diff options
Diffstat (limited to 'contrib/llvm/lib/ExecutionEngine/RuntimeDyld')
9 files changed, 2546 insertions, 0 deletions
diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/GDBRegistrar.cpp b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/GDBRegistrar.cpp new file mode 100644 index 0000000..603c526 --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/GDBRegistrar.cpp @@ -0,0 +1,214 @@ +//===-- GDBRegistrar.cpp - Registers objects with GDB ---------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +#include "JITRegistrar.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/Support/Compiler.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/Mutex.h" +#include "llvm/Support/MutexGuard.h" + +using namespace llvm; + +// This must be kept in sync with gdb/gdb/jit.h . +extern "C" { + + typedef enum { + JIT_NOACTION = 0, + JIT_REGISTER_FN, + JIT_UNREGISTER_FN + } jit_actions_t; + + struct jit_code_entry { + struct jit_code_entry *next_entry; + struct jit_code_entry *prev_entry; + const char *symfile_addr; + uint64_t symfile_size; + }; + + struct jit_descriptor { + uint32_t version; + // This should be jit_actions_t, but we want to be specific about the + // bit-width. + uint32_t action_flag; + struct jit_code_entry *relevant_entry; + struct jit_code_entry *first_entry; + }; + + // We put information about the JITed function in this global, which the + // debugger reads. Make sure to specify the version statically, because the + // debugger checks the version before we can set it during runtime. + struct jit_descriptor __jit_debug_descriptor = { 1, 0, 0, 0 }; + + // Debuggers puts a breakpoint in this function. + LLVM_ATTRIBUTE_NOINLINE void __jit_debug_register_code() { } + +} + +namespace { + +// Buffer for an in-memory object file in executable memory +typedef llvm::DenseMap< const char*, + std::pair<std::size_t, jit_code_entry*> > + RegisteredObjectBufferMap; + +/// Global access point for the JIT debugging interface designed for use with a +/// singleton toolbox. Handles thread-safe registration and deregistration of +/// object files that are in executable memory managed by the client of this +/// class. +class GDBJITRegistrar : public JITRegistrar { + /// A map of in-memory object files that have been registered with the + /// JIT interface. + RegisteredObjectBufferMap ObjectBufferMap; + +public: + /// Instantiates the JIT service. + GDBJITRegistrar() : ObjectBufferMap() {} + + /// Unregisters each object that was previously registered and releases all + /// internal resources. + virtual ~GDBJITRegistrar(); + + /// Creates an entry in the JIT registry for the buffer @p Object, + /// which must contain an object file in executable memory with any + /// debug information for the debugger. + void registerObject(const ObjectBuffer &Object); + + /// Removes the internal registration of @p Object, and + /// frees associated resources. + /// Returns true if @p Object was found in ObjectBufferMap. + bool deregisterObject(const ObjectBuffer &Object); + +private: + /// Deregister the debug info for the given object file from the debugger + /// and delete any temporary copies. This private method does not remove + /// the function from Map so that it can be called while iterating over Map. + void deregisterObjectInternal(RegisteredObjectBufferMap::iterator I); +}; + +/// Lock used to serialize all jit registration events, since they +/// modify global variables. +llvm::sys::Mutex JITDebugLock; + +/// Acquire the lock and do the registration. +void NotifyDebugger(jit_code_entry* JITCodeEntry) { + llvm::MutexGuard locked(JITDebugLock); + __jit_debug_descriptor.action_flag = JIT_REGISTER_FN; + + // Insert this entry at the head of the list. + JITCodeEntry->prev_entry = NULL; + jit_code_entry* NextEntry = __jit_debug_descriptor.first_entry; + JITCodeEntry->next_entry = NextEntry; + if (NextEntry != NULL) { + NextEntry->prev_entry = JITCodeEntry; + } + __jit_debug_descriptor.first_entry = JITCodeEntry; + __jit_debug_descriptor.relevant_entry = JITCodeEntry; + __jit_debug_register_code(); +} + +GDBJITRegistrar::~GDBJITRegistrar() { + // Free all registered object files. + for (RegisteredObjectBufferMap::iterator I = ObjectBufferMap.begin(), E = ObjectBufferMap.end(); + I != E; ++I) { + // Call the private method that doesn't update the map so our iterator + // doesn't break. + deregisterObjectInternal(I); + } + ObjectBufferMap.clear(); +} + +void GDBJITRegistrar::registerObject(const ObjectBuffer &Object) { + + const char *Buffer = Object.getBufferStart(); + size_t Size = Object.getBufferSize(); + + assert(Buffer && "Attempt to register a null object with a debugger."); + assert(ObjectBufferMap.find(Buffer) == ObjectBufferMap.end() && + "Second attempt to perform debug registration."); + jit_code_entry* JITCodeEntry = new jit_code_entry(); + + if (JITCodeEntry == 0) { + llvm::report_fatal_error( + "Allocation failed when registering a JIT entry!\n"); + } + else { + JITCodeEntry->symfile_addr = Buffer; + JITCodeEntry->symfile_size = Size; + + ObjectBufferMap[Buffer] = std::make_pair(Size, JITCodeEntry); + NotifyDebugger(JITCodeEntry); + } +} + +bool GDBJITRegistrar::deregisterObject(const ObjectBuffer& Object) { + const char *Buffer = Object.getBufferStart(); + RegisteredObjectBufferMap::iterator I = ObjectBufferMap.find(Buffer); + + if (I != ObjectBufferMap.end()) { + deregisterObjectInternal(I); + ObjectBufferMap.erase(I); + return true; + } + return false; +} + +void GDBJITRegistrar::deregisterObjectInternal( + RegisteredObjectBufferMap::iterator I) { + + jit_code_entry*& JITCodeEntry = I->second.second; + + // Acquire the lock and do the unregistration. + { + llvm::MutexGuard locked(JITDebugLock); + __jit_debug_descriptor.action_flag = JIT_UNREGISTER_FN; + + // Remove the jit_code_entry from the linked list. + jit_code_entry* PrevEntry = JITCodeEntry->prev_entry; + jit_code_entry* NextEntry = JITCodeEntry->next_entry; + + if (NextEntry) { + NextEntry->prev_entry = PrevEntry; + } + if (PrevEntry) { + PrevEntry->next_entry = NextEntry; + } + else { + assert(__jit_debug_descriptor.first_entry == JITCodeEntry); + __jit_debug_descriptor.first_entry = NextEntry; + } + + // Tell the debugger which entry we removed, and unregister the code. + __jit_debug_descriptor.relevant_entry = JITCodeEntry; + __jit_debug_register_code(); + } + + delete JITCodeEntry; + JITCodeEntry = NULL; +} + +} // end namespace + +namespace llvm { + +JITRegistrar& JITRegistrar::getGDBRegistrar() { + static GDBJITRegistrar* sRegistrar = NULL; + if (sRegistrar == NULL) { + // The mutex is here so that it won't slow down access once the registrar + // is instantiated + llvm::MutexGuard locked(JITDebugLock); + // Check again to be sure another thread didn't create this while we waited + if (sRegistrar == NULL) { + sRegistrar = new GDBJITRegistrar; + } + } + return *sRegistrar; +} + +} // namespace llvm diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/JITRegistrar.h b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/JITRegistrar.h new file mode 100644 index 0000000..69e9dbe --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/JITRegistrar.h @@ -0,0 +1,43 @@ +//===-- JITRegistrar.h - Registers objects with a debugger ----------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_EXECUTION_ENGINE_JIT_REGISTRAR_H +#define LLVM_EXECUTION_ENGINE_JIT_REGISTRAR_H + +#include "llvm/ExecutionEngine/ObjectBuffer.h" + +namespace llvm { + +/// Global access point for the JIT debugging interface. +class JITRegistrar { +public: + /// Instantiates the JIT service. + JITRegistrar() {} + + /// Unregisters each object that was previously registered and releases all + /// internal resources. + virtual ~JITRegistrar() {} + + /// Creates an entry in the JIT registry for the buffer @p Object, + /// which must contain an object file in executable memory with any + /// debug information for the debugger. + virtual void registerObject(const ObjectBuffer &Object) = 0; + + /// Removes the internal registration of @p Object, and + /// frees associated resources. + /// Returns true if @p Object was previously registered. + virtual bool deregisterObject(const ObjectBuffer &Object) = 0; + + /// Returns a reference to a GDB JIT registrar singleton + static JITRegistrar& getGDBRegistrar(); +}; + +} // end namespace llvm + +#endif // LLVM_EXECUTION_ENGINE_JIT_REGISTRAR_H diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/ObjectImageCommon.h b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/ObjectImageCommon.h new file mode 100644 index 0000000..89350cc --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/ObjectImageCommon.h @@ -0,0 +1,78 @@ +//===-- ObjectImageCommon.h - Format independent executuable object image -===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file declares a file format independent ObjectImage class. +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_RUNTIMEDYLD_OBJECTIMAGECOMMON_H +#define LLVM_RUNTIMEDYLD_OBJECTIMAGECOMMON_H + +#include "llvm/ExecutionEngine/ObjectBuffer.h" +#include "llvm/ExecutionEngine/ObjectImage.h" +#include "llvm/Object/ObjectFile.h" + +namespace llvm { + +class ObjectImageCommon : public ObjectImage { + ObjectImageCommon(); // = delete + ObjectImageCommon(const ObjectImageCommon &other); // = delete + +protected: + object::ObjectFile *ObjFile; + + // This form of the constructor allows subclasses to use + // format-specific subclasses of ObjectFile directly + ObjectImageCommon(ObjectBuffer *Input, object::ObjectFile *Obj) + : ObjectImage(Input), // saves Input as Buffer and takes ownership + ObjFile(Obj) + { + } + +public: + ObjectImageCommon(ObjectBuffer* Input) + : ObjectImage(Input) // saves Input as Buffer and takes ownership + { + ObjFile = object::ObjectFile::createObjectFile(Buffer->getMemBuffer()); + } + virtual ~ObjectImageCommon() { delete ObjFile; } + + virtual object::symbol_iterator begin_symbols() const + { return ObjFile->begin_symbols(); } + virtual object::symbol_iterator end_symbols() const + { return ObjFile->end_symbols(); } + + virtual object::section_iterator begin_sections() const + { return ObjFile->begin_sections(); } + virtual object::section_iterator end_sections() const + { return ObjFile->end_sections(); } + + virtual /* Triple::ArchType */ unsigned getArch() const + { return ObjFile->getArch(); } + + virtual StringRef getData() const { return ObjFile->getData(); } + + virtual object::ObjectFile* getObjectFile() const { return ObjFile; } + + // Subclasses can override these methods to update the image with loaded + // addresses for sections and common symbols + virtual void updateSectionAddress(const object::SectionRef &Sec, + uint64_t Addr) {} + virtual void updateSymbolAddress(const object::SymbolRef &Sym, uint64_t Addr) + {} + + // Subclasses can override these methods to provide JIT debugging support + virtual void registerWithDebugger() {} + virtual void deregisterWithDebugger() {} +}; + +} // end namespace llvm + +#endif // LLVM_RUNTIMEDYLD_OBJECT_IMAGE_H + diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp new file mode 100644 index 0000000..409b25f --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp @@ -0,0 +1,537 @@ +//===-- RuntimeDyld.cpp - Run-time dynamic linker for MC-JIT ----*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Implementation of the MC-JIT runtime dynamic linker. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "dyld" +#include "llvm/ExecutionEngine/RuntimeDyld.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 { + +// 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 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("Attempting to remap address of unknown section!"); +} + +// 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); +} + +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; + bool IsCode; + section_iterator si = obj->end_sections(); + Check(i->getFileOffset(FileOffset)); + Check(i->getSection(si)); + if (si == obj->end_sections()) continue; + Check(si->getContents(SectionData)); + Check(si->isText(IsCode)); + 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, IsCode, 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"); + } + + // 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); + } + } + + return obj.take(); +} + +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; + } +} + +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"); + } + + Sections.push_back(SectionEntry(Name, Addr, Allocate, DataSize, + (uintptr_t)pData)); + return SectionID; +} + +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; +} + +void RuntimeDyldImpl::addRelocationForSection(const RelocationEntry &RE, + unsigned SectionID) { + Relocations[SectionID].push_back(RE); +} + +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 { + // 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); + } +} + +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 || Arch == Triple::mips) { + 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::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); + } +} + +void RuntimeDyldImpl::resolveRelocationList(const RelocationList &Relocs, + uint64_t Value) { + for (unsigned i = 0, e = Relocs.size(); i != e; ++i) { + 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()) { + if (Name.size() == 0) { + // This is an absolute symbol, use an address of zero. + DEBUG(dbgs() << "Resolving absolute relocations." << "\n"); + resolveRelocationList(Relocs, 0); + } else { + // This is an external symbol, try to get its 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) { + // 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; +} + +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); +} + +void *RuntimeDyld::getSymbolAddress(StringRef Name) { + return Dyld->getSymbolAddress(Name); +} + +uint64_t RuntimeDyld::getSymbolLoadAddress(StringRef Name) { + return Dyld->getSymbolLoadAddress(Name); +} + +void RuntimeDyld::resolveRelocations() { + Dyld->resolveRelocations(); +} + +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() { + return Dyld->getErrorString(); +} + +} // end namespace llvm diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp new file mode 100644 index 0000000..b8537b1 --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp @@ -0,0 +1,851 @@ +//===-- RuntimeDyldELF.cpp - Run-time dynamic linker for MC-JIT -*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Implementation of ELF support for the MC-JIT runtime dynamic linker. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "dyld" +#include "RuntimeDyldELF.h" +#include "JITRegistrar.h" +#include "ObjectImageCommon.h" +#include "llvm/ADT/IntervalMap.h" +#include "llvm/ADT/OwningPtr.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/StringRef.h" +#include "llvm/ADT/Triple.h" +#include "llvm/ExecutionEngine/ObjectBuffer.h" +#include "llvm/ExecutionEngine/ObjectImage.h" +#include "llvm/Object/ELF.h" +#include "llvm/Object/ObjectFile.h" +#include "llvm/Support/ELF.h" +using namespace llvm; +using namespace llvm::object; + +namespace { + +static inline +error_code check(error_code Err) { + if (Err) { + report_fatal_error(Err.message()); + } + return Err; +} + +template<class ELFT> +class DyldELFObject + : public ELFObjectFile<ELFT> { + LLVM_ELF_IMPORT_TYPES(ELFT) + + typedef Elf_Shdr_Impl<ELFT> Elf_Shdr; + typedef Elf_Sym_Impl<ELFT> Elf_Sym; + typedef + Elf_Rel_Impl<ELFT, false> Elf_Rel; + typedef + Elf_Rel_Impl<ELFT, true> Elf_Rela; + + typedef Elf_Ehdr_Impl<ELFT> Elf_Ehdr; + + typedef typename ELFDataTypeTypedefHelper< + ELFT>::value_type addr_type; + +public: + DyldELFObject(MemoryBuffer *Wrapper, error_code &ec); + + void updateSectionAddress(const SectionRef &Sec, uint64_t Addr); + void updateSymbolAddress(const SymbolRef &Sym, uint64_t Addr); + + // Methods for type inquiry through isa, cast and dyn_cast + static inline bool classof(const Binary *v) { + return (isa<ELFObjectFile<ELFT> >(v) + && classof(cast<ELFObjectFile + <ELFT> >(v))); + } + static inline bool classof( + const ELFObjectFile<ELFT> *v) { + return v->isDyldType(); + } +}; + +template<class ELFT> +class ELFObjectImage : public ObjectImageCommon { + protected: + DyldELFObject<ELFT> *DyldObj; + bool Registered; + + public: + ELFObjectImage(ObjectBuffer *Input, + DyldELFObject<ELFT> *Obj) + : ObjectImageCommon(Input, Obj), + DyldObj(Obj), + Registered(false) {} + + virtual ~ELFObjectImage() { + 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) + { + DyldObj->updateSectionAddress(Sec, Addr); + } + + virtual void updateSymbolAddress(const SymbolRef &Sym, uint64_t Addr) + { + DyldObj->updateSymbolAddress(Sym, Addr); + } + + virtual void registerWithDebugger() + { + JITRegistrar::getGDBRegistrar().registerObject(*Buffer); + Registered = true; + } + virtual void deregisterWithDebugger() + { + JITRegistrar::getGDBRegistrar().deregisterObject(*Buffer); + } +}; + +// The MemoryBuffer passed into this constructor is just a wrapper around the +// actual memory. Ultimately, the Binary parent class will take ownership of +// this MemoryBuffer object but not the underlying memory. +template<class ELFT> +DyldELFObject<ELFT>::DyldELFObject(MemoryBuffer *Wrapper, error_code &ec) + : ELFObjectFile<ELFT>(Wrapper, ec) { + this->isDyldELFObject = true; +} + +template<class ELFT> +void DyldELFObject<ELFT>::updateSectionAddress(const SectionRef &Sec, + uint64_t Addr) { + DataRefImpl ShdrRef = Sec.getRawDataRefImpl(); + Elf_Shdr *shdr = const_cast<Elf_Shdr*>( + reinterpret_cast<const Elf_Shdr *>(ShdrRef.p)); + + // This assumes the address passed in matches the target address bitness + // The template-based type cast handles everything else. + shdr->sh_addr = static_cast<addr_type>(Addr); +} + +template<class ELFT> +void DyldELFObject<ELFT>::updateSymbolAddress(const SymbolRef &SymRef, + uint64_t Addr) { + + Elf_Sym *sym = const_cast<Elf_Sym*>( + ELFObjectFile<ELFT>::getSymbol(SymRef.getRawDataRefImpl())); + + // This assumes the address passed in matches the target address bitness + // The template-based type cast handles everything else. + sym->st_value = static_cast<addr_type>(Addr); +} + +} // namespace + +namespace llvm { + +ObjectImage *RuntimeDyldELF::createObjectImage(ObjectBuffer *Buffer) { + if (Buffer->getBufferSize() < ELF::EI_NIDENT) + llvm_unreachable("Unexpected ELF object size"); + std::pair<unsigned char, unsigned char> Ident = std::make_pair( + (uint8_t)Buffer->getBufferStart()[ELF::EI_CLASS], + (uint8_t)Buffer->getBufferStart()[ELF::EI_DATA]); + error_code ec; + + if (Ident.first == ELF::ELFCLASS32 && Ident.second == ELF::ELFDATA2LSB) { + DyldELFObject<ELFType<support::little, 4, false> > *Obj = + new DyldELFObject<ELFType<support::little, 4, false> >( + Buffer->getMemBuffer(), ec); + return new ELFObjectImage<ELFType<support::little, 4, false> >(Buffer, Obj); + } + else if (Ident.first == ELF::ELFCLASS32 && Ident.second == ELF::ELFDATA2MSB) { + DyldELFObject<ELFType<support::big, 4, false> > *Obj = + new DyldELFObject<ELFType<support::big, 4, false> >( + Buffer->getMemBuffer(), ec); + return new ELFObjectImage<ELFType<support::big, 4, false> >(Buffer, Obj); + } + else if (Ident.first == ELF::ELFCLASS64 && Ident.second == ELF::ELFDATA2MSB) { + DyldELFObject<ELFType<support::big, 8, true> > *Obj = + new DyldELFObject<ELFType<support::big, 8, true> >( + Buffer->getMemBuffer(), ec); + return new ELFObjectImage<ELFType<support::big, 8, true> >(Buffer, Obj); + } + else if (Ident.first == ELF::ELFCLASS64 && Ident.second == ELF::ELFDATA2LSB) { + DyldELFObject<ELFType<support::little, 8, true> > *Obj = + new DyldELFObject<ELFType<support::little, 8, true> >( + Buffer->getMemBuffer(), ec); + return new ELFObjectImage<ELFType<support::little, 8, true> >(Buffer, Obj); + } + else + llvm_unreachable("Unexpected ELF format"); +} + +RuntimeDyldELF::~RuntimeDyldELF() { +} + +void RuntimeDyldELF::resolveX86_64Relocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend) { + switch (Type) { + default: + llvm_unreachable("Relocation type not implemented yet!"); + break; + case ELF::R_X86_64_64: { + uint64_t *Target = reinterpret_cast<uint64_t*>(Section.Address + Offset); + *Target = Value + Addend; + DEBUG(dbgs() << "Writing " << format("%p", (Value + Addend)) + << " at " << format("%p\n",Target)); + break; + } + case ELF::R_X86_64_32: + case ELF::R_X86_64_32S: { + Value += Addend; + assert((Type == ELF::R_X86_64_32 && (Value <= UINT32_MAX)) || + (Type == ELF::R_X86_64_32S && + ((int64_t)Value <= INT32_MAX && (int64_t)Value >= INT32_MIN))); + uint32_t TruncatedAddr = (Value & 0xFFFFFFFF); + uint32_t *Target = reinterpret_cast<uint32_t*>(Section.Address + Offset); + *Target = TruncatedAddr; + DEBUG(dbgs() << "Writing " << format("%p", TruncatedAddr) + << " at " << format("%p\n",Target)); + break; + } + case ELF::R_X86_64_PC32: { + // Get the placeholder value from the generated object since + // a previous relocation attempt may have overwritten the loaded version + uint32_t *Placeholder = reinterpret_cast<uint32_t*>(Section.ObjAddress + + Offset); + uint32_t *Target = reinterpret_cast<uint32_t*>(Section.Address + Offset); + uint64_t FinalAddress = Section.LoadAddress + Offset; + int64_t RealOffset = *Placeholder + Value + Addend - FinalAddress; + assert(RealOffset <= INT32_MAX && RealOffset >= INT32_MIN); + int32_t TruncOffset = (RealOffset & 0xFFFFFFFF); + *Target = TruncOffset; + break; + } + } +} + +void RuntimeDyldELF::resolveX86Relocation(const SectionEntry &Section, + uint64_t Offset, + uint32_t Value, + uint32_t Type, + int32_t Addend) { + switch (Type) { + case ELF::R_386_32: { + // Get the placeholder value from the generated object since + // a previous relocation attempt may have overwritten the loaded version + uint32_t *Placeholder = reinterpret_cast<uint32_t*>(Section.ObjAddress + + Offset); + uint32_t *Target = reinterpret_cast<uint32_t*>(Section.Address + Offset); + *Target = *Placeholder + Value + Addend; + break; + } + case ELF::R_386_PC32: { + // Get the placeholder value from the generated object since + // a previous relocation attempt may have overwritten the loaded version + uint32_t *Placeholder = reinterpret_cast<uint32_t*>(Section.ObjAddress + + Offset); + uint32_t *Target = reinterpret_cast<uint32_t*>(Section.Address + Offset); + uint32_t FinalAddress = ((Section.LoadAddress + Offset) & 0xFFFFFFFF); + uint32_t RealOffset = *Placeholder + Value + Addend - FinalAddress; + *Target = RealOffset; + break; + } + default: + // There are other relocation types, but it appears these are the + // only ones currently used by the LLVM ELF object writer + llvm_unreachable("Relocation type not implemented yet!"); + break; + } +} + +void RuntimeDyldELF::resolveARMRelocation(const SectionEntry &Section, + uint64_t Offset, + uint32_t Value, + uint32_t Type, + int32_t Addend) { + // TODO: Add Thumb relocations. + uint32_t* TargetPtr = (uint32_t*)(Section.Address + Offset); + uint32_t FinalAddress = ((Section.LoadAddress + Offset) & 0xFFFFFFFF); + Value += Addend; + + DEBUG(dbgs() << "resolveARMRelocation, LocalAddress: " + << Section.Address + Offset + << " FinalAddress: " << format("%p",FinalAddress) + << " Value: " << format("%x",Value) + << " Type: " << format("%x",Type) + << " Addend: " << format("%x",Addend) + << "\n"); + + switch(Type) { + default: + llvm_unreachable("Not implemented relocation type!"); + + // Write a 32bit value to relocation address, taking into account the + // implicit addend encoded in the target. + case ELF::R_ARM_TARGET1 : + case ELF::R_ARM_ABS32 : + *TargetPtr += Value; + break; + + // Write first 16 bit of 32 bit value to the mov instruction. + // Last 4 bit should be shifted. + case ELF::R_ARM_MOVW_ABS_NC : + // We are not expecting any other addend in the relocation address. + // Using 0x000F0FFF because MOVW has its 16 bit immediate split into 2 + // non-contiguous fields. + assert((*TargetPtr & 0x000F0FFF) == 0); + Value = Value & 0xFFFF; + *TargetPtr |= Value & 0xFFF; + *TargetPtr |= ((Value >> 12) & 0xF) << 16; + break; + + // Write last 16 bit of 32 bit value to the mov instruction. + // Last 4 bit should be shifted. + case ELF::R_ARM_MOVT_ABS : + // We are not expecting any other addend in the relocation address. + // Use 0x000F0FFF for the same reason as R_ARM_MOVW_ABS_NC. + assert((*TargetPtr & 0x000F0FFF) == 0); + Value = (Value >> 16) & 0xFFFF; + *TargetPtr |= Value & 0xFFF; + *TargetPtr |= ((Value >> 12) & 0xF) << 16; + break; + + // Write 24 bit relative value to the branch instruction. + case ELF::R_ARM_PC24 : // Fall through. + case ELF::R_ARM_CALL : // Fall through. + case ELF::R_ARM_JUMP24 : + int32_t RelValue = static_cast<int32_t>(Value - FinalAddress - 8); + RelValue = (RelValue & 0x03FFFFFC) >> 2; + *TargetPtr &= 0xFF000000; + *TargetPtr |= RelValue; + break; + } +} + +void RuntimeDyldELF::resolveMIPSRelocation(const SectionEntry &Section, + uint64_t Offset, + uint32_t Value, + uint32_t Type, + int32_t Addend) { + uint32_t* TargetPtr = (uint32_t*)(Section.Address + Offset); + Value += Addend; + + DEBUG(dbgs() << "resolveMipselocation, LocalAddress: " + << Section.Address + Offset + << " FinalAddress: " + << format("%p",Section.LoadAddress + Offset) + << " Value: " << format("%x",Value) + << " Type: " << format("%x",Type) + << " Addend: " << format("%x",Addend) + << "\n"); + + switch(Type) { + default: + llvm_unreachable("Not implemented relocation type!"); + break; + case ELF::R_MIPS_32: + *TargetPtr = Value + (*TargetPtr); + break; + case ELF::R_MIPS_26: + *TargetPtr = ((*TargetPtr) & 0xfc000000) | (( Value & 0x0fffffff) >> 2); + break; + case ELF::R_MIPS_HI16: + // Get the higher 16-bits. Also add 1 if bit 15 is 1. + Value += ((*TargetPtr) & 0x0000ffff) << 16; + *TargetPtr = ((*TargetPtr) & 0xffff0000) | + (((Value + 0x8000) >> 16) & 0xffff); + break; + case ELF::R_MIPS_LO16: + Value += ((*TargetPtr) & 0x0000ffff); + *TargetPtr = ((*TargetPtr) & 0xffff0000) | (Value & 0xffff); + break; + } +} + +// Return the .TOC. section address to R_PPC64_TOC relocations. +uint64_t RuntimeDyldELF::findPPC64TOC() const { + // The TOC consists of sections .got, .toc, .tocbss, .plt in that + // order. The TOC starts where the first of these sections starts. + SectionList::const_iterator it = Sections.begin(); + SectionList::const_iterator ite = Sections.end(); + for (; it != ite; ++it) { + if (it->Name == ".got" || + it->Name == ".toc" || + it->Name == ".tocbss" || + it->Name == ".plt") + break; + } + if (it == ite) { + // This may happen for + // * references to TOC base base (sym@toc, .odp relocation) without + // a .toc directive. + // In this case just use the first section (which is usually + // the .odp) since the code won't reference the .toc base + // directly. + it = Sections.begin(); + } + assert (it != ite); + // Per the ppc64-elf-linux ABI, The TOC base is TOC value plus 0x8000 + // thus permitting a full 64 Kbytes segment. + return it->LoadAddress + 0x8000; +} + +// Returns the sections and offset associated with the ODP entry referenced +// by Symbol. +void RuntimeDyldELF::findOPDEntrySection(ObjectImage &Obj, + ObjSectionToIDMap &LocalSections, + RelocationValueRef &Rel) { + // Get the ELF symbol value (st_value) to compare with Relocation offset in + // .opd entries + + error_code err; + for (section_iterator si = Obj.begin_sections(), + se = Obj.end_sections(); si != se; si.increment(err)) { + StringRef SectionName; + check(si->getName(SectionName)); + if (SectionName != ".opd") + continue; + + for (relocation_iterator i = si->begin_relocations(), + e = si->end_relocations(); i != e;) { + check(err); + + // The R_PPC64_ADDR64 relocation indicates the first field + // of a .opd entry + uint64_t TypeFunc; + check(i->getType(TypeFunc)); + if (TypeFunc != ELF::R_PPC64_ADDR64) { + i.increment(err); + continue; + } + + SymbolRef TargetSymbol; + uint64_t TargetSymbolOffset; + int64_t TargetAdditionalInfo; + check(i->getSymbol(TargetSymbol)); + check(i->getOffset(TargetSymbolOffset)); + check(i->getAdditionalInfo(TargetAdditionalInfo)); + + i = i.increment(err); + if (i == e) + break; + check(err); + + // Just check if following relocation is a R_PPC64_TOC + uint64_t TypeTOC; + check(i->getType(TypeTOC)); + if (TypeTOC != ELF::R_PPC64_TOC) + continue; + + // Finally compares the Symbol value and the target symbol offset + // to check if this .opd entry refers to the symbol the relocation + // points to. + if (Rel.Addend != (intptr_t)TargetSymbolOffset) + continue; + + section_iterator tsi(Obj.end_sections()); + check(TargetSymbol.getSection(tsi)); + Rel.SectionID = findOrEmitSection(Obj, (*tsi), true, LocalSections); + Rel.Addend = (intptr_t)TargetAdditionalInfo; + return; + } + } + llvm_unreachable("Attempting to get address of ODP entry!"); +} + +// Relocation masks following the #lo(value), #hi(value), #higher(value), +// and #highest(value) macros defined in section 4.5.1. Relocation Types +// in PPC-elf64abi document. +// +static inline +uint16_t applyPPClo (uint64_t value) +{ + return value & 0xffff; +} + +static inline +uint16_t applyPPChi (uint64_t value) +{ + return (value >> 16) & 0xffff; +} + +static inline +uint16_t applyPPChigher (uint64_t value) +{ + return (value >> 32) & 0xffff; +} + +static inline +uint16_t applyPPChighest (uint64_t value) +{ + return (value >> 48) & 0xffff; +} + +void RuntimeDyldELF::resolvePPC64Relocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend) { + uint8_t* LocalAddress = Section.Address + Offset; + switch (Type) { + default: + llvm_unreachable("Relocation type not implemented yet!"); + break; + case ELF::R_PPC64_ADDR16_LO : + writeInt16BE(LocalAddress, applyPPClo (Value + Addend)); + break; + case ELF::R_PPC64_ADDR16_HI : + writeInt16BE(LocalAddress, applyPPChi (Value + Addend)); + break; + case ELF::R_PPC64_ADDR16_HIGHER : + writeInt16BE(LocalAddress, applyPPChigher (Value + Addend)); + break; + case ELF::R_PPC64_ADDR16_HIGHEST : + writeInt16BE(LocalAddress, applyPPChighest (Value + Addend)); + break; + case ELF::R_PPC64_ADDR14 : { + assert(((Value + Addend) & 3) == 0); + // Preserve the AA/LK bits in the branch instruction + uint8_t aalk = *(LocalAddress+3); + writeInt16BE(LocalAddress + 2, (aalk & 3) | ((Value + Addend) & 0xfffc)); + } break; + case ELF::R_PPC64_ADDR32 : { + int32_t Result = static_cast<int32_t>(Value + Addend); + if (SignExtend32<32>(Result) != Result) + llvm_unreachable("Relocation R_PPC64_ADDR32 overflow"); + writeInt32BE(LocalAddress, Result); + } break; + case ELF::R_PPC64_REL24 : { + uint64_t FinalAddress = (Section.LoadAddress + Offset); + int32_t delta = static_cast<int32_t>(Value - FinalAddress + Addend); + if (SignExtend32<24>(delta) != delta) + llvm_unreachable("Relocation R_PPC64_REL24 overflow"); + // Generates a 'bl <address>' instruction + writeInt32BE(LocalAddress, 0x48000001 | (delta & 0x03FFFFFC)); + } break; + case ELF::R_PPC64_REL32 : { + uint64_t FinalAddress = (Section.LoadAddress + Offset); + int32_t delta = static_cast<int32_t>(Value - FinalAddress + Addend); + if (SignExtend32<32>(delta) != delta) + llvm_unreachable("Relocation R_PPC64_REL32 overflow"); + writeInt32BE(LocalAddress, delta); + } break; + case ELF::R_PPC64_ADDR64 : + writeInt64BE(LocalAddress, Value + Addend); + break; + case ELF::R_PPC64_TOC : + writeInt64BE(LocalAddress, findPPC64TOC()); + break; + case ELF::R_PPC64_TOC16 : { + uint64_t TOCStart = findPPC64TOC(); + Value = applyPPClo((Value + Addend) - TOCStart); + writeInt16BE(LocalAddress, applyPPClo(Value)); + } break; + case ELF::R_PPC64_TOC16_DS : { + uint64_t TOCStart = findPPC64TOC(); + Value = ((Value + Addend) - TOCStart); + writeInt16BE(LocalAddress, applyPPClo(Value)); + } break; + } +} + +void RuntimeDyldELF::resolveRelocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend) { + switch (Arch) { + case Triple::x86_64: + resolveX86_64Relocation(Section, Offset, Value, Type, Addend); + break; + case Triple::x86: + resolveX86Relocation(Section, Offset, + (uint32_t)(Value & 0xffffffffL), Type, + (uint32_t)(Addend & 0xffffffffL)); + break; + case Triple::arm: // Fall through. + case Triple::thumb: + resolveARMRelocation(Section, Offset, + (uint32_t)(Value & 0xffffffffL), Type, + (uint32_t)(Addend & 0xffffffffL)); + break; + case Triple::mips: // Fall through. + case Triple::mipsel: + resolveMIPSRelocation(Section, Offset, + (uint32_t)(Value & 0xffffffffL), Type, + (uint32_t)(Addend & 0xffffffffL)); + break; + case Triple::ppc64: + resolvePPC64Relocation(Section, Offset, Value, Type, Addend); + break; + default: llvm_unreachable("Unsupported CPU type!"); + } +} + +void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, + ObjectImage &Obj, + ObjSectionToIDMap &ObjSectionToID, + const SymbolTableMap &Symbols, + StubMap &Stubs) { + + uint32_t RelType = (uint32_t)(Rel.Type & 0xffffffffL); + intptr_t Addend = (intptr_t)Rel.AdditionalInfo; + const SymbolRef &Symbol = Rel.Symbol; + + // Obtain the symbol name which is referenced in the relocation + StringRef TargetName; + Symbol.getName(TargetName); + DEBUG(dbgs() << "\t\tRelType: " << RelType + << " Addend: " << Addend + << " TargetName: " << TargetName + << "\n"); + RelocationValueRef Value; + // First search for the symbol in the local symbol table + SymbolTableMap::const_iterator lsi = Symbols.find(TargetName.data()); + SymbolRef::Type SymType; + Symbol.getType(SymType); + if (lsi != Symbols.end()) { + Value.SectionID = lsi->second.first; + Value.Addend = lsi->second.second; + } 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; + } else { + switch (SymType) { + case SymbolRef::ST_Debug: { + // TODO: Now ELF SymbolRef::ST_Debug = STT_SECTION, it's not obviously + // and can be changed by another developers. Maybe best way is add + // a new symbol type ST_Section to SymbolRef and use it. + section_iterator si(Obj.end_sections()); + Symbol.getSection(si); + if (si == Obj.end_sections()) + llvm_unreachable("Symbol section not found, bad object file format!"); + DEBUG(dbgs() << "\t\tThis is section symbol\n"); + // Default to 'true' in case isText fails (though it never does). + bool isCode = true; + si->isText(isCode); + Value.SectionID = findOrEmitSection(Obj, + (*si), + isCode, + ObjSectionToID); + Value.Addend = Addend; + break; + } + case SymbolRef::ST_Unknown: { + Value.SymbolName = TargetName.data(); + Value.Addend = Addend; + break; + } + default: + llvm_unreachable("Unresolved symbol type!"); + break; + } + } + } + DEBUG(dbgs() << "\t\tRel.SectionID: " << Rel.SectionID + << " Rel.Offset: " << Rel.Offset + << "\n"); + if (Arch == Triple::arm && + (RelType == ELF::R_ARM_PC24 || + RelType == ELF::R_ARM_CALL || + RelType == ELF::R_ARM_JUMP24)) { + // This is an ARM branch relocation, need to use a stub function. + DEBUG(dbgs() << "\t\tThis is an ARM branch relocation."); + SectionEntry &Section = Sections[Rel.SectionID]; + + // Look for an existing stub. + StubMap::const_iterator i = Stubs.find(Value); + if (i != Stubs.end()) { + resolveRelocation(Section, Rel.Offset, + (uint64_t)Section.Address + i->second, RelType, 0); + DEBUG(dbgs() << " Stub function found\n"); + } else { + // Create a new stub function. + DEBUG(dbgs() << " Create a new stub function\n"); + Stubs[Value] = Section.StubOffset; + uint8_t *StubTargetAddr = createStubFunction(Section.Address + + Section.StubOffset); + RelocationEntry RE(Rel.SectionID, StubTargetAddr - Section.Address, + ELF::R_ARM_ABS32, Value.Addend); + if (Value.SymbolName) + addRelocationForSymbol(RE, Value.SymbolName); + else + addRelocationForSection(RE, Value.SectionID); + + resolveRelocation(Section, Rel.Offset, + (uint64_t)Section.Address + Section.StubOffset, + RelType, 0); + Section.StubOffset += getMaxStubSize(); + } + } else if ((Arch == Triple::mipsel || Arch == Triple::mips) && + RelType == ELF::R_MIPS_26) { + // This is an Mips branch relocation, need to use a stub function. + DEBUG(dbgs() << "\t\tThis is a Mips branch relocation."); + SectionEntry &Section = Sections[Rel.SectionID]; + uint8_t *Target = Section.Address + Rel.Offset; + uint32_t *TargetAddress = (uint32_t *)Target; + + // Extract the addend from the instruction. + uint32_t Addend = ((*TargetAddress) & 0x03ffffff) << 2; + + Value.Addend += Addend; + + // Look up for existing stub. + StubMap::const_iterator i = Stubs.find(Value); + if (i != Stubs.end()) { + resolveRelocation(Section, Rel.Offset, + (uint64_t)Section.Address + i->second, RelType, 0); + DEBUG(dbgs() << " Stub function found\n"); + } else { + // Create a new stub function. + DEBUG(dbgs() << " Create a new stub function\n"); + Stubs[Value] = Section.StubOffset; + uint8_t *StubTargetAddr = createStubFunction(Section.Address + + Section.StubOffset); + + // Creating Hi and Lo relocations for the filled stub instructions. + RelocationEntry REHi(Rel.SectionID, + StubTargetAddr - Section.Address, + ELF::R_MIPS_HI16, Value.Addend); + RelocationEntry RELo(Rel.SectionID, + StubTargetAddr - Section.Address + 4, + ELF::R_MIPS_LO16, Value.Addend); + + if (Value.SymbolName) { + addRelocationForSymbol(REHi, Value.SymbolName); + addRelocationForSymbol(RELo, Value.SymbolName); + } else { + addRelocationForSection(REHi, Value.SectionID); + addRelocationForSection(RELo, Value.SectionID); + } + + resolveRelocation(Section, Rel.Offset, + (uint64_t)Section.Address + Section.StubOffset, + RelType, 0); + Section.StubOffset += getMaxStubSize(); + } + } else if (Arch == Triple::ppc64) { + if (RelType == ELF::R_PPC64_REL24) { + // A PPC branch relocation will need a stub function if the target is + // an external symbol (Symbol::ST_Unknown) or if the target address + // is not within the signed 24-bits branch address. + SectionEntry &Section = Sections[Rel.SectionID]; + uint8_t *Target = Section.Address + Rel.Offset; + bool RangeOverflow = false; + if (SymType != SymbolRef::ST_Unknown) { + // A function call may points to the .opd entry, so the final symbol value + // in calculated based in the relocation values in .opd section. + findOPDEntrySection(Obj, ObjSectionToID, Value); + uint8_t *RelocTarget = Sections[Value.SectionID].Address + Value.Addend; + int32_t delta = static_cast<int32_t>(Target - RelocTarget); + // If it is within 24-bits branch range, just set the branch target + if (SignExtend32<24>(delta) == delta) { + RelocationEntry RE(Rel.SectionID, Rel.Offset, RelType, Value.Addend); + if (Value.SymbolName) + addRelocationForSymbol(RE, Value.SymbolName); + else + addRelocationForSection(RE, Value.SectionID); + } else { + RangeOverflow = true; + } + } + if (SymType == SymbolRef::ST_Unknown || RangeOverflow == true) { + // It is an external symbol (SymbolRef::ST_Unknown) or within a range + // larger than 24-bits. + StubMap::const_iterator i = Stubs.find(Value); + if (i != Stubs.end()) { + // Symbol function stub already created, just relocate to it + resolveRelocation(Section, Rel.Offset, + (uint64_t)Section.Address + i->second, RelType, 0); + DEBUG(dbgs() << " Stub function found\n"); + } else { + // Create a new stub function. + DEBUG(dbgs() << " Create a new stub function\n"); + Stubs[Value] = Section.StubOffset; + uint8_t *StubTargetAddr = createStubFunction(Section.Address + + Section.StubOffset); + RelocationEntry RE(Rel.SectionID, StubTargetAddr - Section.Address, + ELF::R_PPC64_ADDR64, Value.Addend); + + // Generates the 64-bits address loads as exemplified in section + // 4.5.1 in PPC64 ELF ABI. + RelocationEntry REhst(Rel.SectionID, + StubTargetAddr - Section.Address + 2, + ELF::R_PPC64_ADDR16_HIGHEST, Value.Addend); + RelocationEntry REhr(Rel.SectionID, + StubTargetAddr - Section.Address + 6, + ELF::R_PPC64_ADDR16_HIGHER, Value.Addend); + RelocationEntry REh(Rel.SectionID, + StubTargetAddr - Section.Address + 14, + ELF::R_PPC64_ADDR16_HI, Value.Addend); + RelocationEntry REl(Rel.SectionID, + StubTargetAddr - Section.Address + 18, + ELF::R_PPC64_ADDR16_LO, Value.Addend); + + if (Value.SymbolName) { + addRelocationForSymbol(REhst, Value.SymbolName); + addRelocationForSymbol(REhr, Value.SymbolName); + addRelocationForSymbol(REh, Value.SymbolName); + addRelocationForSymbol(REl, Value.SymbolName); + } else { + addRelocationForSection(REhst, Value.SectionID); + addRelocationForSection(REhr, Value.SectionID); + addRelocationForSection(REh, Value.SectionID); + addRelocationForSection(REl, Value.SectionID); + } + + resolveRelocation(Section, Rel.Offset, + (uint64_t)Section.Address + Section.StubOffset, + RelType, 0); + if (SymType == SymbolRef::ST_Unknown) + // Restore the TOC for external calls + writeInt32BE(Target+4, 0xE8410028); // ld r2,40(r1) + Section.StubOffset += getMaxStubSize(); + } + } + } else { + RelocationEntry RE(Rel.SectionID, Rel.Offset, RelType, Value.Addend); + // Extra check to avoid relocation againt empty symbols (usually + // the R_PPC64_TOC). + if (Value.SymbolName && !TargetName.empty()) + addRelocationForSymbol(RE, Value.SymbolName); + else + addRelocationForSection(RE, Value.SectionID); + } + } else { + RelocationEntry RE(Rel.SectionID, Rel.Offset, RelType, Value.Addend); + if (Value.SymbolName) + addRelocationForSymbol(RE, Value.SymbolName); + else + addRelocationForSection(RE, Value.SectionID); + } +} + +unsigned RuntimeDyldELF::getCommonSymbolAlignment(const SymbolRef &Sym) { + // In ELF, the value of an SHN_COMMON symbol is its alignment requirement. + uint64_t Align; + Check(Sym.getValue(Align)); + return Align; +} + +bool RuntimeDyldELF::isCompatibleFormat(const ObjectBuffer *Buffer) const { + if (Buffer->getBufferSize() < strlen(ELF::ElfMagic)) + return false; + return (memcmp(Buffer->getBufferStart(), ELF::ElfMagic, strlen(ELF::ElfMagic))) == 0; +} +} // namespace llvm diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h new file mode 100644 index 0000000..07e704b --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h @@ -0,0 +1,97 @@ +//===-- RuntimeDyldELF.h - Run-time dynamic linker for MC-JIT ---*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// ELF support for MC-JIT runtime dynamic linker. +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_RUNTIME_DYLD_ELF_H +#define LLVM_RUNTIME_DYLD_ELF_H + +#include "RuntimeDyldImpl.h" + +using namespace llvm; + +namespace llvm { + +namespace { + // Helper for extensive error checking in debug builds. + error_code Check(error_code Err) { + if (Err) { + report_fatal_error(Err.message()); + } + return Err; + } +} // end anonymous namespace + +class RuntimeDyldELF : public RuntimeDyldImpl { +protected: + void resolveX86_64Relocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend); + + void resolveX86Relocation(const SectionEntry &Section, + uint64_t Offset, + uint32_t Value, + uint32_t Type, + int32_t Addend); + + void resolveARMRelocation(const SectionEntry &Section, + uint64_t Offset, + uint32_t Value, + uint32_t Type, + int32_t Addend); + + void resolveMIPSRelocation(const SectionEntry &Section, + uint64_t Offset, + uint32_t Value, + uint32_t Type, + int32_t Addend); + + void resolvePPC64Relocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend); + + virtual void resolveRelocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend); + + virtual void processRelocationRef(const ObjRelocationInfo &Rel, + ObjectImage &Obj, + ObjSectionToIDMap &ObjSectionToID, + const SymbolTableMap &Symbols, + StubMap &Stubs); + + unsigned getCommonSymbolAlignment(const SymbolRef &Sym); + + virtual ObjectImage *createObjectImage(ObjectBuffer *InputBuffer); + + uint64_t findPPC64TOC() const; + void findOPDEntrySection(ObjectImage &Obj, + ObjSectionToIDMap &LocalSections, + RelocationValueRef &Rel); + +public: + RuntimeDyldELF(RTDyldMemoryManager *mm) + : RuntimeDyldImpl(mm) {} + + virtual ~RuntimeDyldELF(); + + bool isCompatibleFormat(const ObjectBuffer *Buffer) const; +}; + +} // end namespace llvm + +#endif diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h new file mode 100644 index 0000000..f100994 --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h @@ -0,0 +1,344 @@ +//===-- RuntimeDyldImpl.h - Run-time dynamic linker for MC-JIT --*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Interface for the implementations of runtime dynamic linker facilities. +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_RUNTIME_DYLD_IMPL_H +#define LLVM_RUNTIME_DYLD_IMPL_H + +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/StringMap.h" +#include "llvm/ADT/Triple.h" +#include "llvm/ExecutionEngine/ObjectImage.h" +#include "llvm/ExecutionEngine/RuntimeDyld.h" +#include "llvm/Object/ObjectFile.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/Format.h" +#include "llvm/Support/Host.h" +#include "llvm/Support/SwapByteOrder.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Support/system_error.h" +#include <map> + +using namespace llvm; +using namespace llvm::object; + +namespace llvm { + +class ObjectBuffer; +class Twine; + + +/// SectionEntry - represents a section emitted into memory by the dynamic +/// linker. +class SectionEntry { +public: + /// Name - section name. + StringRef Name; + + /// Address - address in the linker's memory where the section resides. + uint8_t *Address; + + /// Size - section size. + size_t Size; + + /// LoadAddress - the address of the section in the target process's memory. + /// Used for situations in which JIT-ed code is being executed in the address + /// space of a separate process. If the code executes in the same address + /// space where it was JIT-ed, this just equals Address. + uint64_t LoadAddress; + + /// StubOffset - used for architectures with stub functions for far + /// relocations (like ARM). + uintptr_t StubOffset; + + /// ObjAddress - address of the section in the in-memory object file. Used + /// for calculating relocations in some object formats (like MachO). + uintptr_t ObjAddress; + + SectionEntry(StringRef name, uint8_t *address, size_t size, + uintptr_t stubOffset, uintptr_t objAddress) + : Name(name), Address(address), Size(size), LoadAddress((uintptr_t)address), + StubOffset(stubOffset), ObjAddress(objAddress) {} +}; + +/// RelocationEntry - used to represent relocations internally in the dynamic +/// linker. +class RelocationEntry { +public: + /// SectionID - the section this relocation points to. + unsigned SectionID; + + /// Offset - offset into the section. + uintptr_t Offset; + + /// RelType - relocation type. + uint32_t RelType; + + /// Addend - the relocation addend encoded in the instruction itself. Also + /// used to make a relocation section relative instead of symbol relative. + intptr_t Addend; + + RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend) + : SectionID(id), Offset(offset), RelType(type), Addend(addend) {} +}; + +/// ObjRelocationInfo - relocation information as read from the object file. +/// Used to pass around data taken from object::RelocationRef, together with +/// the section to which the relocation points (represented by a SectionID). +class ObjRelocationInfo { +public: + unsigned SectionID; + uint64_t Offset; + SymbolRef Symbol; + uint64_t Type; + int64_t AdditionalInfo; +}; + +class RelocationValueRef { +public: + unsigned SectionID; + intptr_t Addend; + const char *SymbolName; + RelocationValueRef(): SectionID(0), Addend(0), SymbolName(0) {} + + inline bool operator==(const RelocationValueRef &Other) const { + return std::memcmp(this, &Other, sizeof(RelocationValueRef)) == 0; + } + inline bool operator <(const RelocationValueRef &Other) const { + return std::memcmp(this, &Other, sizeof(RelocationValueRef)) < 0; + } +}; + +class RuntimeDyldImpl { +protected: + // The MemoryManager to load objects into. + RTDyldMemoryManager *MemMgr; + + // A list of all sections emitted by the dynamic linker. These sections are + // referenced in the code by means of their index in this list - SectionID. + typedef SmallVector<SectionEntry, 64> SectionList; + SectionList Sections; + + // Keep a map of sections from object file to the SectionID which + // references it. + typedef std::map<SectionRef, unsigned> ObjSectionToIDMap; + + // A global symbol table for symbols from all loaded modules. Maps the + // symbol name to a (SectionID, offset in section) pair. + typedef std::pair<unsigned, uintptr_t> SymbolLoc; + typedef StringMap<SymbolLoc> SymbolTableMap; + SymbolTableMap GlobalSymbolTable; + + // Pair representing the size and alignment requirement for a common symbol. + typedef std::pair<unsigned, unsigned> CommonSymbolInfo; + // Keep a map of common symbols to their info pairs + typedef std::map<SymbolRef, CommonSymbolInfo> CommonSymbolMap; + + // 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. + // The symbol (or section) the relocation is sourced from is the Key + // in the relocation list where it's stored. + typedef SmallVector<RelocationEntry, 64> RelocationList; + // Relocations to sections already loaded. Indexed by SectionID which is the + // source of the address. The target where the address will be written is + // SectionID/Offset in the relocation itself. + DenseMap<unsigned, RelocationList> Relocations; + + // Relocations to external symbols that are not yet resolved. Symbols are + // external when they aren't found in the global symbol table of all loaded + // modules. This map is indexed by symbol name. + StringMap<RelocationList> ExternalSymbolRelocations; + + typedef std::map<RelocationValueRef, uintptr_t> StubMap; + + Triple::ArchType Arch; + + inline unsigned getMaxStubSize() { + if (Arch == Triple::arm || Arch == Triple::thumb) + return 8; // 32-bit instruction and 32-bit address + else if (Arch == Triple::mipsel || Arch == Triple::mips) + return 16; + else if (Arch == Triple::ppc64) + return 44; + else + return 0; + } + + 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; + } + + uint64_t getSectionLoadAddress(unsigned SectionID) { + return Sections[SectionID].LoadAddress; + } + + uint8_t *getSectionAddress(unsigned SectionID) { + return (uint8_t*)Sections[SectionID].Address; + } + + // Subclasses can override this method to get the alignment requirement of + // a common symbol. Returns no alignment requirement if not implemented. + virtual unsigned getCommonSymbolAlignment(const SymbolRef &Sym) { + return 0; + } + + + void writeInt16BE(uint8_t *Addr, uint16_t Value) { + if (sys::isLittleEndianHost()) + Value = sys::SwapByteOrder(Value); + *Addr = (Value >> 8) & 0xFF; + *(Addr+1) = Value & 0xFF; + } + + void writeInt32BE(uint8_t *Addr, uint32_t Value) { + if (sys::isLittleEndianHost()) + Value = sys::SwapByteOrder(Value); + *Addr = (Value >> 24) & 0xFF; + *(Addr+1) = (Value >> 16) & 0xFF; + *(Addr+2) = (Value >> 8) & 0xFF; + *(Addr+3) = Value & 0xFF; + } + + void writeInt64BE(uint8_t *Addr, uint64_t Value) { + if (sys::isLittleEndianHost()) + Value = sys::SwapByteOrder(Value); + *Addr = (Value >> 56) & 0xFF; + *(Addr+1) = (Value >> 48) & 0xFF; + *(Addr+2) = (Value >> 40) & 0xFF; + *(Addr+3) = (Value >> 32) & 0xFF; + *(Addr+4) = (Value >> 24) & 0xFF; + *(Addr+5) = (Value >> 16) & 0xFF; + *(Addr+6) = (Value >> 8) & 0xFF; + *(Addr+7) = Value & 0xFF; + } + + /// \brief Given the common symbols discovered in the object file, emit a + /// new section for them and update the symbol mappings in the object and + /// symbol table. + void emitCommonSymbols(ObjectImage &Obj, + const CommonSymbolMap &CommonSymbols, + uint64_t TotalSize, + SymbolTableMap &SymbolTable); + + /// \brief Emits section data from the object file to the MemoryManager. + /// \param IsCode if it's true then allocateCodeSection() will be + /// used for emits, else allocateDataSection() will be used. + /// \return SectionID. + unsigned emitSection(ObjectImage &Obj, + const SectionRef &Section, + bool IsCode); + + /// \brief Find Section in LocalSections. If the secton is not found - emit + /// it and store in LocalSections. + /// \param IsCode if it's true then allocateCodeSection() will be + /// used for emmits, else allocateDataSection() will be used. + /// \return SectionID. + unsigned findOrEmitSection(ObjectImage &Obj, + const SectionRef &Section, + bool IsCode, + ObjSectionToIDMap &LocalSections); + + // \brief Add a relocation entry that uses the given section. + void addRelocationForSection(const RelocationEntry &RE, unsigned SectionID); + + // \brief Add a relocation entry that uses the given symbol. This symbol may + // be found in the global symbol table, or it may be external. + void addRelocationForSymbol(const RelocationEntry &RE, StringRef SymbolName); + + /// \brief Emits long jump instruction to Addr. + /// \return Pointer to the memory area for emitting target address. + uint8_t* createStubFunction(uint8_t *Addr); + + /// \brief Resolves relocations from Relocs list with address from Value. + void resolveRelocationList(const RelocationList &Relocs, uint64_t Value); + void resolveRelocationEntry(const RelocationEntry &RE, uint64_t Value); + + /// \brief A object file specific relocation resolver + /// \param Section The section where the relocation is being applied + /// \param Offset The offset into the section for this relocation + /// \param Value Target symbol address to apply the relocation action + /// \param Type object file specific relocation type + /// \param Addend A constant addend used to compute the value to be stored + /// into the relocatable field + virtual void resolveRelocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend) = 0; + + /// \brief Parses the object file relocation and stores it to Relocations + /// or SymbolRelocations (this depends on the object file type). + virtual void processRelocationRef(const ObjRelocationInfo &Rel, + ObjectImage &Obj, + ObjSectionToIDMap &ObjSectionToID, + const SymbolTableMap &Symbols, + StubMap &Stubs) = 0; + + /// \brief Resolve relocations to external symbols. + void resolveExternalSymbols(); + virtual ObjectImage *createObjectImage(ObjectBuffer *InputBuffer); +public: + RuntimeDyldImpl(RTDyldMemoryManager *mm) : MemMgr(mm), HasError(false) {} + + virtual ~RuntimeDyldImpl(); + + ObjectImage *loadObject(ObjectBuffer *InputBuffer); + + void *getSymbolAddress(StringRef Name) { + // FIXME: Just look up as a function for now. Overly simple of course. + // Work in progress. + if (GlobalSymbolTable.find(Name) == GlobalSymbolTable.end()) + return 0; + SymbolLoc Loc = GlobalSymbolTable.lookup(Name); + return getSectionAddress(Loc.first) + Loc.second; + } + + uint64_t getSymbolLoadAddress(StringRef Name) { + // FIXME: Just look up as a function for now. Overly simple of course. + // Work in progress. + if (GlobalSymbolTable.find(Name) == GlobalSymbolTable.end()) + return 0; + SymbolLoc Loc = GlobalSymbolTable.lookup(Name); + return getSectionLoadAddress(Loc.first) + Loc.second; + } + + void resolveRelocations(); + + void reassignSectionAddress(unsigned SectionID, uint64_t Addr); + + void mapSectionAddress(const void *LocalAddress, uint64_t TargetAddress); + + // 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; } + + virtual bool isCompatibleFormat(const ObjectBuffer *Buffer) const = 0; +}; + +} // end namespace llvm + + +#endif diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp new file mode 100644 index 0000000..bcc3df1 --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp @@ -0,0 +1,311 @@ +//===-- RuntimeDyldMachO.cpp - Run-time dynamic linker for MC-JIT -*- C++ -*-=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Implementation of the MC-JIT runtime dynamic linker. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "dyld" +#include "RuntimeDyldMachO.h" +#include "llvm/ADT/OwningPtr.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/StringRef.h" +using namespace llvm; +using namespace llvm::object; + +namespace llvm { + +void RuntimeDyldMachO::resolveRelocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend) { + uint8_t *LocalAddress = Section.Address + Offset; + uint64_t FinalAddress = Section.LoadAddress + Offset; + bool isPCRel = (Type >> 24) & 1; + unsigned MachoType = (Type >> 28) & 0xf; + unsigned Size = 1 << ((Type >> 25) & 3); + + DEBUG(dbgs() << "resolveRelocation LocalAddress: " + << format("%p", LocalAddress) + << " FinalAddress: " << format("%p", FinalAddress) + << " Value: " << format("%p", Value) + << " Addend: " << Addend + << " isPCRel: " << isPCRel + << " MachoType: " << MachoType + << " Size: " << Size + << "\n"); + + // This just dispatches to the proper target specific routine. + switch (Arch) { + default: llvm_unreachable("Unsupported CPU type!"); + case Triple::x86_64: + resolveX86_64Relocation(LocalAddress, + FinalAddress, + (uintptr_t)Value, + isPCRel, + MachoType, + Size, + Addend); + break; + case Triple::x86: + resolveI386Relocation(LocalAddress, + FinalAddress, + (uintptr_t)Value, + isPCRel, + MachoType, + Size, + Addend); + break; + case Triple::arm: // Fall through. + case Triple::thumb: + resolveARMRelocation(LocalAddress, + FinalAddress, + (uintptr_t)Value, + isPCRel, + MachoType, + Size, + Addend); + break; + } +} + +bool RuntimeDyldMachO::resolveI386Relocation(uint8_t *LocalAddress, + uint64_t FinalAddress, + uint64_t Value, + bool isPCRel, + unsigned Type, + unsigned Size, + int64_t Addend) { + if (isPCRel) + Value -= FinalAddress + 4; // see resolveX86_64Relocation + + switch (Type) { + default: + llvm_unreachable("Invalid relocation type!"); + case macho::RIT_Vanilla: { + uint8_t *p = LocalAddress; + uint64_t ValueToWrite = Value + Addend; + for (unsigned i = 0; i < Size; ++i) { + *p++ = (uint8_t)(ValueToWrite & 0xff); + ValueToWrite >>= 8; + } + return false; + } + case macho::RIT_Difference: + case macho::RIT_Generic_LocalDifference: + case macho::RIT_Generic_PreboundLazyPointer: + return Error("Relocation type not implemented yet!"); + } +} + +bool RuntimeDyldMachO::resolveX86_64Relocation(uint8_t *LocalAddress, + uint64_t FinalAddress, + uint64_t Value, + bool isPCRel, + unsigned Type, + unsigned Size, + int64_t Addend) { + // 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 -= FinalAddress + 4; + + switch(Type) { + 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*)LocalAddress; + 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: + return Error("Relocation type not implemented yet!"); + } +} + +bool RuntimeDyldMachO::resolveARMRelocation(uint8_t *LocalAddress, + uint64_t FinalAddress, + uint64_t Value, + bool isPCRel, + unsigned Type, + unsigned Size, + int64_t Addend) { + // If the relocation is PC-relative, the value to be encoded is the + // pointer difference. + if (isPCRel) { + 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) { + default: + llvm_unreachable("Invalid relocation type!"); + case macho::RIT_Vanilla: { + // 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*)LocalAddress; + for (unsigned i = 0; i < Size; ++i) { + *p++ = (uint8_t)Value; + Value >>= 8; + } + break; + } + case macho::RIT_ARM_Branch24Bit: { + // Mask the value into the target address. We know instructions are + // 32-bit aligned, so we can do it all at once. + uint32_t *p = (uint32_t*)LocalAddress; + // 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; +} + +void RuntimeDyldMachO::processRelocationRef(const ObjRelocationInfo &Rel, + ObjectImage &Obj, + ObjSectionToIDMap &ObjSectionToID, + const SymbolTableMap &Symbols, + StubMap &Stubs) { + + uint32_t RelType = (uint32_t) (Rel.Type & 0xffffffffL); + RelocationValueRef Value; + SectionEntry &Section = Sections[Rel.SectionID]; + + bool isExtern = (RelType >> 27) & 1; + if (isExtern) { + // Obtain the symbol name which is referenced in the relocation + StringRef TargetName; + const SymbolRef &Symbol = Rel.Symbol; + 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; + } 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; + } else + Value.SymbolName = TargetName.data(); + } + } else { + error_code err; + uint8_t sectionIndex = static_cast<uint8_t>(RelType & 0xFF); + section_iterator si = Obj.begin_sections(), + se = Obj.end_sections(); + for (uint8_t i = 1; i < sectionIndex; i++) { + error_code err; + si.increment(err); + if (si == se) + break; + } + assert(si != se && "No section containing relocation!"); + Value.SectionID = findOrEmitSection(Obj, *si, true, ObjSectionToID); + Value.Addend = 0; + // FIXME: The size and type of the relocation determines if we can + // encode an Addend in the target location itself, and if so, how many + // bytes we should read in order to get it. We don't yet support doing + // that, and just assuming it's sizeof(intptr_t) is blatantly wrong. + //Value.Addend = *(const intptr_t *)Target; + if (Value.Addend) { + // The MachO addend is an offset from the current section. We need it + // to be an offset from the destination section + Value.Addend += Section.ObjAddress - Sections[Value.SectionID].ObjAddress; + } + } + + if (Arch == Triple::arm && (RelType & 0xf) == macho::RIT_ARM_Branch24Bit) { + // This is an ARM branch relocation, need to use a stub function. + + // Look up for existing stub. + StubMap::const_iterator i = Stubs.find(Value); + if (i != Stubs.end()) + resolveRelocation(Section, Rel.Offset, + (uint64_t)Section.Address + i->second, + RelType, 0); + else { + // Create a new stub function. + Stubs[Value] = Section.StubOffset; + uint8_t *StubTargetAddr = createStubFunction(Section.Address + + Section.StubOffset); + RelocationEntry RE(Rel.SectionID, StubTargetAddr - Section.Address, + macho::RIT_Vanilla, Value.Addend); + if (Value.SymbolName) + addRelocationForSymbol(RE, Value.SymbolName); + else + addRelocationForSection(RE, Value.SectionID); + resolveRelocation(Section, Rel.Offset, + (uint64_t)Section.Address + Section.StubOffset, + RelType, 0); + Section.StubOffset += getMaxStubSize(); + } + } else { + RelocationEntry RE(Rel.SectionID, Rel.Offset, RelType, Value.Addend); + if (Value.SymbolName) + addRelocationForSymbol(RE, Value.SymbolName); + else + addRelocationForSection(RE, Value.SectionID); + } +} + + +bool RuntimeDyldMachO::isCompatibleFormat( + const ObjectBuffer *InputBuffer) const { + if (InputBuffer->getBufferSize() < 4) + return false; + StringRef Magic(InputBuffer->getBufferStart(), 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; +} + +} // end namespace llvm diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h new file mode 100644 index 0000000..62d8487 --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h @@ -0,0 +1,71 @@ +//===-- RuntimeDyldMachO.h - Run-time dynamic linker for MC-JIT ---*- C++ -*-=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// MachO support for MC-JIT runtime dynamic linker. +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_RUNTIME_DYLD_MACHO_H +#define LLVM_RUNTIME_DYLD_MACHO_H + +#include "RuntimeDyldImpl.h" +#include "llvm/ADT/IndexedMap.h" +#include "llvm/Object/MachOObject.h" +#include "llvm/Support/Format.h" + +using namespace llvm; +using namespace llvm::object; + + +namespace llvm { +class RuntimeDyldMachO : public RuntimeDyldImpl { +protected: + bool resolveI386Relocation(uint8_t *LocalAddress, + uint64_t FinalAddress, + uint64_t Value, + bool isPCRel, + unsigned Type, + unsigned Size, + int64_t Addend); + bool resolveX86_64Relocation(uint8_t *LocalAddress, + uint64_t FinalAddress, + uint64_t Value, + bool isPCRel, + unsigned Type, + unsigned Size, + int64_t Addend); + bool resolveARMRelocation(uint8_t *LocalAddress, + uint64_t FinalAddress, + uint64_t Value, + bool isPCRel, + unsigned Type, + unsigned Size, + int64_t Addend); + + virtual void processRelocationRef(const ObjRelocationInfo &Rel, + ObjectImage &Obj, + ObjSectionToIDMap &ObjSectionToID, + const SymbolTableMap &Symbols, + StubMap &Stubs); + +public: + virtual void resolveRelocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend); + + RuntimeDyldMachO(RTDyldMemoryManager *mm) : RuntimeDyldImpl(mm) {} + + bool isCompatibleFormat(const ObjectBuffer *Buffer) const; +}; + +} // end namespace llvm + +#endif |
