c7558a2c0d
This makes all functions end up in the same text section, which is important for TCling on macOS to catch exceptions from constructors: Stack unwinding requires information about program addresses to find out which objects to destroy and what code should be called to handle the exception. These addresses are relocated against a single __text section when loading the produced MachO binary, which breaks if the call sites of global constructors end up in a separate init section. Fixes ROOT-10703 and ROOT-10962
478 lines
18 KiB
C++
478 lines
18 KiB
C++
//--------------------------------------------------------------------*- C++ -*-
|
|
// CLING - the C++ LLVM-based InterpreterG :)
|
|
// author: Axel Naumann <axel@cern.ch>
|
|
//
|
|
// This file is dual-licensed: you can choose to license it under the University
|
|
// of Illinois Open Source License or the GNU Lesser General Public License. See
|
|
// LICENSE.TXT for details.
|
|
//------------------------------------------------------------------------------
|
|
|
|
#include "IncrementalJIT.h"
|
|
|
|
#include "IncrementalExecutor.h"
|
|
#include "cling/Utils/Platform.h"
|
|
|
|
#include "llvm/ExecutionEngine/Orc/LambdaResolver.h"
|
|
#include "llvm/ExecutionEngine/SectionMemoryManager.h"
|
|
#include "llvm/Support/DynamicLibrary.h"
|
|
|
|
#if defined(__APPLE__) || defined (_MSC_VER)
|
|
// Apple and Windows add an extra '_'
|
|
# define MANGLE_PREFIX "_"
|
|
#endif
|
|
|
|
using namespace llvm;
|
|
|
|
namespace {
|
|
|
|
///\brief Memory manager providing the lop-level link to the
|
|
/// IncrementalExecutor, handles missing or special / replaced symbols.
|
|
class ClingMemoryManager: public SectionMemoryManager {
|
|
public:
|
|
ClingMemoryManager(cling::IncrementalExecutor& Exe) {}
|
|
|
|
///\brief Simply wraps the base class's function setting AbortOnFailure
|
|
/// to false and instead using the error handling mechanism to report it.
|
|
void* getPointerToNamedFunction(const std::string &Name,
|
|
bool /*AbortOnFailure*/ =true) override {
|
|
return SectionMemoryManager::getPointerToNamedFunction(Name, false);
|
|
}
|
|
};
|
|
|
|
class NotifyFinalizedT {
|
|
public:
|
|
NotifyFinalizedT(cling::IncrementalJIT &jit) : m_JIT(jit) {}
|
|
void operator()(llvm::orc::RTDyldObjectLinkingLayerBase::ObjHandleT H) {
|
|
m_JIT.RemoveUnfinalizedSection(H);
|
|
}
|
|
|
|
private:
|
|
cling::IncrementalJIT &m_JIT;
|
|
};
|
|
|
|
} // unnamed namespace
|
|
|
|
namespace cling {
|
|
|
|
///\brief Memory manager for the OrcJIT layers to resolve symbols from the
|
|
/// common IncrementalJIT. I.e. the master of the Orcs.
|
|
/// Each ObjectLayer instance has one Azog object.
|
|
class Azog: public RTDyldMemoryManager {
|
|
cling::IncrementalJIT& m_jit;
|
|
|
|
struct AllocInfo {
|
|
uint8_t *m_Start = nullptr;
|
|
uint8_t *m_End = nullptr;
|
|
uint8_t *m_Current = nullptr;
|
|
|
|
void allocate(RTDyldMemoryManager *exeMM,
|
|
uintptr_t Size, uint32_t Align,
|
|
bool code, bool isReadOnly) {
|
|
|
|
uintptr_t RequiredSize = Size;
|
|
if (code)
|
|
m_Start = exeMM->allocateCodeSection(RequiredSize, Align,
|
|
0 /* SectionID */,
|
|
"codeReserve");
|
|
else if (isReadOnly)
|
|
m_Start = exeMM->allocateDataSection(RequiredSize, Align,
|
|
0 /* SectionID */,
|
|
"rodataReserve",isReadOnly);
|
|
else
|
|
m_Start = exeMM->allocateDataSection(RequiredSize, Align,
|
|
0 /* SectionID */,
|
|
"rwataReserve",isReadOnly);
|
|
m_Current = m_Start;
|
|
m_End = m_Start + RequiredSize;
|
|
}
|
|
|
|
uint8_t* getNextAddr(uintptr_t Size, unsigned Alignment) {
|
|
if (!Alignment)
|
|
Alignment = 16;
|
|
|
|
assert(!(Alignment & (Alignment - 1)) && "Alignment must be a power of two.");
|
|
|
|
uintptr_t RequiredSize = Alignment * ((Size + Alignment - 1)/Alignment + 1);
|
|
if ( (m_Current + RequiredSize) > m_End ) {
|
|
// This must be the last block.
|
|
if ((m_Current + Size) <= m_End) {
|
|
RequiredSize = Size;
|
|
} else {
|
|
cling::errs() << "Error in block allocation by Azog. "
|
|
<< "Not enough memory was reserved for the current module. "
|
|
<< Size << " (with alignment: " << RequiredSize
|
|
<< " ) is needed but\n"
|
|
<< "we only have " << (m_End - m_Current) << ".\n";
|
|
return nullptr;
|
|
}
|
|
}
|
|
|
|
uintptr_t Addr = (uintptr_t)m_Current;
|
|
|
|
// Align the address.
|
|
Addr = (Addr + Alignment - 1) & ~(uintptr_t)(Alignment - 1);
|
|
|
|
m_Current = (uint8_t*)(Addr + Size);
|
|
|
|
return (uint8_t*)Addr;
|
|
}
|
|
|
|
operator bool() {
|
|
return m_Current != nullptr;
|
|
}
|
|
};
|
|
|
|
AllocInfo m_Code;
|
|
AllocInfo m_ROData;
|
|
AllocInfo m_RWData;
|
|
|
|
#ifdef CLING_WIN_SEH_EXCEPTIONS
|
|
uintptr_t getBaseAddr() const {
|
|
if (LLVM_LIKELY(m_Code.m_Start && m_ROData.m_Start && m_RWData.m_Start)) {
|
|
return uintptr_t(std::min(std::min(m_Code.m_Start, m_ROData.m_Start),
|
|
m_RWData.m_Start));
|
|
}
|
|
if (LLVM_LIKELY(m_Code.m_Start)) {
|
|
return uintptr_t(m_ROData.m_Start
|
|
? std::min(m_Code.m_Start, m_ROData.m_Start)
|
|
: std::min(m_Code.m_Start, m_RWData.m_Start));
|
|
}
|
|
return uintptr_t(m_ROData.m_Start && m_RWData.m_Start
|
|
? std::min(m_ROData.m_Start, m_RWData.m_Start)
|
|
: std::max(m_ROData.m_Start, m_RWData.m_Start));
|
|
}
|
|
|
|
// FIXME: This is directly mirroring a structure in RTDyldMemoryManager that
|
|
// is private. Get Win64 exceptions into LLVM or add an accessor for it.
|
|
platform::windows::EHFrameInfos m_EHFrames;
|
|
#endif
|
|
|
|
public:
|
|
Azog(cling::IncrementalJIT& Jit): m_jit(Jit) {}
|
|
|
|
RTDyldMemoryManager* getExeMM() const { return m_jit.m_ExeMM.get(); }
|
|
|
|
uint8_t *allocateCodeSection(uintptr_t Size, unsigned Alignment,
|
|
unsigned SectionID,
|
|
StringRef SectionName) override {
|
|
uint8_t *Addr = nullptr;
|
|
if (m_Code) {
|
|
Addr = m_Code.getNextAddr(Size, Alignment);
|
|
}
|
|
if (!Addr) {
|
|
Addr = getExeMM()->allocateCodeSection(Size, Alignment, SectionID, SectionName);
|
|
m_jit.m_SectionsAllocatedSinceLastLoad.insert(Addr);
|
|
}
|
|
|
|
return Addr;
|
|
}
|
|
|
|
uint8_t *allocateDataSection(uintptr_t Size, unsigned Alignment,
|
|
unsigned SectionID, StringRef SectionName,
|
|
bool IsReadOnly) override {
|
|
|
|
uint8_t *Addr = nullptr;
|
|
if (IsReadOnly && m_ROData) {
|
|
Addr = m_ROData.getNextAddr(Size,Alignment);
|
|
} else if (m_RWData) {
|
|
Addr = m_RWData.getNextAddr(Size,Alignment);
|
|
}
|
|
if (!Addr) {
|
|
Addr = getExeMM()->allocateDataSection(Size, Alignment, SectionID,
|
|
SectionName, IsReadOnly);
|
|
m_jit.m_SectionsAllocatedSinceLastLoad.insert(Addr);
|
|
}
|
|
return Addr;
|
|
}
|
|
|
|
void reserveAllocationSpace(uintptr_t CodeSize, uint32_t CodeAlign,
|
|
uintptr_t RODataSize, uint32_t RODataAlign,
|
|
uintptr_t RWDataSize, uint32_t RWDataAlign) override {
|
|
m_Code.allocate(getExeMM(),CodeSize, CodeAlign, true, false);
|
|
m_ROData.allocate(getExeMM(),RODataSize, RODataAlign, false, true);
|
|
m_RWData.allocate(getExeMM(),RWDataSize, RWDataAlign, false, false);
|
|
|
|
m_jit.m_SectionsAllocatedSinceLastLoad.insert(m_Code.m_Start);
|
|
m_jit.m_SectionsAllocatedSinceLastLoad.insert(m_ROData.m_Start);
|
|
m_jit.m_SectionsAllocatedSinceLastLoad.insert(m_RWData.m_Start);
|
|
}
|
|
|
|
bool needsToReserveAllocationSpace() override {
|
|
return true; // getExeMM()->needsToReserveAllocationSpace();
|
|
}
|
|
|
|
void registerEHFrames(uint8_t *Addr, uint64_t LoadAddr,
|
|
size_t Size) override {
|
|
#ifdef CLING_WIN_SEH_EXCEPTIONS
|
|
const platform::windows::RuntimePRFunction PRFunc = { Addr, Size };
|
|
m_EHFrames.emplace_back(PRFunc);
|
|
#else
|
|
return getExeMM()->registerEHFrames(Addr, LoadAddr, Size);
|
|
#endif
|
|
}
|
|
|
|
void deregisterEHFrames() override {
|
|
#ifdef CLING_WIN_SEH_EXCEPTIONS
|
|
platform::DeRegisterEHFrames(getBaseAddr(), m_EHFrames);
|
|
platform::windows::EHFrameInfos().swap(m_EHFrames);
|
|
#else
|
|
return getExeMM()->deregisterEHFrames();
|
|
#endif
|
|
}
|
|
|
|
uint64_t getSymbolAddress(const std::string &Name) override {
|
|
// FIXME: We should decide if we want to handle the error here or make the
|
|
// return type of the function llvm::Expected<uint64_t> relying on the
|
|
// users to decide how to handle the error.
|
|
if (auto Addr = m_jit.getSymbolAddressWithoutMangling(Name,
|
|
true /*also use dlsym*/)
|
|
.getAddress())
|
|
return *Addr;
|
|
|
|
llvm_unreachable("Handle the error case");
|
|
return ~0U;
|
|
}
|
|
|
|
void *getPointerToNamedFunction(const std::string &Name,
|
|
bool AbortOnFailure = true) override {
|
|
return getExeMM()->getPointerToNamedFunction(Name, AbortOnFailure);
|
|
}
|
|
|
|
using llvm::RuntimeDyld::MemoryManager::notifyObjectLoaded;
|
|
|
|
void notifyObjectLoaded(ExecutionEngine *EE,
|
|
const object::ObjectFile &O) override {
|
|
return getExeMM()->notifyObjectLoaded(EE, O);
|
|
}
|
|
|
|
bool finalizeMemory(std::string *ErrMsg = nullptr) override {
|
|
// Each set of objects loaded will be finalized exactly once, but since
|
|
// symbol lookup during relocation may recursively trigger the
|
|
// loading/relocation of other modules, and since we're forwarding all
|
|
// finalizeMemory calls to a single underlying memory manager, we need to
|
|
// defer forwarding the call on until all necessary objects have been
|
|
// loaded. Otherwise, during the relocation of a leaf object, we will end
|
|
// up finalizing memory, causing a crash further up the stack when we
|
|
// attempt to apply relocations to finalized memory.
|
|
// To avoid finalizing too early, look at how many objects have been
|
|
// loaded but not yet finalized. This is a bit of a hack that relies on
|
|
// the fact that we're lazily emitting object files: The only way you can
|
|
// get more than one set of objects loaded but not yet finalized is if
|
|
// they were loaded during relocation of another set.
|
|
if (m_jit.m_UnfinalizedSections.size() == 1) {
|
|
#ifdef CLING_WIN_SEH_EXCEPTIONS
|
|
platform::RegisterEHFrames(getBaseAddr(), m_EHFrames, true);
|
|
#endif
|
|
return getExeMM()->finalizeMemory(ErrMsg);
|
|
}
|
|
return false;
|
|
};
|
|
|
|
}; // class Azog
|
|
|
|
IncrementalJIT::IncrementalJIT(IncrementalExecutor& exe,
|
|
std::unique_ptr<TargetMachine> TM):
|
|
m_Parent(exe),
|
|
m_TM(std::move(TM)),
|
|
m_TMDataLayout(m_TM->createDataLayout()),
|
|
m_ExeMM(std::make_shared<ClingMemoryManager>(m_Parent)),
|
|
m_NotifyObjectLoaded(*this),
|
|
m_ObjectLayer(m_SymbolMap, [this] () { return llvm::make_unique<Azog>(*this); },
|
|
m_NotifyObjectLoaded, NotifyFinalizedT(*this)),
|
|
m_CompileLayer(m_ObjectLayer, llvm::orc::SimpleCompiler(*m_TM)),
|
|
m_LazyEmitLayer(m_CompileLayer) {
|
|
// Libraries might get exposed through ExposeHiddenSharedLibrarySymbols(),
|
|
// make them available to the JIT, even though their symbols cannot be
|
|
// resolved through the process.
|
|
llvm::sys::DynamicLibrary::SearchOrder
|
|
= llvm::sys::DynamicLibrary::SO_LoadedLast;
|
|
|
|
// Enable JIT symbol resolution from the binary.
|
|
llvm::sys::DynamicLibrary::LoadLibraryPermanently(0, 0);
|
|
|
|
// Make debug symbols available.
|
|
m_GDBListener = 0; // JITEventListener::createGDBRegistrationListener();
|
|
|
|
// #if MCJIT
|
|
// llvm::EngineBuilder builder(std::move(m));
|
|
|
|
// std::string errMsg;
|
|
// builder.setErrorStr(&errMsg);
|
|
// builder.setOptLevel(llvm::CodeGenOpt::Less);
|
|
// builder.setEngineKind(llvm::EngineKind::JIT);
|
|
// std::unique_ptr<llvm::RTDyldMemoryManager>
|
|
// MemMan(new ClingMemoryManager(*this));
|
|
// builder.setMCJITMemoryManager(std::move(MemMan));
|
|
|
|
// // EngineBuilder uses default c'ted TargetOptions, too:
|
|
// llvm::TargetOptions TargetOpts;
|
|
// TargetOpts.NoFramePointerElim = 1;
|
|
// TargetOpts.JITEmitDebugInfo = 1;
|
|
|
|
// builder.setTargetOptions(TargetOpts);
|
|
|
|
// m_engine.reset(builder.create());
|
|
// assert(m_engine && "Cannot create module!");
|
|
// #endif
|
|
}
|
|
|
|
|
|
llvm::JITSymbol
|
|
IncrementalJIT::getInjectedSymbols(const std::string& Name) const {
|
|
using JITSymbol = llvm::JITSymbol;
|
|
auto SymMapI = m_SymbolMap.find(Name);
|
|
if (SymMapI != m_SymbolMap.end())
|
|
return JITSymbol(SymMapI->second, llvm::JITSymbolFlags::Exported);
|
|
|
|
return JITSymbol(nullptr);
|
|
}
|
|
|
|
std::pair<void*, bool>
|
|
IncrementalJIT::lookupSymbol(llvm::StringRef Name, void *InAddr, bool Jit) {
|
|
// FIXME: See comments on DLSym below.
|
|
#if !defined(LLVM_ON_WIN32)
|
|
void* Addr = llvm::sys::DynamicLibrary::SearchForAddressOfSymbol(Name);
|
|
#else
|
|
void* Addr = const_cast<void*>(platform::DLSym(Name));
|
|
#endif
|
|
|
|
if (InAddr && (!Addr || Jit)) {
|
|
if (Jit) {
|
|
std::string Key(Name);
|
|
#ifdef MANGLE_PREFIX
|
|
Key.insert(0, MANGLE_PREFIX);
|
|
#endif
|
|
m_SymbolMap[Key] = llvm::JITTargetAddress(InAddr);
|
|
}
|
|
llvm::sys::DynamicLibrary::AddSymbol(Name, InAddr);
|
|
return std::make_pair(InAddr, true);
|
|
}
|
|
return std::make_pair(Addr, false);
|
|
}
|
|
|
|
llvm::JITSymbol
|
|
IncrementalJIT::getSymbolAddressWithoutMangling(const std::string& Name,
|
|
bool AlsoInProcess) {
|
|
if (auto Sym = getInjectedSymbols(Name))
|
|
return Sym;
|
|
|
|
if (AlsoInProcess) {
|
|
if (llvm::JITSymbol SymInfo = m_ExeMM->findSymbol(Name)) {
|
|
if (auto AddrOrErr = SymInfo.getAddress())
|
|
return llvm::JITSymbol(*AddrOrErr, llvm::JITSymbolFlags::Exported);
|
|
else
|
|
llvm_unreachable("Handle the error case");
|
|
}
|
|
#ifdef LLVM_ON_WIN32
|
|
// FIXME: DLSym symbol lookup can overlap m_ExeMM->findSymbol wasting time
|
|
// looking for a symbol in libs where it is already known not to exist.
|
|
// Perhaps a better solution would be to have IncrementalJIT own the
|
|
// DynamicLibraryManger instance (or at least have a reference) that will
|
|
// look only through user loaded libraries.
|
|
// An upside to doing it this way is RTLD_GLOBAL won't need to be used
|
|
// allowing libs with competing symbols to co-exists.
|
|
if (const void* Sym = platform::DLSym(Name))
|
|
return llvm::JITSymbol(llvm::JITTargetAddress(Sym),
|
|
llvm::JITSymbolFlags::Exported);
|
|
#endif
|
|
}
|
|
|
|
if (auto Sym = m_LazyEmitLayer.findSymbol(Name, false))
|
|
return Sym;
|
|
|
|
return llvm::JITSymbol(nullptr);
|
|
}
|
|
|
|
void IncrementalJIT::addModule(const std::shared_ptr<llvm::Module>& module) {
|
|
// If this module doesn't have a DataLayout attached then attach the
|
|
// default.
|
|
module->setDataLayout(m_TMDataLayout);
|
|
|
|
// Reset the sections of all functions so that they end up in the same text
|
|
// section. This is important for TCling on macOS to catch exceptions raised
|
|
// by constructors, which requires unwinding information. The addresses in
|
|
// the __eh_frame table are relocated against a single __text section when
|
|
// loading the MachO binary, which breaks if the call sites of constructors
|
|
// end up in a separate init section.
|
|
// (see clang::TargetInfo::getStaticInitSectionSpecifier())
|
|
for (auto &Fn : module->functions()) {
|
|
if (Fn.hasSection()) {
|
|
// dbgs() << "Resetting section '" << Fn.getSection() << "' of function "
|
|
// << Fn.getName() << "\n";
|
|
Fn.setSection("");
|
|
}
|
|
}
|
|
|
|
// LLVM MERGE FIXME: update this to use new interfaces.
|
|
auto Resolver = llvm::orc::createLambdaResolver(
|
|
[&](const std::string &S) {
|
|
if (auto Sym = getInjectedSymbols(S)) {
|
|
if (auto AddrOrErr = Sym.getAddress())
|
|
return JITSymbol((uint64_t)*AddrOrErr, Sym.getFlags());
|
|
else
|
|
llvm_unreachable("Handle the error case");
|
|
}
|
|
return m_ExeMM->findSymbol(S);
|
|
},
|
|
[&](const std::string &Name) {
|
|
if (auto Sym = getSymbolAddressWithoutMangling(Name, true)) {
|
|
if (auto AddrOrErr = Sym.getAddress())
|
|
return JITSymbol(*AddrOrErr, Sym.getFlags());
|
|
else
|
|
llvm_unreachable("Handle the error case");
|
|
}
|
|
|
|
const std::string* NameNP = &Name;
|
|
#ifdef MANGLE_PREFIX
|
|
std::string NameNoPrefix;
|
|
const size_t PrfxLen = strlen(MANGLE_PREFIX);
|
|
if (!Name.compare(0, PrfxLen, MANGLE_PREFIX)) {
|
|
NameNoPrefix = Name.substr(PrfxLen);
|
|
NameNP = &NameNoPrefix;
|
|
}
|
|
#endif
|
|
|
|
/// This method returns the address of the specified function or variable
|
|
/// that could not be resolved by getSymbolAddress() or by resolving
|
|
/// possible weak symbols by the ExecutionEngine.
|
|
/// It is used to resolve symbols during module linking.
|
|
///
|
|
/// This might trigger autoloading and in turn jitting during static
|
|
/// initialization. That jitted initialization must be run before
|
|
/// NotifyLazyFunctionCreators() returns. The nested jitting must thus
|
|
/// finalize its memory without finalizing the current (outer) JIT
|
|
/// memory. To separate the outer and inner levels' memory, create a
|
|
/// dedicated ExeMM (with its CodeMem etc) and unfinalizedSection (used
|
|
/// to book-keep emitted sections in the IncrementalJIT). (ROOT-10426)
|
|
decltype(m_ExeMM) outerExeMM;
|
|
swap(outerExeMM, m_ExeMM);
|
|
m_ExeMM = std::make_shared<ClingMemoryManager>(m_Parent);
|
|
decltype(m_UnfinalizedSections) outerUnfinalizedSections;
|
|
swap(m_UnfinalizedSections, outerUnfinalizedSections);
|
|
uint64_t addr = uint64_t(getParent().NotifyLazyFunctionCreators(*NameNP));
|
|
swap(outerExeMM, m_ExeMM);
|
|
swap(m_UnfinalizedSections, outerUnfinalizedSections);
|
|
return JITSymbol(addr, llvm::JITSymbolFlags::Weak);
|
|
});
|
|
|
|
if (auto H = m_LazyEmitLayer.addModule(module, std::move(Resolver)))
|
|
m_UnloadPoints[module.get()] = *H;
|
|
else
|
|
llvm_unreachable("Handle the error case");
|
|
}
|
|
|
|
llvm::Error
|
|
IncrementalJIT::removeModule(const std::shared_ptr<llvm::Module>& module) {
|
|
// FIXME: Track down what calls this routine on a not-yet-added module. Once
|
|
// this is resolved we can remove this check enabling the assert.
|
|
auto IUnload = m_UnloadPoints.find(module.get());
|
|
if (IUnload == m_UnloadPoints.end())
|
|
return llvm::Error::success();
|
|
auto Handle = IUnload->second;
|
|
assert(*Handle && "Trying to remove a non existent module!");
|
|
m_UnloadPoints.erase(IUnload);
|
|
return m_LazyEmitLayer.removeModule(Handle);
|
|
}
|
|
|
|
}// end namespace cling
|