-//===-- RuntimeDyld.h - Run-time dynamic linker for MC-JIT ------*- C++ -*-===//
+//===-- RuntimeDyld.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/SmallVector.h"
-#include "llvm/ADT/StringMap.h"
-#include "llvm/ADT/StringRef.h"
-#include "llvm/ADT/STLExtras.h"
-#include "llvm/ADT/Twine.h"
#include "llvm/ExecutionEngine/RuntimeDyld.h"
-#include "llvm/Object/MachOObject.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/Format.h"
-#include "llvm/Support/Memory.h"
-#include "llvm/Support/MemoryBuffer.h"
-#include "llvm/Support/system_error.h"
-#include "llvm/Support/raw_ostream.h"
+#include "ObjectImageCommon.h"
+#include "RuntimeDyldELF.h"
+#include "RuntimeDyldImpl.h"
+#include "RuntimeDyldMachO.h"
+#include "llvm/Support/MathExtras.h"
+#include "llvm/Support/Path.h"
+
using namespace llvm;
using namespace llvm::object;
// Empty out-of-line virtual destructor as the key function.
RTDyldMemoryManager::~RTDyldMemoryManager() {}
+RuntimeDyldImpl::~RuntimeDyldImpl() {}
namespace llvm {
-class RuntimeDyldImpl {
- unsigned CPUType;
- unsigned CPUSubtype;
-
- // The MemoryManager to load objects into.
- RTDyldMemoryManager *MemMgr;
-
- // FIXME: This all assumes we're dealing with external symbols for anything
- // explicitly referenced. I.e., we can index by name and things
- // will work out. In practice, this may not be the case, so we
- // should find a way to effectively generalize.
-
- // For each function, we have a MemoryBlock of it's instruction data.
- StringMap<sys::MemoryBlock> Functions;
-
- // Master symbol table. As modules are loaded and external symbols are
- // resolved, their addresses are stored here.
- StringMap<uint8_t*> SymbolTable;
-
- // For each symbol, keep a list of relocations based on it. Anytime
- // its address is reassigned (the JIT re-compiled the function, e.g.),
- // the relocations get re-resolved.
- struct RelocationEntry {
- std::string Target; // Object this relocation is contained in.
- uint64_t Offset; // Offset into the object for the relocation.
- uint32_t Data; // Second word of the raw macho relocation entry.
- int64_t Addend; // Addend encoded in the instruction itself, if any.
- bool isResolved; // Has this relocation been resolved previously?
-
- RelocationEntry(StringRef t, uint64_t offset, uint32_t data, int64_t addend)
- : Target(t), Offset(offset), Data(data), Addend(addend),
- isResolved(false) {}
- };
- typedef SmallVector<RelocationEntry, 4> RelocationList;
- StringMap<RelocationList> Relocations;
-
- // FIXME: Also keep a map of all the relocations contained in an object. Use
- // this to dynamically answer whether all of the relocations in it have
- // been resolved or not.
-
- bool HasError;
- std::string ErrorStr;
-
- // Set the error state and record an error string.
- bool Error(const Twine &Msg) {
- ErrorStr = Msg.str();
- HasError = true;
- return true;
- }
- void extractFunction(StringRef Name, uint8_t *StartAddress,
- uint8_t *EndAddress);
- bool resolveRelocation(uint8_t *Address, uint8_t *Value, bool isPCRel,
- unsigned Type, unsigned Size);
- bool resolveX86_64Relocation(uintptr_t Address, uintptr_t Value, bool isPCRel,
- unsigned Type, unsigned Size);
- bool resolveARMRelocation(uintptr_t Address, uintptr_t Value, bool isPCRel,
- unsigned Type, unsigned Size);
-
- bool loadSegment32(const MachOObject *Obj,
- const MachOObject::LoadCommandInfo *SegmentLCI,
- const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC);
- bool loadSegment64(const MachOObject *Obj,
- const MachOObject::LoadCommandInfo *SegmentLCI,
- const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC);
-
-public:
- RuntimeDyldImpl(RTDyldMemoryManager *mm) : MemMgr(mm), HasError(false) {}
-
- bool loadObject(MemoryBuffer *InputBuffer);
-
- void *getSymbolAddress(StringRef Name) {
- // FIXME: Just look up as a function for now. Overly simple of course.
- // Work in progress.
- return SymbolTable.lookup(Name);
+// Resolve the relocations for all symbols we currently know about.
+void RuntimeDyldImpl::resolveRelocations() {
+ // First, resolve relocations associated with external symbols.
+ resolveExternalSymbols();
+
+ // Just iterate over the sections we have and resolve all the relocations
+ // in them. Gross overkill, but it gets the job done.
+ for (int i = 0, e = Sections.size(); i != e; ++i) {
+ uint64_t Addr = Sections[i].LoadAddress;
+ DEBUG(dbgs() << "Resolving relocations Section #" << i
+ << "\t" << format("%p", (uint8_t *)Addr)
+ << "\n");
+ resolveRelocationList(Relocations[i], Addr);
}
-
- void resolveRelocations();
-
- void reassignSymbolAddress(StringRef Name, uint8_t *Addr);
-
- // Is the linker in an error state?
- bool hasError() { return HasError; }
-
- // Mark the error condition as handled and continue.
- void clearError() { HasError = false; }
-
- // Get the error message.
- StringRef getErrorString() { return ErrorStr; }
-};
-
-void RuntimeDyldImpl::extractFunction(StringRef Name, uint8_t *StartAddress,
- uint8_t *EndAddress) {
- // Allocate memory for the function via the memory manager.
- uintptr_t Size = EndAddress - StartAddress + 1;
- uint8_t *Mem = MemMgr->startFunctionBody(Name.data(), Size);
- assert(Size >= (uint64_t)(EndAddress - StartAddress + 1) &&
- "Memory manager failed to allocate enough memory!");
- // Copy the function payload into the memory block.
- memcpy(Mem, StartAddress, EndAddress - StartAddress + 1);
- MemMgr->endFunctionBody(Name.data(), Mem, Mem + Size);
- // Remember where we put it.
- Functions[Name] = sys::MemoryBlock(Mem, Size);
- // Default the assigned address for this symbol to wherever this
- // allocated it.
- SymbolTable[Name] = Mem;
- DEBUG(dbgs() << " allocated to " << Mem << "\n");
}
-bool RuntimeDyldImpl::
-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);
+void RuntimeDyldImpl::mapSectionAddress(const void *LocalAddress,
+ uint64_t TargetAddress) {
+ for (unsigned i = 0, e = Sections.size(); i != e; ++i) {
+ if (Sections[i].Address == LocalAddress) {
+ reassignSectionAddress(i, TargetAddress);
+ return;
+ }
}
- llvm_unreachable("");
+ llvm_unreachable("Attempting to remap address of unknown section!");
}
-bool RuntimeDyldImpl::
-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;
+// Subclasses can implement this method to create specialized image instances.
+// The caller owns the pointer that is returned.
+ObjectImage *RuntimeDyldImpl::createObjectImage(ObjectBuffer *InputBuffer) {
+ return new ObjectImageCommon(InputBuffer);
}
-bool RuntimeDyldImpl::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;
+ObjectImage *RuntimeDyldImpl::loadObject(ObjectBuffer *InputBuffer) {
+ OwningPtr<ObjectImage> obj(createObjectImage(InputBuffer));
+ if (!obj)
+ report_fatal_error("Unable to create object image from memory buffer!");
+
+ Arch = (Triple::ArchType)obj->getArch();
+
+ // Symbols found in this object
+ StringMap<SymbolLoc> LocalSymbols;
+ // Used sections from the object file
+ ObjSectionToIDMap LocalSections;
+
+ // Common symbols requiring allocation, with their sizes and alignments
+ CommonSymbolMap CommonSymbols;
+ // Maximum required total memory to allocate all common symbols
+ uint64_t CommonSize = 0;
+
+ error_code err;
+ // Parse symbols
+ DEBUG(dbgs() << "Parse symbols:\n");
+ for (symbol_iterator i = obj->begin_symbols(), e = obj->end_symbols();
+ i != e; i.increment(err)) {
+ Check(err);
+ object::SymbolRef::Type SymType;
+ StringRef Name;
+ Check(i->getType(SymType));
+ Check(i->getName(Name));
+
+ uint32_t flags;
+ Check(i->getFlags(flags));
+
+ bool isCommon = flags & SymbolRef::SF_Common;
+ if (isCommon) {
+ // Add the common symbols to a list. We'll allocate them all below.
+ uint64_t Align = getCommonSymbolAlignment(*i);
+ uint64_t Size = 0;
+ Check(i->getSize(Size));
+ CommonSize += Size + Align;
+ CommonSymbols[*i] = CommonSymbolInfo(Size, Align);
+ } else {
+ if (SymType == object::SymbolRef::ST_Function ||
+ SymType == object::SymbolRef::ST_Data ||
+ SymType == object::SymbolRef::ST_Unknown) {
+ uint64_t FileOffset;
+ StringRef SectionData;
+ section_iterator si = obj->end_sections();
+ Check(i->getFileOffset(FileOffset));
+ Check(i->getSection(si));
+ if (si == obj->end_sections()) continue;
+ Check(si->getContents(SectionData));
+ const uint8_t* SymPtr = (const uint8_t*)InputBuffer->getBufferStart() +
+ (uintptr_t)FileOffset;
+ uintptr_t SectOffset = (uintptr_t)(SymPtr -
+ (const uint8_t*)SectionData.begin());
+ unsigned SectionID =
+ findOrEmitSection(*obj,
+ *si,
+ SymType == object::SymbolRef::ST_Function,
+ LocalSections);
+ LocalSymbols[Name.data()] = SymbolLoc(SectionID, SectOffset);
+ DEBUG(dbgs() << "\tFileOffset: " << format("%p", (uintptr_t)FileOffset)
+ << " flags: " << flags
+ << " SID: " << SectionID
+ << " Offset: " << format("%p", SectOffset));
+ GlobalSymbolTable[Name] = SymbolLoc(SectionID, SectOffset);
+ }
+ }
+ DEBUG(dbgs() << "\tType: " << SymType << " Name: " << Name << "\n");
}
- 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;
+ // Allocate common symbols
+ if (CommonSize != 0)
+ emitCommonSymbols(*obj, CommonSymbols, CommonSize, LocalSymbols);
+
+ // Parse and process relocations
+ DEBUG(dbgs() << "Parse relocations:\n");
+ for (section_iterator si = obj->begin_sections(),
+ se = obj->end_sections(); si != se; si.increment(err)) {
+ Check(err);
+ bool isFirstRelocation = true;
+ unsigned SectionID = 0;
+ StubMap Stubs;
+
+ for (relocation_iterator i = si->begin_relocations(),
+ e = si->end_relocations(); i != e; i.increment(err)) {
+ Check(err);
+
+ // If it's the first relocation in this section, find its SectionID
+ if (isFirstRelocation) {
+ SectionID = findOrEmitSection(*obj, *si, true, LocalSections);
+ DEBUG(dbgs() << "\tSectionID: " << SectionID << "\n");
+ isFirstRelocation = false;
+ }
+
+ ObjRelocationInfo RI;
+ RI.SectionID = SectionID;
+ Check(i->getAdditionalInfo(RI.AdditionalInfo));
+ Check(i->getOffset(RI.Offset));
+ Check(i->getSymbol(RI.Symbol));
+ Check(i->getType(RI.Type));
+
+ DEBUG(dbgs() << "\t\tAddend: " << RI.AdditionalInfo
+ << " Offset: " << format("%p", (uintptr_t)RI.Offset)
+ << " Type: " << (uint32_t)(RI.Type & 0xffffffffL)
+ << "\n");
+ processRelocationRef(RI, *obj, LocalSections, LocalSymbols, Stubs);
}
- 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;
+
+ return obj.take();
}
-bool RuntimeDyldImpl::
-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: Improve check.
- if (Sect->Flags != 0x80000400)
- return Error("unsupported section type!");
-
- // 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.
- return Error("unexpected symbol type!");
- // Flags == 0x8 marks a thumb function for ARM, which is fine as it
- // doesn't require any special handling here.
- if (STE->Flags != 0x0 && STE->Flags != 0x8)
- return Error("unexpected symbol type!");
-
- // 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());
-
- // 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");
+void RuntimeDyldImpl::emitCommonSymbols(ObjectImage &Obj,
+ const CommonSymbolMap &CommonSymbols,
+ uint64_t TotalSize,
+ SymbolTableMap &SymbolTable) {
+ // Allocate memory for the section
+ unsigned SectionID = Sections.size();
+ uint8_t *Addr = MemMgr->allocateDataSection(TotalSize, sizeof(void*),
+ SectionID, false);
+ if (!Addr)
+ report_fatal_error("Unable to allocate memory for common symbols!");
+ uint64_t Offset = 0;
+ Sections.push_back(SectionEntry(StringRef(), Addr, TotalSize, TotalSize, 0));
+ memset(Addr, 0, TotalSize);
+
+ DEBUG(dbgs() << "emitCommonSection SectionID: " << SectionID
+ << " new addr: " << format("%p", Addr)
+ << " DataSize: " << TotalSize
+ << "\n");
+
+ // Assign the address of each symbol
+ for (CommonSymbolMap::const_iterator it = CommonSymbols.begin(),
+ itEnd = CommonSymbols.end(); it != itEnd; it++) {
+ uint64_t Size = it->second.first;
+ uint64_t Align = it->second.second;
+ StringRef Name;
+ it->first.getName(Name);
+ if (Align) {
+ // This symbol has an alignment requirement.
+ uint64_t AlignOffset = OffsetToAlignment((uint64_t)Addr, Align);
+ Addr += AlignOffset;
+ Offset += AlignOffset;
+ DEBUG(dbgs() << "Allocating common symbol " << Name << " address " <<
+ format("%p\n", Addr));
}
+ Obj.updateSymbolAddress(it->first, (uint64_t)Addr);
+ SymbolTable[Name.data()] = SymbolLoc(SectionID, Offset);
+ Offset += Size;
+ Addr += Size;
}
- return false;
}
-
-bool RuntimeDyldImpl::
-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");
-
- 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) + "'");
-
- // FIXME: Improve check.
- if (Sect->Flags != 0x80000400)
- return Error("unsupported section type!");
-
- // 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.
- return Error("unexpected symbol type!");
- if (STE->Flags != 0x0)
- return Error("unexpected symbol type!");
-
- // 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());
-
- // 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");
- }
+unsigned RuntimeDyldImpl::emitSection(ObjectImage &Obj,
+ const SectionRef &Section,
+ bool IsCode) {
+
+ unsigned StubBufSize = 0,
+ StubSize = getMaxStubSize();
+ error_code err;
+ if (StubSize > 0) {
+ for (relocation_iterator i = Section.begin_relocations(),
+ e = Section.end_relocations(); i != e; i.increment(err), Check(err))
+ StubBufSize += StubSize;
+ }
+ StringRef data;
+ uint64_t Alignment64;
+ Check(Section.getContents(data));
+ Check(Section.getAlignment(Alignment64));
+
+ unsigned Alignment = (unsigned)Alignment64 & 0xffffffffL;
+ bool IsRequired;
+ bool IsVirtual;
+ bool IsZeroInit;
+ bool IsReadOnly;
+ uint64_t DataSize;
+ StringRef Name;
+ Check(Section.isRequiredForExecution(IsRequired));
+ Check(Section.isVirtual(IsVirtual));
+ Check(Section.isZeroInit(IsZeroInit));
+ Check(Section.isReadOnlyData(IsReadOnly));
+ Check(Section.getSize(DataSize));
+ Check(Section.getName(Name));
+
+ unsigned Allocate;
+ unsigned SectionID = Sections.size();
+ uint8_t *Addr;
+ const char *pData = 0;
+
+ // Some sections, such as debug info, don't need to be loaded for execution.
+ // Leave those where they are.
+ if (IsRequired) {
+ Allocate = DataSize + StubBufSize;
+ Addr = IsCode
+ ? MemMgr->allocateCodeSection(Allocate, Alignment, SectionID)
+ : MemMgr->allocateDataSection(Allocate, Alignment, SectionID, IsReadOnly);
+ if (!Addr)
+ report_fatal_error("Unable to allocate section memory!");
+
+ // Virtual sections have no data in the object image, so leave pData = 0
+ if (!IsVirtual)
+ pData = data.data();
+
+ // Zero-initialize or copy the data from the image
+ if (IsZeroInit || IsVirtual)
+ memset(Addr, 0, DataSize);
+ else
+ memcpy(Addr, pData, DataSize);
+
+ DEBUG(dbgs() << "emitSection SectionID: " << SectionID
+ << " Name: " << Name
+ << " obj addr: " << format("%p", pData)
+ << " new addr: " << format("%p", Addr)
+ << " DataSize: " << DataSize
+ << " StubBufSize: " << StubBufSize
+ << " Allocate: " << Allocate
+ << "\n");
+ Obj.updateSectionAddress(Section, (uint64_t)Addr);
+ }
+ else {
+ // Even if we didn't load the section, we need to record an entry for it
+ // to handle later processing (and by 'handle' I mean don't do anything
+ // with these sections).
+ Allocate = 0;
+ Addr = 0;
+ DEBUG(dbgs() << "emitSection SectionID: " << SectionID
+ << " Name: " << Name
+ << " obj addr: " << format("%p", data.data())
+ << " new addr: 0"
+ << " DataSize: " << DataSize
+ << " StubBufSize: " << StubBufSize
+ << " Allocate: " << Allocate
+ << "\n");
}
- return false;
+
+ Sections.push_back(SectionEntry(Name, Addr, Allocate, DataSize,
+ (uintptr_t)pData));
+ return SectionID;
}
-bool RuntimeDyldImpl::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");
- }
+unsigned RuntimeDyldImpl::findOrEmitSection(ObjectImage &Obj,
+ const SectionRef &Section,
+ bool IsCode,
+ ObjSectionToIDMap &LocalSections) {
+
+ unsigned SectionID = 0;
+ ObjSectionToIDMap::iterator i = LocalSections.find(Section);
+ if (i != LocalSections.end())
+ SectionID = i->second;
+ else {
+ SectionID = emitSection(Obj, Section, IsCode);
+ LocalSections[Section] = SectionID;
}
+ return SectionID;
+}
- 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");
- }
+void RuntimeDyldImpl::addRelocationForSection(const RelocationEntry &RE,
+ unsigned SectionID) {
+ Relocations[SectionID].push_back(RE);
+}
- // Load the segment load command.
- if (SegmentLCI->Command.Type == macho::LCT_Segment) {
- if (loadSegment32(Obj.get(), SegmentLCI, SymtabLC))
- return true;
+void RuntimeDyldImpl::addRelocationForSymbol(const RelocationEntry &RE,
+ StringRef SymbolName) {
+ // Relocation by symbol. If the symbol is found in the global symbol table,
+ // create an appropriate section relocation. Otherwise, add it to
+ // ExternalSymbolRelocations.
+ SymbolTableMap::const_iterator Loc =
+ GlobalSymbolTable.find(SymbolName);
+ if (Loc == GlobalSymbolTable.end()) {
+ ExternalSymbolRelocations[SymbolName].push_back(RE);
} else {
- if (loadSegment64(Obj.get(), SegmentLCI, SymtabLC))
- return true;
+ // Copy the RE since we want to modify its addend.
+ RelocationEntry RECopy = RE;
+ RECopy.Addend += Loc->second.second;
+ Relocations[Loc->second.first].push_back(RECopy);
}
-
- return false;
}
-// Resolve the relocations for all symbols we currently know about.
-void RuntimeDyldImpl::resolveRelocations() {
- // Just iterate over the symbols in our symbol table and assign their
- // addresses.
- StringMap<uint8_t*>::iterator i = SymbolTable.begin();
- StringMap<uint8_t*>::iterator e = SymbolTable.end();
- for (;i != e; ++i)
- reassignSymbolAddress(i->getKey(), i->getValue());
+uint8_t *RuntimeDyldImpl::createStubFunction(uint8_t *Addr) {
+ if (Arch == Triple::arm) {
+ // TODO: There is only ARM far stub now. We should add the Thumb stub,
+ // and stubs for branches Thumb - ARM and ARM - Thumb.
+ uint32_t *StubAddr = (uint32_t*)Addr;
+ *StubAddr = 0xe51ff004; // ldr pc,<label>
+ return (uint8_t*)++StubAddr;
+ } else if (Arch == Triple::mipsel) {
+ uint32_t *StubAddr = (uint32_t*)Addr;
+ // 0: 3c190000 lui t9,%hi(addr).
+ // 4: 27390000 addiu t9,t9,%lo(addr).
+ // 8: 03200008 jr t9.
+ // c: 00000000 nop.
+ const unsigned LuiT9Instr = 0x3c190000, AdduiT9Instr = 0x27390000;
+ const unsigned JrT9Instr = 0x03200008, NopInstr = 0x0;
+
+ *StubAddr = LuiT9Instr;
+ StubAddr++;
+ *StubAddr = AdduiT9Instr;
+ StubAddr++;
+ *StubAddr = JrT9Instr;
+ StubAddr++;
+ *StubAddr = NopInstr;
+ return Addr;
+ } else if (Arch == Triple::ppc64) {
+ // PowerPC64 stub: the address points to a function descriptor
+ // instead of the function itself. Load the function address
+ // on r11 and sets it to control register. Also loads the function
+ // TOC in r2 and environment pointer to r11.
+ writeInt32BE(Addr, 0x3D800000); // lis r12, highest(addr)
+ writeInt32BE(Addr+4, 0x618C0000); // ori r12, higher(addr)
+ writeInt32BE(Addr+8, 0x798C07C6); // sldi r12, r12, 32
+ writeInt32BE(Addr+12, 0x658C0000); // oris r12, r12, h(addr)
+ writeInt32BE(Addr+16, 0x618C0000); // ori r12, r12, l(addr)
+ writeInt32BE(Addr+20, 0xF8410028); // std r2, 40(r1)
+ writeInt32BE(Addr+24, 0xE96C0000); // ld r11, 0(r12)
+ writeInt32BE(Addr+28, 0xE84C0008); // ld r2, 0(r12)
+ writeInt32BE(Addr+32, 0x7D6903A6); // mtctr r11
+ writeInt32BE(Addr+36, 0xE96C0010); // ld r11, 16(r2)
+ writeInt32BE(Addr+40, 0x4E800420); // bctr
+
+ return Addr;
+ }
+ return Addr;
}
// Assign an address to a symbol name and resolve all the relocations
// associated with it.
-void RuntimeDyldImpl::reassignSymbolAddress(StringRef Name, uint8_t *Addr) {
- // Assign the address in our symbol table.
- SymbolTable[Name] = Addr;
+void RuntimeDyldImpl::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. Relocations can't
+ // be applied until all the sections have been moved. The client must
+ // trigger this with a call to MCJIT::finalize() or
+ // RuntimeDyld::resolveRelocations().
+ //
+ // 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.
+ Sections[SectionID].LoadAddress = Addr;
+}
+
+void RuntimeDyldImpl::resolveRelocationEntry(const RelocationEntry &RE,
+ uint64_t Value) {
+ // Ignore relocations for sections that were not loaded
+ if (Sections[RE.SectionID].Address != 0) {
+ DEBUG(dbgs() << "\tSectionID: " << RE.SectionID
+ << " + " << RE.Offset << " ("
+ << format("%p", Sections[RE.SectionID].Address + RE.Offset) << ")"
+ << " RelType: " << RE.RelType
+ << " Addend: " << RE.Addend
+ << "\n");
+
+ resolveRelocation(Sections[RE.SectionID], RE.Offset,
+ Value, RE.RelType, RE.Addend);
+ }
+}
- RelocationList &Relocs = Relocations[Name];
+void RuntimeDyldImpl::resolveRelocationList(const RelocationList &Relocs,
+ uint64_t Value) {
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;
+ resolveRelocationEntry(Relocs[i], Value);
}
}
+void RuntimeDyldImpl::resolveExternalSymbols() {
+ StringMap<RelocationList>::iterator i = ExternalSymbolRelocations.begin(),
+ e = ExternalSymbolRelocations.end();
+ for (; i != e; i++) {
+ StringRef Name = i->first();
+ RelocationList &Relocs = i->second;
+ SymbolTableMap::const_iterator Loc = GlobalSymbolTable.find(Name);
+ if (Loc == GlobalSymbolTable.end()) {
+ // This is an external symbol, try to get it address from
+ // MemoryManager.
+ uint8_t *Addr = (uint8_t*) MemMgr->getPointerToNamedFunction(Name.data(),
+ true);
+ DEBUG(dbgs() << "Resolving relocations Name: " << Name
+ << "\t" << format("%p", Addr)
+ << "\n");
+ resolveRelocationList(Relocs, (uintptr_t)Addr);
+ } else {
+ report_fatal_error("Expected external symbol");
+ }
+ }
+}
+
+
//===----------------------------------------------------------------------===//
// RuntimeDyld class implementation
-RuntimeDyld::RuntimeDyld(RTDyldMemoryManager *MM) {
- Dyld = new RuntimeDyldImpl(MM);
+RuntimeDyld::RuntimeDyld(RTDyldMemoryManager *mm) {
+ // FIXME: There's a potential issue lurking here if a single instance of
+ // RuntimeDyld is used to load multiple objects. The current implementation
+ // associates a single memory manager with a RuntimeDyld instance. Even
+ // though the public class spawns a new 'impl' instance for each load,
+ // they share a single memory manager. This can become a problem when page
+ // permissions are applied.
+ Dyld = 0;
+ MM = mm;
}
RuntimeDyld::~RuntimeDyld() {
delete Dyld;
}
-bool RuntimeDyld::loadObject(MemoryBuffer *InputBuffer) {
+ObjectImage *RuntimeDyld::loadObject(ObjectBuffer *InputBuffer) {
+ if (!Dyld) {
+ sys::LLVMFileType type = sys::IdentifyFileType(
+ InputBuffer->getBufferStart(),
+ static_cast<unsigned>(InputBuffer->getBufferSize()));
+ switch (type) {
+ case sys::ELF_Relocatable_FileType:
+ case sys::ELF_Executable_FileType:
+ case sys::ELF_SharedObject_FileType:
+ case sys::ELF_Core_FileType:
+ Dyld = new RuntimeDyldELF(MM);
+ break;
+ case sys::Mach_O_Object_FileType:
+ case sys::Mach_O_Executable_FileType:
+ case sys::Mach_O_FixedVirtualMemorySharedLib_FileType:
+ case sys::Mach_O_Core_FileType:
+ case sys::Mach_O_PreloadExecutable_FileType:
+ case sys::Mach_O_DynamicallyLinkedSharedLib_FileType:
+ case sys::Mach_O_DynamicLinker_FileType:
+ case sys::Mach_O_Bundle_FileType:
+ case sys::Mach_O_DynamicallyLinkedSharedLibStub_FileType:
+ case sys::Mach_O_DSYMCompanion_FileType:
+ Dyld = new RuntimeDyldMachO(MM);
+ break;
+ case sys::Unknown_FileType:
+ case sys::Bitcode_FileType:
+ case sys::Archive_FileType:
+ case sys::COFF_FileType:
+ report_fatal_error("Incompatible object format!");
+ }
+ } else {
+ if (!Dyld->isCompatibleFormat(InputBuffer))
+ report_fatal_error("Incompatible object format!");
+ }
+
return Dyld->loadObject(InputBuffer);
}
return Dyld->getSymbolAddress(Name);
}
+uint64_t RuntimeDyld::getSymbolLoadAddress(StringRef Name) {
+ return Dyld->getSymbolLoadAddress(Name);
+}
+
void RuntimeDyld::resolveRelocations() {
Dyld->resolveRelocations();
}
-void RuntimeDyld::reassignSymbolAddress(StringRef Name, uint8_t *Addr) {
- Dyld->reassignSymbolAddress(Name, Addr);
+void RuntimeDyld::reassignSectionAddress(unsigned SectionID,
+ uint64_t Addr) {
+ Dyld->reassignSectionAddress(SectionID, Addr);
+}
+
+void RuntimeDyld::mapSectionAddress(const void *LocalAddress,
+ uint64_t TargetAddress) {
+ Dyld->mapSectionAddress(LocalAddress, TargetAddress);
}
StringRef RuntimeDyld::getErrorString() {
projects / oota-llvm.git / blobdiff