)(PCRelVal >> 9) & 0x00FFFFE0;
+ *p = (*p & 0x9F00001F) | ImmHiValue | ImmLoValue;
+ break;
+ }
+ case MachO::ARM64_RELOC_GOT_LOAD_PAGEOFF12:
+ case MachO::ARM64_RELOC_PAGEOFF12: {
+ assert(!RE.IsPCRel && "PCRel and ARM64_RELOC_PAGEOFF21 not supported");
+ // 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*)LocalAddress;
+ // Check if the addend is encoded in the instruction.
+ uint32_t EncodedAddend = *p & 0x003FFC00;
+ if (EncodedAddend != 0) {
+ if (RE.Addend == 0)
+ llvm_unreachable("adrp instruction has embedded addend.");
+ else
+ llvm_unreachable("adrp instruction has embedded addend and" \
+ "ARM64_RELOC_ADDEND.");
+ }
+ // Add the offset from the symbol.
+ Value += RE.Addend;
+ // Mask out the page address and only use the lower 12 bits.
+ Value &= 0xFFF;
+ // Check which instruction we are updating to obtain the implicit shift
+ // factor from LDR/STR instructions.
+ if (*p & 0x08000000) {
+ uint32_t ImplicitShift = ((*p >> 30) & 0x3);
+ switch (ImplicitShift) {
+ case 0:
+ // Check if this a vector op.
+ if ((*p & 0x04800000) == 0x04800000) {
+ ImplicitShift = 4;
+ assert(((Value & 0xF) == 0) &&
+ "128-bit LDR/STR not 16-byte aligned.");
+ }
+ break;
+ case 1:
+ assert(((Value & 0x1) == 0) && "16-bit LDR/STR not 2-byte aligned.");
+ case 2:
+ assert(((Value & 0x3) == 0) && "32-bit LDR/STR not 4-byte aligned.");
+ case 3:
+ assert(((Value & 0x7) == 0) && "64-bit LDR/STR not 8-byte aligned.");
}
+ // Compensate for implicit shift.
+ Value >>= ImplicitShift;
}
+ // Insert the value into the instruction.
+ *p = (*p & 0xFFC003FF) | ((uint32_t)(Value << 10) & 0x003FFC00);
+ break;
+ }
+ case MachO::ARM64_RELOC_SUBTRACTOR:
+ case MachO::ARM64_RELOC_POINTER_TO_GOT:
+ case MachO::ARM64_RELOC_TLVP_LOAD_PAGE21:
+ case MachO::ARM64_RELOC_TLVP_LOAD_PAGEOFF12:
+ llvm_unreachable("Relocation type not implemented yet!");
+ return Error("Relocation type not implemented yet!");
+ case MachO::ARM64_RELOC_ADDEND:
+ llvm_unreachable("ARM64_RELOC_ADDEND should have been handeled by " \
+ "processRelocationRef!");
}
-
- // Resolve the addresses of any symbols that were defined in this segment.
- for (int i = 0, e = SymbolNames.size(); i != e; ++i)
- resolveSymbol(SymbolNames[i]);
-
return false;
}
-bool RuntimeDyldMachO::
-processSymbols32(const MachOObject *Obj,
- SmallVectorImpl<unsigned> &SectionMap,
- SmallVectorImpl<StringRef> &SymbolNames,
- const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) {
- // FIXME: Combine w/ processSymbols64. Factor 64/32 datatype and such.
- 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) + "'");
- // Get the symbol name.
- StringRef Name = Obj->getStringAtIndex(STE->StringIndex);
- SymbolNames.push_back(Name);
-
- // FIXME: Check the symbol type and flags.
- if (STE->Type != 0xF) // external, defined in this segment.
- continue;
- // Flags in the upper nibble we don't care about.
- if ((STE->Flags & 0xf) != 0x0)
- continue;
-
- // Remember the symbol.
- uint32_t SectionID = SectionMap[STE->SectionIndex - 1];
- SymbolTable[Name] = SymbolLoc(SectionID, STE->Value);
+void RuntimeDyldMachO::populateJumpTable(MachOObjectFile &Obj,
+ const SectionRef &JTSection,
+ unsigned JTSectionID) {
+ assert(!Obj.is64Bit() &&
+ "__jump_table section not supported in 64-bit MachO.");
+
+ MachO::dysymtab_command DySymTabCmd = Obj.getDysymtabLoadCommand();
+ MachO::section Sec32 = Obj.getSection(JTSection.getRawDataRefImpl());
+ uint32_t JTSectionSize = Sec32.size;
+ unsigned FirstIndirectSymbol = Sec32.reserved1;
+ unsigned JTEntrySize = Sec32.reserved2;
+ unsigned NumJTEntries = JTSectionSize / JTEntrySize;
+ uint8_t* JTSectionAddr = getSectionAddress(JTSectionID);
+ unsigned JTEntryOffset = 0;
+
+ assert((JTSectionSize % JTEntrySize) == 0 &&
+ "Jump-table section does not contain a whole number of stubs?");
+
+ for (unsigned i = 0; i < NumJTEntries; ++i) {
+ unsigned SymbolIndex =
+ Obj.getIndirectSymbolTableEntry(DySymTabCmd, FirstIndirectSymbol + i);
+ symbol_iterator SI = Obj.getSymbolByIndex(SymbolIndex);
+ StringRef IndirectSymbolName;
+ SI->getName(IndirectSymbolName);
+ uint8_t* JTEntryAddr = JTSectionAddr + JTEntryOffset;
+ createStubFunction(JTEntryAddr);
+ RelocationEntry RE(JTSectionID, JTEntryOffset + 1,
+ MachO::GENERIC_RELOC_VANILLA, 0, true, 2);
+ addRelocationForSymbol(RE, IndirectSymbolName);
+ JTEntryOffset += JTEntrySize;
+ }
+}
- DEBUG(dbgs() << "Symbol: '" << Name << "' @ "
- << (getSectionAddress(SectionID) + STE->Value)
- << "\n");
+void RuntimeDyldMachO::populatePointersSection(MachOObjectFile &Obj,
+ const SectionRef &PTSection,
+ unsigned PTSectionID) {
+ assert(!Obj.is64Bit() &&
+ "__pointers 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 __pointers, 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::
-processSymbols64(const MachOObject *Obj,
- SmallVectorImpl<unsigned> &SectionMap,
- SmallVectorImpl<StringRef> &SymbolNames,
- const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) {
- 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) + "'");
- // Get the symbol name.
- StringRef Name = Obj->getStringAtIndex(STE->StringIndex);
- SymbolNames.push_back(Name);
-
- // FIXME: Check the symbol type and flags.
- if (STE->Type != 0xF) // external, defined in this segment.
- continue;
- // Flags in the upper nibble we don't care about.
- if ((STE->Flags & 0xf) != 0x0)
- continue;
- // Remember the symbol.
- uint32_t SectionID = SectionMap[STE->SectionIndex - 1];
- SymbolTable[Name] = SymbolLoc(SectionID, STE->Value);
+section_iterator getSectionByAddress(const MachOObjectFile &Obj,
+ uint64_t Addr) {
+ section_iterator SI = Obj.section_begin();
+ section_iterator SE = Obj.section_end();
- DEBUG(dbgs() << "Symbol: '" << Name << "' @ "
- << (getSectionAddress(SectionID) + STE->Value)
- << "\n");
+ for (; SI != SE; ++SI) {
+ uint64_t SAddr, SSize;
+ SI->getAddress(SAddr);
+ SI->getSize(SSize);
+ if ((Addr >= SAddr) && (Addr < SAddr + SSize))
+ return SI;
}
- return false;
+
+ return SE;
}
-// resolveSymbol - Resolve any relocations to the specified symbol if
-// we know where it lives.
-void RuntimeDyldMachO::resolveSymbol(StringRef Name) {
- StringMap<SymbolLoc>::const_iterator Loc = SymbolTable.find(Name);
- if (Loc == SymbolTable.end())
- return;
+relocation_iterator RuntimeDyldMachO::processSECTDIFFRelocation(
+ unsigned SectionID,
+ relocation_iterator RelI,
+ ObjectImage &Obj,
+ ObjSectionToIDMap &ObjSectionToID) {
+ const MachOObjectFile *MachO =
+ static_cast<const MachOObjectFile*>(Obj.getObjectFile());
+ MachO::any_relocation_info RE =
+ MachO->getRelocation(RelI->getRawDataRefImpl());
+
+ SectionEntry &Section = Sections[SectionID];
+ uint32_t RelocType = MachO->getAnyRelocationType(RE);
+ bool IsPCRel = MachO->getAnyRelocationPCRel(RE);
+ unsigned Size = MachO->getAnyRelocationLength(RE);
+ uint64_t Offset;
+ RelI->getOffset(Offset);
+ uint8_t *LocalAddress = Section.Address + Offset;
+ unsigned NumBytes = 1 << Size;
+ int64_t Addend = 0;
+ memcpy(&Addend, LocalAddress, NumBytes);
+
+ ++RelI;
+ MachO::any_relocation_info RE2 =
+ MachO->getRelocation(RelI->getRawDataRefImpl());
+
+ uint32_t AddrA = MachO->getScatteredRelocationValue(RE);
+ section_iterator SAI = getSectionByAddress(*MachO, AddrA);
+ assert(SAI != MachO->section_end() && "Can't find section for address A");
+ uint64_t SectionABase;
+ SAI->getAddress(SectionABase);
+ uint64_t SectionAOffset = AddrA - SectionABase;
+ SectionRef SectionA = *SAI;
+ bool IsCode;
+ SectionA.isText(IsCode);
+ uint32_t SectionAID = findOrEmitSection(Obj, SectionA, IsCode,
+ ObjSectionToID);
+
+ uint32_t AddrB = MachO->getScatteredRelocationValue(RE2);
+ section_iterator SBI = getSectionByAddress(*MachO, AddrB);
+ assert(SBI != MachO->section_end() && "Can't find section for address B");
+ uint64_t SectionBBase;
+ SBI->getAddress(SectionBBase);
+ uint64_t SectionBOffset = AddrB - SectionBBase;
+ SectionRef SectionB = *SBI;
+ uint32_t SectionBID = findOrEmitSection(Obj, SectionB, IsCode,
+ ObjSectionToID);
+
+ if (Addend != AddrA - AddrB)
+ Error("Unexpected SECTDIFF relocation addend.");
+
+ DEBUG(dbgs() << "Found SECTDIFF: AddrA: " << AddrA << ", AddrB: " << AddrB
+ << ", Addend: " << Addend << ", SectionA ID: "
+ << SectionAID << ", SectionAOffset: " << SectionAOffset
+ << ", SectionB ID: " << SectionBID << ", SectionBOffset: "
+ << SectionBOffset << "\n");
+ RelocationEntry R(SectionID, Offset, RelocType, 0,
+ SectionAID, SectionAOffset, SectionBID, SectionBOffset,
+ IsPCRel, Size);
+
+ addRelocationForSection(R, SectionAID);
+ addRelocationForSection(R, SectionBID);
+
+ return ++RelI;
+}
- RelocationList &Relocs = UnresolvedRelocations[Name];
- DEBUG(dbgs() << "Resolving symbol '" << Name << "'\n");
- for (int i = 0, e = Relocs.size(); i != e; ++i) {
- // Change the relocation to be section relative rather than symbol
- // relative and move it to the resolved relocation list.
- RelocationEntry Entry = Relocs[i];
- Entry.Addend += Loc->second.second;
- Relocations[Loc->second.first].push_back(Entry);
- }
- // FIXME: Keep a worklist of the relocations we've added so that we can
- // resolve more selectively later.
- Relocs.clear();
+relocation_iterator RuntimeDyldMachO::processI386ScatteredVANILLA(
+ unsigned SectionID,
+ relocation_iterator RelI,
+ ObjectImage &Obj,
+ ObjSectionToIDMap &ObjSectionToID) {
+ const MachOObjectFile *MachO =
+ static_cast<const MachOObjectFile*>(Obj.getObjectFile());
+ MachO::any_relocation_info RE =
+ MachO->getRelocation(RelI->getRawDataRefImpl());
+
+ SectionEntry &Section = Sections[SectionID];
+ uint32_t RelocType = MachO->getAnyRelocationType(RE);
+ bool IsPCRel = MachO->getAnyRelocationPCRel(RE);
+ unsigned Size = MachO->getAnyRelocationLength(RE);
+ uint64_t Offset;
+ RelI->getOffset(Offset);
+ uint8_t *LocalAddress = Section.Address + Offset;
+ unsigned NumBytes = 1 << Size;
+ int64_t Addend = 0;
+ memcpy(&Addend, LocalAddress, NumBytes);
+
+ unsigned SymbolBaseAddr = MachO->getScatteredRelocationValue(RE);
+ section_iterator TargetSI = getSectionByAddress(*MachO, SymbolBaseAddr);
+ assert(TargetSI != MachO->section_end() && "Can't find section for symbol");
+ uint64_t SectionBaseAddr;
+ TargetSI->getAddress(SectionBaseAddr);
+ SectionRef TargetSection = *TargetSI;
+ bool IsCode;
+ TargetSection.isText(IsCode);
+ uint32_t TargetSectionID = findOrEmitSection(Obj, TargetSection, IsCode,
+ ObjSectionToID);
+
+ Addend -= SectionBaseAddr;
+ RelocationEntry R(SectionID, Offset, RelocType, Addend,
+ IsPCRel, Size);
+
+ addRelocationForSection(R, TargetSectionID);
+
+ return ++RelI;
}
-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");
+relocation_iterator RuntimeDyldMachO::processRelocationRef(
+ unsigned SectionID, relocation_iterator RelI, ObjectImage &Obj,
+ ObjSectionToIDMap &ObjSectionToID, const SymbolTableMap &Symbols,
+ StubMap &Stubs) {
+ const ObjectFile *OF = Obj.getObjectFile();
+ const MachOObjectImage &MachOObj = *static_cast<MachOObjectImage *>(&Obj);
+ const MachOObjectFile *MachO = static_cast<const MachOObjectFile *>(OF);
+ MachO::any_relocation_info RE =
+ MachO->getRelocation(RelI->getRawDataRefImpl());
+ int64_t RelocAddendValue = 0;
+ bool HasRelocAddendValue = false;
+
+ uint32_t RelType = MachO->getAnyRelocationType(RE);
+ if (Arch == Triple::arm64) {
+ // ARM64_RELOC_ADDEND provides the offset (addend) that will be used by the
+ // next relocation entry. Save the value and advance to the next relocation
+ // entry.
+ if (RelType == MachO::ARM64_RELOC_ADDEND) {
+ assert(!MachO->getPlainRelocationExternal(RE));
+ assert(!MachO->getAnyRelocationPCRel(RE));
+ assert(MachO->getAnyRelocationLength(RE) == 2);
+ uint64_t RawAddend = MachO->getPlainRelocationSymbolNum(RE);
+ // Sign-extend the 24-bit to 64-bit.
+ RelocAddendValue = RawAddend << 40;
+ RelocAddendValue >>= 40;
+ HasRelocAddendValue = true;
+
+ // Get the next entry.
+ RE = MachO->getRelocation((++RelI)->getRawDataRefImpl());
+ RelType = MachO->getAnyRelocationType(RE);
+ assert(RelType == MachO::ARM64_RELOC_BRANCH26 ||
+ RelType == MachO::ARM64_RELOC_PAGE21 ||
+ RelType == MachO::ARM64_RELOC_PAGEOFF12);
+
+ } else if (RelType == MachO::ARM64_RELOC_BRANCH26 ||
+ RelType == MachO::ARM64_RELOC_PAGE21 ||
+ RelType == MachO::ARM64_RELOC_PAGEOFF12 ||
+ RelType == MachO::ARM64_RELOC_GOT_LOAD_PAGE21 ||
+ RelType == MachO::ARM64_RELOC_GOT_LOAD_PAGEOFF12) {
+ RelocAddendValue = 0;
+ HasRelocAddendValue = true;
}
}
- if (!SymtabLCI)
- return Error("no symbol table found in object");
- if (!SegmentLCI)
- return Error("no segments 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");
- }
-
- // Load the segment load command.
- if (SegmentLCI->Command.Type == macho::LCT_Segment) {
- if (loadSegment32(Obj.get(), SegmentLCI, SymtabLC))
- return true;
+ // FIXME: Properly handle scattered relocations.
+ // Special case the couple of scattered relocations that we know how
+ // to handle: SECTDIFF relocations, and scattered VANILLA relocations
+ // on I386.
+ // For all other scattered relocations, just bail out and hope for the
+ // best, since the offsets computed by scattered relocations have often
+ // been optimisticaly filled in by the compiler. This will fail
+ // horribly where the relocations *do* need to be applied, but that was
+ // already the case.
+ if (MachO->isRelocationScattered(RE)) {
+ if (RelType == MachO::GENERIC_RELOC_SECTDIFF ||
+ RelType == MachO::GENERIC_RELOC_LOCAL_SECTDIFF)
+ return processSECTDIFFRelocation(SectionID, RelI, Obj, ObjSectionToID);
+ else if (Arch == Triple::x86 && RelType == MachO::GENERIC_RELOC_VANILLA)
+ return processI386ScatteredVANILLA(SectionID, RelI, Obj, ObjSectionToID);
+ else
+ return ++RelI;
+ }
+
+ RelocationValueRef Value;
+ SectionEntry &Section = Sections[SectionID];
+
+ bool IsExtern = MachO->getPlainRelocationExternal(RE);
+ bool IsPCRel = MachO->getAnyRelocationPCRel(RE);
+ unsigned Size = MachO->getAnyRelocationLength(RE);
+ uint64_t Offset;
+ RelI->getOffset(Offset);
+ uint8_t *LocalAddress = Section.Address + Offset;
+ unsigned NumBytes = 1 << Size;
+ int64_t Addend = 0;
+ if (HasRelocAddendValue)
+ Addend = RelocAddendValue;
+ else
+ memcpy(&Addend, LocalAddress, NumBytes);
+
+ if (IsExtern) {
+ // Obtain the symbol name which is referenced in the relocation
+ symbol_iterator Symbol = RelI->getSymbol();
+ StringRef TargetName;
+ Symbol->getName(TargetName);
+ // First search for the symbol in the local symbol table
+ SymbolTableMap::const_iterator lsi = Symbols.find(TargetName.data());
+ if (lsi != Symbols.end()) {
+ Value.SectionID = lsi->second.first;
+ Value.Addend = lsi->second.second + Addend;
+ } else {
+ // Search for the symbol in the global symbol table
+ SymbolTableMap::const_iterator gsi =
+ GlobalSymbolTable.find(TargetName.data());
+ if (gsi != GlobalSymbolTable.end()) {
+ Value.SectionID = gsi->second.first;
+ Value.Addend = gsi->second.second + Addend;
+ } else {
+ Value.SymbolName = TargetName.data();
+ Value.Addend = Addend;
+ }
+ }
+
+ // Addends for external, PC-rel relocations on i386 point back to the zero
+ // offset. Calculate the final offset from the relocation target instead.
+ // This allows us to use the same logic for both external and internal
+ // relocations in resolveI386RelocationRef.
+ if (Arch == Triple::x86 && IsPCRel) {
+ uint64_t RelocAddr = 0;
+ RelI->getAddress(RelocAddr);
+ Value.Addend += RelocAddr + 4;
+ }
+
} else {
- if (loadSegment64(Obj.get(), SegmentLCI, SymtabLC))
- return true;
+ SectionRef Sec = MachO->getRelocationSection(RE);
+ bool IsCode = false;
+ Sec.isText(IsCode);
+ Value.SectionID = findOrEmitSection(Obj, Sec, IsCode, ObjSectionToID);
+ uint64_t Addr = MachOObj.getOldSectionAddr(Sec);
+ DEBUG(dbgs() << "\nAddr: " << Addr << "\nAddend: " << Addend);
+ Value.Addend = Addend - Addr;
+ if (IsPCRel)
+ Value.Addend += Offset + NumBytes;
}
- // Assign the addresses of the sections from the object so that any
- // relocations to them get set properly.
- // FIXME: This is done directly from the client at the moment. We should
- // default the values to the local storage, at least when the target arch
- // is the same as the host arch.
+ if (Arch == Triple::x86_64 && (RelType == MachO::X86_64_RELOC_GOT ||
+ RelType == MachO::X86_64_RELOC_GOT_LOAD)) {
+ assert(IsPCRel);
+ assert(Size == 2);
+
+ // FIXME: Teach the generic code above not to prematurely conflate
+ // relocation addends and symbol offsets.
+ Value.Addend -= Addend;
+ StubMap::const_iterator i = Stubs.find(Value);
+ uint8_t *Addr;
+ if (i != Stubs.end()) {
+ Addr = Section.Address + i->second;
+ } else {
+ Stubs[Value] = Section.StubOffset;
+ uint8_t *GOTEntry = Section.Address + Section.StubOffset;
+ RelocationEntry GOTRE(SectionID, Section.StubOffset,
+ MachO::X86_64_RELOC_UNSIGNED, Value.Addend, false,
+ 3);
+ if (Value.SymbolName)
+ addRelocationForSymbol(GOTRE, Value.SymbolName);
+ else
+ addRelocationForSection(GOTRE, Value.SectionID);
+ Section.StubOffset += 8;
+ Addr = GOTEntry;
+ }
+ RelocationEntry TargetRE(SectionID, Offset,
+ MachO::X86_64_RELOC_UNSIGNED, Addend, true,
+ 2);
+ resolveRelocation(TargetRE, (uint64_t)Addr);
+ } else if (Arch == Triple::arm && (RelType & 0xf) == MachO::ARM_RELOC_BR24) {
+ // This is an ARM branch relocation, need to use a stub function.
+
+ // Look up for existing stub.
+ StubMap::const_iterator i = Stubs.find(Value);
+ uint8_t *Addr;
+ if (i != Stubs.end()) {
+ Addr = Section.Address + i->second;
+ } else {
+ // Create a new stub function.
+ Stubs[Value] = Section.StubOffset;
+ uint8_t *StubTargetAddr =
+ createStubFunction(Section.Address + Section.StubOffset);
+ RelocationEntry StubRE(SectionID, StubTargetAddr - Section.Address,
+ MachO::GENERIC_RELOC_VANILLA, Value.Addend);
+ if (Value.SymbolName)
+ addRelocationForSymbol(StubRE, Value.SymbolName);
+ else
+ addRelocationForSection(StubRE, Value.SectionID);
+ Addr = Section.Address + Section.StubOffset;
+ Section.StubOffset += getMaxStubSize();
+ }
+ RelocationEntry TargetRE(Value.SectionID, Offset, RelType, 0, IsPCRel,
+ Size);
+ resolveRelocation(TargetRE, (uint64_t)Addr);
+ } else if (Arch == Triple::arm64 &&
+ (RelType == MachO::ARM64_RELOC_GOT_LOAD_PAGE21 ||
+ RelType == MachO::ARM64_RELOC_GOT_LOAD_PAGEOFF12)) {
+ assert(Size == 2);
+ StubMap::const_iterator i = Stubs.find(Value);
+ uint8_t *Addr;
+ if (i != Stubs.end())
+ Addr = Section.Address + i->second;
+ else {
+ // FIXME: There must be a better way to do this then to check and fix the
+ // alignment every time!!!
+ uintptr_t BaseAddress = uintptr_t(Section.Address);
+ uintptr_t StubAlignment = getStubAlignment();
+ uintptr_t StubAddress
+ = (BaseAddress + Section.StubOffset + StubAlignment - 1) &
+ -StubAlignment;
+ unsigned StubOffset = StubAddress - BaseAddress;
+ Stubs[Value] = StubOffset;
+ assert(((StubAddress % getStubAlignment()) == 0) &&
+ "GOT entry not aligned");
+ RelocationEntry GOTRE(SectionID, StubOffset, MachO::ARM64_RELOC_UNSIGNED,
+ Value.Addend, /*IsPCRel=*/false, /*Size=*/3);
+ if (Value.SymbolName)
+ addRelocationForSymbol(GOTRE, Value.SymbolName);
+ else
+ addRelocationForSection(GOTRE, Value.SectionID);
+ Section.StubOffset = StubOffset + getMaxStubSize();
+
+ Addr = (uint8_t *)StubAddress;
+ }
+ RelocationEntry TargetRE(SectionID, Offset, RelType, /*Addend=*/0, IsPCRel,
+ Size);
+ resolveRelocation(TargetRE, (uint64_t)Addr);
+ } else {
- return false;
+ RelocationEntry RE(SectionID, Offset, RelType, Value.Addend, IsPCRel, Size);
+ if (Value.SymbolName)
+ addRelocationForSymbol(RE, Value.SymbolName);
+ else
+ addRelocationForSection(RE, Value.SectionID);
+ }
+ return ++RelI;
}
-// Assign an address to a symbol name and resolve all the relocations
-// associated with it.
-void RuntimeDyldMachO::reassignSectionAddress(unsigned SectionID,
- uint64_t Addr) {
- // The address to use for relocation resolution is not
- // the address of the local section buffer. We must be doing
- // a remote execution environment of some sort. Re-apply any
- // relocations referencing this section with the given address.
- //
- // Addr is a uint64_t because we can't assume the pointer width
- // of the target is the same as that of the host. Just use a generic
- // "big enough" type.
-
- SectionLoadAddress[SectionID] = Addr;
-
- RelocationList &Relocs = Relocations[SectionID];
- for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
- RelocationEntry &RE = Relocs[i];
- uint8_t *Target = (uint8_t*)Sections[RE.SectionID].base() + 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 Section #" << RE.SectionID
- << " + " << RE.Offset << " (" << format("%p", Target) << ")"
- << " from Section #" << SectionID << " (" << format("%p", Addr) << ")"
- << "(" << (isPCRel ? "pcrel" : "absolute")
- << ", type: " << Type << ", Size: " << Size << ", Addend: "
- << RE.Addend << ").\n");
-
- resolveRelocation(Target, Addr, isPCRel, Type, Size, RE.Addend);
- }
+bool
+RuntimeDyldMachO::isCompatibleFormat(const ObjectBuffer *InputBuffer) const {
+ uint32_t Magic = readMachOMagic(InputBuffer->getBufferStart(),
+ InputBuffer->getBufferSize());
+ return (Magic == 0xFEEDFACE || Magic == 0xFEEDFACF);
}
-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;
+bool RuntimeDyldMachO::isCompatibleFile(const object::ObjectFile *Obj) const {
+ return Obj->isMachO();
}
} // end namespace llvm