[RuntimeDyld] Work around a bug in RuntimeDyldELF exposed by r234839.
[oota-llvm.git] / tools / llvm-rtdyld / llvm-rtdyld.cpp
index 7b5bd0388d882d58424c4aa754ea0096a18c89b8..480032d3b6fe72fd1266d135ff054904c5adf11e 100644 (file)
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ADT/StringMap.h"
-#include "llvm/ADT/OwningPtr.h"
+#include "llvm/DebugInfo/DWARF/DIContext.h"
+#include "llvm/ExecutionEngine/RTDyldMemoryManager.h"
 #include "llvm/ExecutionEngine/RuntimeDyld.h"
-#include "llvm/ExecutionEngine/ObjectImage.h"
-#include "llvm/ExecutionEngine/ObjectBuffer.h"
-#include "llvm/Object/MachOObject.h"
+#include "llvm/ExecutionEngine/RuntimeDyldChecker.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCDisassembler.h"
+#include "llvm/MC/MCInstPrinter.h"
+#include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/Object/MachO.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Support/DynamicLibrary.h"
 #include "llvm/Support/ManagedStatic.h"
 #include "llvm/Support/Memory.h"
 #include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/PrettyStackTrace.h"
+#include "llvm/Support/Signals.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/TargetSelect.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Support/system_error.h"
+#include <list>
+#include <system_error>
+
 using namespace llvm;
 using namespace llvm::object;
 
@@ -31,7 +44,9 @@ InputFileList(cl::Positional, cl::ZeroOrMore,
               cl::desc("<input file>"));
 
 enum ActionType {
-  AC_Execute
+  AC_Execute,
+  AC_PrintLineInfo,
+  AC_Verify
 };
 
 static cl::opt<ActionType>
@@ -39,6 +54,10 @@ Action(cl::desc("Action to perform:"),
        cl::init(AC_Execute),
        cl::values(clEnumValN(AC_Execute, "execute",
                              "Load, link, and execute the inputs."),
+                  clEnumValN(AC_PrintLineInfo, "printline",
+                             "Load, link, and print line information for each function."),
+                  clEnumValN(AC_Verify, "verify",
+                             "Load, link and verify the resulting memory image."),
                   clEnumValEnd));
 
 static cl::opt<std::string>
@@ -46,6 +65,44 @@ EntryPoint("entry",
            cl::desc("Function to call as entry point."),
            cl::init("_main"));
 
+static cl::list<std::string>
+Dylibs("dylib",
+       cl::desc("Add library."),
+       cl::ZeroOrMore);
+
+static cl::opt<std::string>
+TripleName("triple", cl::desc("Target triple for disassembler"));
+
+static cl::list<std::string>
+CheckFiles("check",
+           cl::desc("File containing RuntimeDyld verifier checks."),
+           cl::ZeroOrMore);
+
+static cl::opt<uint64_t>
+TargetAddrStart("target-addr-start",
+                cl::desc("For -verify only: start of phony target address "
+                         "range."),
+                cl::init(4096), // Start at "page 1" - no allocating at "null".
+                cl::Hidden);
+
+static cl::opt<uint64_t>
+TargetAddrEnd("target-addr-end",
+              cl::desc("For -verify only: end of phony target address range."),
+              cl::init(~0ULL),
+              cl::Hidden);
+
+static cl::opt<uint64_t>
+TargetSectionSep("target-section-sep",
+                 cl::desc("For -verify only: Separation between sections in "
+                          "phony target address space."),
+                 cl::init(0),
+                 cl::Hidden);
+
+static cl::list<std::string>
+SpecificSectionMappings("map-section",
+                        cl::desc("Map a section to a specific address."),
+                        cl::ZeroOrMore);
+
 /* *** */
 
 // A trivial memory manager that doesn't do anything fancy, just uses the
@@ -56,15 +113,19 @@ public:
   SmallVector<sys::MemoryBlock, 16> DataMemory;
 
   uint8_t *allocateCodeSection(uintptr_t Size, unsigned Alignment,
-                               unsigned SectionID);
+                               unsigned SectionID,
+                               StringRef SectionName) override;
   uint8_t *allocateDataSection(uintptr_t Size, unsigned Alignment,
-                               unsigned SectionID);
+                               unsigned SectionID, StringRef SectionName,
+                               bool IsReadOnly) override;
 
-  virtual void *getPointerToNamedFunction(const std::string &Name,
-                                          bool AbortOnFailure = true) {
-    return 0;
+  void *getPointerToNamedFunction(const std::string &Name,
+                                  bool AbortOnFailure = true) override {
+    return nullptr;
   }
 
+  bool finalizeMemory(std::string *ErrMsg) override { return false; }
+
   // Invalidate instruction cache for sections with execute permissions.
   // Some platforms with separate data cache and instruction cache require
   // explicit cache flush, otherwise JIT code manipulations (like resolved
@@ -74,16 +135,19 @@ public:
 
 uint8_t *TrivialMemoryManager::allocateCodeSection(uintptr_t Size,
                                                    unsigned Alignment,
-                                                   unsigned SectionID) {
-  sys::MemoryBlock MB = sys::Memory::AllocateRWX(Size, 0, 0);
+                                                   unsigned SectionID,
+                                                   StringRef SectionName) {
+  sys::MemoryBlock MB = sys::Memory::AllocateRWX(Size, nullptr, nullptr);
   FunctionMemory.push_back(MB);
   return (uint8_t*)MB.base();
 }
 
 uint8_t *TrivialMemoryManager::allocateDataSection(uintptr_t Size,
                                                    unsigned Alignment,
-                                                   unsigned SectionID) {
-  sys::MemoryBlock MB = sys::Memory::AllocateRWX(Size, 0, 0);
+                                                   unsigned SectionID,
+                                                   StringRef SectionName,
+                                                   bool IsReadOnly) {
+  sys::MemoryBlock MB = sys::Memory::AllocateRWX(Size, nullptr, nullptr);
   DataMemory.push_back(MB);
   return (uint8_t*)MB.base();
 }
@@ -109,27 +173,126 @@ static int Error(const Twine &Msg) {
   return 1;
 }
 
+static void loadDylibs() {
+  for (const std::string &Dylib : Dylibs) {
+    if (sys::fs::is_regular_file(Dylib)) {
+      std::string ErrMsg;
+      if (sys::DynamicLibrary::LoadLibraryPermanently(Dylib.c_str(), &ErrMsg))
+        llvm::errs() << "Error loading '" << Dylib << "': "
+                     << ErrMsg << "\n";
+    } else
+      llvm::errs() << "Dylib not found: '" << Dylib << "'.\n";
+  }
+}
+
 /* *** */
 
+static int printLineInfoForInput() {
+  // Load any dylibs requested on the command line.
+  loadDylibs();
+
+  // If we don't have any input files, read from stdin.
+  if (!InputFileList.size())
+    InputFileList.push_back("-");
+  for(unsigned i = 0, e = InputFileList.size(); i != e; ++i) {
+    // Instantiate a dynamic linker.
+    TrivialMemoryManager MemMgr;
+    RuntimeDyld Dyld(MemMgr, MemMgr);
+
+    // Load the input memory buffer.
+
+    ErrorOr<std::unique_ptr<MemoryBuffer>> InputBuffer =
+        MemoryBuffer::getFileOrSTDIN(InputFileList[i]);
+    if (std::error_code EC = InputBuffer.getError())
+      return Error("unable to read input: '" + EC.message() + "'");
+
+    ErrorOr<std::unique_ptr<ObjectFile>> MaybeObj(
+      ObjectFile::createObjectFile((*InputBuffer)->getMemBufferRef()));
+
+    if (std::error_code EC = MaybeObj.getError())
+      return Error("unable to create object file: '" + EC.message() + "'");
+
+    ObjectFile &Obj = **MaybeObj;
+
+    // Load the object file
+    std::unique_ptr<RuntimeDyld::LoadedObjectInfo> LoadedObjInfo =
+      Dyld.loadObject(Obj);
+
+    if (Dyld.hasError())
+      return Error(Dyld.getErrorString());
+
+    // Resolve all the relocations we can.
+    Dyld.resolveRelocations();
+
+    OwningBinary<ObjectFile> DebugObj = LoadedObjInfo->getObjectForDebug(Obj);
+
+    std::unique_ptr<DIContext> Context(
+      DIContext::getDWARFContext(*DebugObj.getBinary()));
+
+    // Use symbol info to iterate functions in the object.
+    for (object::symbol_iterator I = DebugObj.getBinary()->symbol_begin(),
+                                 E = DebugObj.getBinary()->symbol_end();
+         I != E; ++I) {
+      object::SymbolRef::Type SymType;
+      if (I->getType(SymType)) continue;
+      if (SymType == object::SymbolRef::ST_Function) {
+        StringRef  Name;
+        uint64_t   Addr;
+        uint64_t   Size;
+        if (I->getName(Name)) continue;
+        if (I->getAddress(Addr)) continue;
+        if (I->getSize(Size)) continue;
+
+        outs() << "Function: " << Name << ", Size = " << Size << "\n";
+
+        DILineInfoTable Lines = Context->getLineInfoForAddressRange(Addr, Size);
+        DILineInfoTable::iterator  Begin = Lines.begin();
+        DILineInfoTable::iterator  End = Lines.end();
+        for (DILineInfoTable::iterator It = Begin; It != End; ++It) {
+          outs() << "  Line info @ " << It->first - Addr << ": "
+                 << It->second.FileName << ", line:" << It->second.Line << "\n";
+        }
+      }
+    }
+  }
+
+  return 0;
+}
+
 static int executeInput() {
+  // Load any dylibs requested on the command line.
+  loadDylibs();
+
   // Instantiate a dynamic linker.
-  TrivialMemoryManager *MemMgr = new TrivialMemoryManager;
-  RuntimeDyld Dyld(MemMgr);
+  TrivialMemoryManager MemMgr;
+  RuntimeDyld Dyld(MemMgr, MemMgr);
+
+  // FIXME: Preserve buffers until resolveRelocations time to work around a bug
+  //        in RuntimeDyldELF.
+  // This fixme should be fixed ASAP. This is a very brittle workaround.
+  std::vector<std::unique_ptr<MemoryBuffer>> InputBuffers;
 
   // If we don't have any input files, read from stdin.
   if (!InputFileList.size())
     InputFileList.push_back("-");
   for(unsigned i = 0, e = InputFileList.size(); i != e; ++i) {
     // Load the input memory buffer.
-    OwningPtr<MemoryBuffer> InputBuffer;
-    OwningPtr<ObjectImage>  LoadedObject;
-    if (error_code ec = MemoryBuffer::getFileOrSTDIN(InputFileList[i],
-                                                     InputBuffer))
-      return Error("unable to read input: '" + ec.message() + "'");
+    ErrorOr<std::unique_ptr<MemoryBuffer>> InputBuffer =
+        MemoryBuffer::getFileOrSTDIN(InputFileList[i]);
+    if (std::error_code EC = InputBuffer.getError())
+      return Error("unable to read input: '" + EC.message() + "'");
+    ErrorOr<std::unique_ptr<ObjectFile>> MaybeObj(
+      ObjectFile::createObjectFile((*InputBuffer)->getMemBufferRef()));
+
+    if (std::error_code EC = MaybeObj.getError())
+      return Error("unable to create object file: '" + EC.message() + "'");
+
+    ObjectFile &Obj = **MaybeObj;
+    InputBuffers.push_back(std::move(*InputBuffer));
 
     // Load the object file
-    LoadedObject.reset(Dyld.loadObject(new ObjectBuffer(InputBuffer.take())));
-    if (!LoadedObject) {
+    Dyld.loadObject(Obj);
+    if (Dyld.hasError()) {
       return Error(Dyld.getErrorString());
     }
   }
@@ -137,18 +300,18 @@ static int executeInput() {
   // Resolve all the relocations we can.
   Dyld.resolveRelocations();
   // Clear instruction cache before code will be executed.
-  MemMgr->invalidateInstructionCache();
+  MemMgr.invalidateInstructionCache();
 
   // FIXME: Error out if there are unresolved relocations.
 
   // Get the address of the entry point (_main by default).
-  void *MainAddress = Dyld.getSymbolAddress(EntryPoint);
-  if (MainAddress == 0)
+  void *MainAddress = Dyld.getSymbolLocalAddress(EntryPoint);
+  if (!MainAddress)
     return Error("no definition for '" + EntryPoint + "'");
 
   // Invalidate the instruction cache for each loaded function.
-  for (unsigned i = 0, e = MemMgr->FunctionMemory.size(); i != e; ++i) {
-    sys::MemoryBlock &Data = MemMgr->FunctionMemory[i];
+  for (unsigned i = 0, e = MemMgr.FunctionMemory.size(); i != e; ++i) {
+    sys::MemoryBlock &Data = MemMgr.FunctionMemory[i];
     // Make sure the memory is executable.
     std::string ErrorStr;
     sys::Memory::InvalidateInstructionCache(Data.base(), Data.size());
@@ -164,18 +327,274 @@ static int executeInput() {
   const char **Argv = new const char*[2];
   // Use the name of the first input object module as argv[0] for the target.
   Argv[0] = InputFileList[0].c_str();
-  Argv[1] = 0;
+  Argv[1] = nullptr;
   return Main(1, Argv);
 }
 
+static int checkAllExpressions(RuntimeDyldChecker &Checker) {
+  for (const auto& CheckerFileName : CheckFiles) {
+    ErrorOr<std::unique_ptr<MemoryBuffer>> CheckerFileBuf =
+        MemoryBuffer::getFileOrSTDIN(CheckerFileName);
+    if (std::error_code EC = CheckerFileBuf.getError())
+      return Error("unable to read input '" + CheckerFileName + "': " +
+                   EC.message());
+
+    if (!Checker.checkAllRulesInBuffer("# rtdyld-check:",
+                                       CheckerFileBuf.get().get()))
+      return Error("some checks in '" + CheckerFileName + "' failed");
+  }
+  return 0;
+}
+
+static std::map<void *, uint64_t>
+applySpecificSectionMappings(RuntimeDyldChecker &Checker) {
+
+  std::map<void*, uint64_t> SpecificMappings;
+
+  for (StringRef Mapping : SpecificSectionMappings) {
+
+    size_t EqualsIdx = Mapping.find_first_of("=");
+    StringRef SectionIDStr = Mapping.substr(0, EqualsIdx);
+    size_t ComaIdx = Mapping.find_first_of(",");
+
+    if (ComaIdx == StringRef::npos) {
+      errs() << "Invalid section specification '" << Mapping
+             << "'. Should be '<file name>,<section name>=<addr>'\n";
+      exit(1);
+    }
+
+    StringRef FileName = SectionIDStr.substr(0, ComaIdx);
+    StringRef SectionName = SectionIDStr.substr(ComaIdx + 1);
+
+    uint64_t OldAddrInt;
+    std::string ErrorMsg;
+    std::tie(OldAddrInt, ErrorMsg) =
+      Checker.getSectionAddr(FileName, SectionName, true);
+
+    if (ErrorMsg != "") {
+      errs() << ErrorMsg;
+      exit(1);
+    }
+
+    void* OldAddr = reinterpret_cast<void*>(static_cast<uintptr_t>(OldAddrInt));
+
+    StringRef NewAddrStr = Mapping.substr(EqualsIdx + 1);
+    uint64_t NewAddr;
+
+    if (NewAddrStr.getAsInteger(0, NewAddr)) {
+      errs() << "Invalid section address in mapping: " << Mapping << "\n";
+      exit(1);
+    }
+
+    Checker.getRTDyld().mapSectionAddress(OldAddr, NewAddr);
+    SpecificMappings[OldAddr] = NewAddr;
+  }
+
+  return SpecificMappings;
+}
+
+// Scatter sections in all directions!
+// Remaps section addresses for -verify mode. The following command line options
+// can be used to customize the layout of the memory within the phony target's
+// address space:
+// -target-addr-start <s> -- Specify where the phony target addres range starts.
+// -target-addr-end   <e> -- Specify where the phony target address range ends.
+// -target-section-sep <d> -- Specify how big a gap should be left between the
+//                            end of one section and the start of the next.
+//                            Defaults to zero. Set to something big
+//                            (e.g. 1 << 32) to stress-test stubs, GOTs, etc.
+//
+static void remapSections(const llvm::Triple &TargetTriple,
+                          const TrivialMemoryManager &MemMgr,
+                          RuntimeDyldChecker &Checker) {
+
+  // Set up a work list (section addr/size pairs).
+  typedef std::list<std::pair<void*, uint64_t>> WorklistT;
+  WorklistT Worklist;
+
+  for (const auto& CodeSection : MemMgr.FunctionMemory)
+    Worklist.push_back(std::make_pair(CodeSection.base(), CodeSection.size()));
+  for (const auto& DataSection : MemMgr.DataMemory)
+    Worklist.push_back(std::make_pair(DataSection.base(), DataSection.size()));
+
+  // Apply any section-specific mappings that were requested on the command
+  // line.
+  typedef std::map<void*, uint64_t> AppliedMappingsT;
+  AppliedMappingsT AppliedMappings = applySpecificSectionMappings(Checker);
+
+  // Keep an "already allocated" mapping of section target addresses to sizes.
+  // Sections whose address mappings aren't specified on the command line will
+  // allocated around the explicitly mapped sections while maintaining the
+  // minimum separation.
+  std::map<uint64_t, uint64_t> AlreadyAllocated;
+
+  // Move the previously applied mappings into the already-allocated map.
+  for (WorklistT::iterator I = Worklist.begin(), E = Worklist.end();
+       I != E;) {
+    WorklistT::iterator Tmp = I;
+    ++I;
+    AppliedMappingsT::iterator AI = AppliedMappings.find(Tmp->first);
+
+    if (AI != AppliedMappings.end()) {
+      AlreadyAllocated[AI->second] = Tmp->second;
+      Worklist.erase(Tmp);
+    }
+  }
+
+  // If the -target-addr-end option wasn't explicitly passed, then set it to a
+  // sensible default based on the target triple.
+  if (TargetAddrEnd.getNumOccurrences() == 0) {
+    if (TargetTriple.isArch16Bit())
+      TargetAddrEnd = (1ULL << 16) - 1;
+    else if (TargetTriple.isArch32Bit())
+      TargetAddrEnd = (1ULL << 32) - 1;
+    // TargetAddrEnd already has a sensible default for 64-bit systems, so
+    // there's nothing to do in the 64-bit case.
+  }
+
+  // Process any elements remaining in the worklist.
+  while (!Worklist.empty()) {
+    std::pair<void*, uint64_t> CurEntry = Worklist.front();
+    Worklist.pop_front();
+
+    uint64_t NextSectionAddr = TargetAddrStart;
+
+    for (const auto &Alloc : AlreadyAllocated)
+      if (NextSectionAddr + CurEntry.second + TargetSectionSep <= Alloc.first)
+        break;
+      else
+        NextSectionAddr = Alloc.first + Alloc.second + TargetSectionSep;
+
+    AlreadyAllocated[NextSectionAddr] = CurEntry.second;
+    Checker.getRTDyld().mapSectionAddress(CurEntry.first, NextSectionAddr);
+  }
+
+}
+
+// Load and link the objects specified on the command line, but do not execute
+// anything. Instead, attach a RuntimeDyldChecker instance and call it to
+// verify the correctness of the linked memory.
+static int linkAndVerify() {
+
+  // Check for missing triple.
+  if (TripleName == "") {
+    llvm::errs() << "Error: -triple required when running in -verify mode.\n";
+    return 1;
+  }
+
+  // Look up the target and build the disassembler.
+  Triple TheTriple(Triple::normalize(TripleName));
+  std::string ErrorStr;
+  const Target *TheTarget =
+    TargetRegistry::lookupTarget("", TheTriple, ErrorStr);
+  if (!TheTarget) {
+    llvm::errs() << "Error accessing target '" << TripleName << "': "
+                 << ErrorStr << "\n";
+    return 1;
+  }
+  TripleName = TheTriple.getTriple();
+
+  std::unique_ptr<MCSubtargetInfo> STI(
+    TheTarget->createMCSubtargetInfo(TripleName, "", ""));
+  assert(STI && "Unable to create subtarget info!");
+
+  std::unique_ptr<MCRegisterInfo> MRI(TheTarget->createMCRegInfo(TripleName));
+  assert(MRI && "Unable to create target register info!");
+
+  std::unique_ptr<MCAsmInfo> MAI(TheTarget->createMCAsmInfo(*MRI, TripleName));
+  assert(MAI && "Unable to create target asm info!");
+
+  MCContext Ctx(MAI.get(), MRI.get(), nullptr);
+
+  std::unique_ptr<MCDisassembler> Disassembler(
+    TheTarget->createMCDisassembler(*STI, Ctx));
+  assert(Disassembler && "Unable to create disassembler!");
+
+  std::unique_ptr<MCInstrInfo> MII(TheTarget->createMCInstrInfo());
+
+  std::unique_ptr<MCInstPrinter> InstPrinter(
+      TheTarget->createMCInstPrinter(Triple(TripleName), 0, *MAI, *MII, *MRI));
+
+  // Load any dylibs requested on the command line.
+  loadDylibs();
+
+  // Instantiate a dynamic linker.
+  TrivialMemoryManager MemMgr;
+  RuntimeDyld Dyld(MemMgr, MemMgr);
+  Dyld.setProcessAllSections(true);
+  RuntimeDyldChecker Checker(Dyld, Disassembler.get(), InstPrinter.get(),
+                             llvm::dbgs());
+
+  // FIXME: Preserve buffers until resolveRelocations time to work around a bug
+  //        in RuntimeDyldELF.
+  // This fixme should be fixed ASAP. This is a very brittle workaround.
+  std::vector<std::unique_ptr<MemoryBuffer>> InputBuffers;
+
+  // If we don't have any input files, read from stdin.
+  if (!InputFileList.size())
+    InputFileList.push_back("-");
+  for(unsigned i = 0, e = InputFileList.size(); i != e; ++i) {
+    // Load the input memory buffer.
+    ErrorOr<std::unique_ptr<MemoryBuffer>> InputBuffer =
+        MemoryBuffer::getFileOrSTDIN(InputFileList[i]);
+
+    if (std::error_code EC = InputBuffer.getError())
+      return Error("unable to read input: '" + EC.message() + "'");
+
+    ErrorOr<std::unique_ptr<ObjectFile>> MaybeObj(
+      ObjectFile::createObjectFile((*InputBuffer)->getMemBufferRef()));
+
+    if (std::error_code EC = MaybeObj.getError())
+      return Error("unable to create object file: '" + EC.message() + "'");
+
+    ObjectFile &Obj = **MaybeObj;
+    InputBuffers.push_back(std::move(*InputBuffer));
+
+    // Load the object file
+    Dyld.loadObject(Obj);
+    if (Dyld.hasError()) {
+      return Error(Dyld.getErrorString());
+    }
+  }
+
+  // Re-map the section addresses into the phony target address space.
+  remapSections(TheTriple, MemMgr, Checker);
+
+  // Resolve all the relocations we can.
+  Dyld.resolveRelocations();
+
+  // Register EH frames.
+  Dyld.registerEHFrames();
+
+  int ErrorCode = checkAllExpressions(Checker);
+  if (Dyld.hasError()) {
+    errs() << "RTDyld reported an error applying relocations:\n  "
+           << Dyld.getErrorString() << "\n";
+    ErrorCode = 1;
+  }
+
+  return ErrorCode;
+}
+
 int main(int argc, char **argv) {
+  sys::PrintStackTraceOnErrorSignal();
+  PrettyStackTraceProgram X(argc, argv);
+
   ProgramName = argv[0];
   llvm_shutdown_obj Y;  // Call llvm_shutdown() on exit.
 
+  llvm::InitializeAllTargetInfos();
+  llvm::InitializeAllTargetMCs();
+  llvm::InitializeAllDisassemblers();
+
   cl::ParseCommandLineOptions(argc, argv, "llvm MC-JIT tool\n");
 
   switch (Action) {
   case AC_Execute:
     return executeInput();
+  case AC_PrintLineInfo:
+    return printLineInfoForInput();
+  case AC_Verify:
+    return linkAndVerify();
   }
 }