Re-factored RuntimeDyld.

Added ExecutionEngine/MCJIT tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@153221 91177308-0d34-0410-b5e6-96231b3b80d8

56 files changed, 1823 insertions, 1071 deletions

diff --git a/include/llvm/ExecutionEngine/RuntimeDyld.h b/include/llvm/ExecutionEngine/RuntimeDyld.h
index 81acc08673..aabfd0341c 100644
--- a/include/llvm/ExecutionEngine/RuntimeDyld.h
+++ b/include/llvm/ExecutionEngine/RuntimeDyld.h
@@ -45,19 +45,9 @@ public:
   virtual uint8_t *allocateDataSection(uintptr_t Size, unsigned Alignment,
                                        unsigned SectionID) = 0;
 
-  // Allocate ActualSize bytes, or more, for the named function. Return
-  // a pointer to the allocated memory and update Size to reflect how much
-  // memory was acutally allocated.
-  virtual uint8_t *startFunctionBody(const char *Name, uintptr_t &Size) = 0;
-
-  // Mark the end of the function, including how much of the allocated
-  // memory was actually used.
-  virtual void endFunctionBody(const char *Name, uint8_t *FunctionStart,
-                               uint8_t *FunctionEnd) = 0;
-
-  
   virtual void *getPointerToNamedFunction(const std::string &Name,
                                           bool AbortOnFailure = true) = 0;
+
 };
 
 class RuntimeDyld {
diff --git a/lib/ExecutionEngine/MCJIT/MCJITMemoryManager.h b/lib/ExecutionEngine/MCJIT/MCJITMemoryManager.h
index 9ff6811c4b..dac8b26291 100644
--- a/lib/ExecutionEngine/MCJIT/MCJITMemoryManager.h
+++ b/lib/ExecutionEngine/MCJIT/MCJITMemoryManager.h
@@ -33,12 +33,12 @@ public:
 
   uint8_t *allocateDataSection(uintptr_t Size, unsigned Alignment,
                                unsigned SectionID) {
-    return JMM->allocateDataSection(Size, Alignment, SectionID);
+    return JMM->allocateSpace(Size, Alignment);
   }
 
   uint8_t *allocateCodeSection(uintptr_t Size, unsigned Alignment,
                                unsigned SectionID) {
-    return JMM->allocateCodeSection(Size, Alignment, SectionID);
+    return JMM->allocateSpace(Size, Alignment);
   }
 
   virtual void *getPointerToNamedFunction(const std::string &Name,
@@ -46,40 +46,6 @@ public:
     return JMM->getPointerToNamedFunction(Name, AbortOnFailure);
   }
 
-  // Allocate ActualSize bytes, or more, for the named function. Return
-  // a pointer to the allocated memory and update Size to reflect how much
-  // memory was acutally allocated.
-  uint8_t *startFunctionBody(const char *Name, uintptr_t &Size) {
-    // FIXME: This should really reference the MCAsmInfo to get the global
-    //        prefix.
-    if (Name[0] == '_') ++Name;
-    Function *F = M->getFunction(Name);
-    // Some ObjC names have a prefixed \01 in the IR. If we failed to find
-    // the symbol and it's of the ObjC conventions (starts with "-" or 
-    // "+"), try prepending a \01 and see if we can find it that way.
-    if (!F && (Name[0] == '-' || Name[0] == '+'))
-      F = M->getFunction((Twine("\1") + Name).str());
-    assert(F && "No matching function in JIT IR Module!");
-    return JMM->startFunctionBody(F, Size);
-  }
-
-  // Mark the end of the function, including how much of the allocated
-  // memory was actually used.
-  void endFunctionBody(const char *Name, uint8_t *FunctionStart,
-                       uint8_t *FunctionEnd) {
-    // FIXME: This should really reference the MCAsmInfo to get the global
-    //        prefix.
-    if (Name[0] == '_') ++Name;
-    Function *F = M->getFunction(Name);
-    // Some ObjC names have a prefixed \01 in the IR. If we failed to find
-    // the symbol and it's of the ObjC conventions (starts with "-" or
-    // "+"), try prepending a \01 and see if we can find it that way.
-    if (!F && (Name[0] == '-' || Name[0] == '+'))
-      F = M->getFunction((Twine("\1") + Name).str());
-    assert(F && "No matching function in JIT IR Module!");
-    JMM->endFunctionBody(F, FunctionStart, FunctionEnd);
-  }
-
 };
 
 } // End llvm namespace
diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp
index 2896c2d556..ff4a2c847e 100644
--- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp
+++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp
@@ -26,45 +26,290 @@ RuntimeDyldImpl::~RuntimeDyldImpl() {}
 
 namespace llvm {
 
-void RuntimeDyldImpl::extractFunction(StringRef Name, uint8_t *StartAddress,
-                                      uint8_t *EndAddress) {
-  // FIXME: DEPRECATED in favor of by-section allocation.
-  // Allocate memory for the function via the memory manager.
-  uintptr_t Size = EndAddress - StartAddress + 1;
-  uintptr_t AllocSize = Size;
-  uint8_t *Mem = MemMgr->startFunctionBody(Name.data(), AllocSize);
-  assert(Size >= (uint64_t)(EndAddress - StartAddress + 1) &&
-         "Memory manager failed to allocate enough memory!");
-  // Copy the function payload into the memory block.
-  memcpy(Mem, StartAddress, Size);
-  MemMgr->endFunctionBody(Name.data(), Mem, Mem + Size);
-  // Remember where we put it.
-  unsigned SectionID = Sections.size();
-  Sections.push_back(sys::MemoryBlock(Mem, Size));
 
-  // Default the assigned address for this symbol to wherever this
-  // allocated it.
-  SymbolTable[Name] = SymbolLoc(SectionID, 0);
-  DEBUG(dbgs() << "    allocated to [" << Mem << ", " << Mem + Size << "]\n");
-}
 
 // Resolve the relocations for all symbols we currently know about.
 void RuntimeDyldImpl::resolveRelocations() {
+  // First, resolve relocations assotiated with external symbols.
+  resolveSymbols();
+
   // 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) {
-    reassignSectionAddress(i, SectionLoadAddress[i]);
+    reassignSectionAddress(i, Sections[i].LoadAddress);
   }
 }
 
 void RuntimeDyldImpl::mapSectionAddress(void *LocalAddress,
                                         uint64_t TargetAddress) {
-  assert(SectionLocalMemToID.count(LocalAddress) &&
-         "Attempting to remap address of unknown section!");
-  unsigned SectionID = SectionLocalMemToID[LocalAddress];
-  reassignSectionAddress(SectionID, 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!");
+}
+
+bool RuntimeDyldImpl::loadObject(const MemoryBuffer *InputBuffer) {
+  // FIXME: ObjectFile don't modify MemoryBuffer.
+  //        It should use const MemoryBuffer as parameter.
+  ObjectFile *obj = ObjectFile::
+                      createObjectFile(const_cast<MemoryBuffer*>(InputBuffer));
+
+  Arch = (Triple::ArchType)obj->getArch();
+
+  LocalSymbolMap LocalSymbols;     // Functions and data symbols from the
+                                   // object file.
+  ObjSectionToIDMap LocalSections; // Used sections from the object file
+
+  error_code err;
+
+
+  // Parse symbols
+  DEBUG(dbgs() << "Parse symbols:\n");
+  for (symbol_iterator it = obj->begin_symbols(), itEnd = obj->end_symbols();
+       it != itEnd; it.increment(err)) {
+    if (err) break;
+    object::SymbolRef::Type SymType;
+    StringRef Name;
+    if ((bool)(err = it->getType(SymType))) break;
+    if ((bool)(err = it->getName(Name))) break;
+
+    if (SymType == object::SymbolRef::ST_Function ||
+        SymType == object::SymbolRef::ST_Data) {
+      uint64_t FileOffset;
+      uint32_t flags;
+      StringRef sData;
+      section_iterator sIt = obj->end_sections();
+      if ((bool)(err = it->getFileOffset(FileOffset))) break;
+      if ((bool)(err = it->getFlags(flags))) break;
+      if ((bool)(err = it->getSection(sIt))) break;
+      if (sIt == obj->end_sections()) continue;
+      if ((bool)(err = sIt->getContents(sData))) break;
+      const uint8_t* SymPtr = (const uint8_t*)InputBuffer->getBufferStart() +
+                              (uintptr_t)FileOffset;
+      uintptr_t SectOffset = (uintptr_t)(SymPtr - (const uint8_t*)sData.begin());
+      unsigned SectionID =
+        findOrEmitSection(*sIt,
+                          SymType == object::SymbolRef::ST_Function,
+                          LocalSections);
+      bool isGlobal = flags & SymbolRef::SF_Global;
+      LocalSymbols[Name.data()] = SymbolLoc(SectionID, SectOffset);
+      DEBUG(dbgs() << "\tFileOffset: " << format("%p", (uintptr_t)FileOffset)
+                   << " flags: " << flags
+                   << " SID: " << SectionID
+                   << " Offset: " << format("%p", SectOffset));
+      if (isGlobal)
+        SymbolTable[Name] = SymbolLoc(SectionID, SectOffset);
+    }
+    DEBUG(dbgs() << "\tType: " << SymType << " Name: " << Name << "\n");
+  }
+  if (err) {
+    report_fatal_error(err.message());
+  }
+
+  // Parse and proccess relocations
+  DEBUG(dbgs() << "Parse relocations:\n");
+  for (section_iterator sIt = obj->begin_sections(),
+       sItEnd = obj->end_sections(); sIt != sItEnd; sIt.increment(err)) {
+    if (err) break;
+    bool isFirstRelocation = true;
+    unsigned SectionID = 0;
+    StubMap Stubs;
+
+    for (relocation_iterator it = sIt->begin_relocations(),
+         itEnd = sIt->end_relocations(); it != itEnd; it.increment(err)) {
+      if (err) break;
+
+      // If it's first relocation in this section, find its SectionID
+      if (isFirstRelocation) {
+        SectionID = findOrEmitSection(*sIt, true, LocalSections);
+        DEBUG(dbgs() << "\tSectionID: " << SectionID << "\n");
+        isFirstRelocation = false;
+      }
+
+      ObjRelocationInfo RI;
+      RI.SectionID = SectionID;
+      if ((bool)(err = it->getAdditionalInfo(RI.AdditionalInfo))) break;
+      if ((bool)(err = it->getOffset(RI.Offset))) break;
+      if ((bool)(err = it->getSymbol(RI.Symbol))) break;
+      if ((bool)(err = it->getType(RI.Type))) break;
+
+      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);
+    }
+    if (err) {
+      report_fatal_error(err.message());
+    }
+  }
+  return false;
+}
+
+unsigned RuntimeDyldImpl::emitSection(const SectionRef &Section,
+                                      bool IsCode) {
+
+  unsigned StubBufSize = 0,
+           StubSize = getMaxStubSize();
+  error_code err;
+  if (StubSize > 0) {
+    for (relocation_iterator it = Section.begin_relocations(),
+         itEnd = Section.end_relocations(); it != itEnd; it.increment(err))
+      StubBufSize += StubSize;
+  }
+  StringRef data;
+  uint64_t Alignment64;
+  if ((bool)(err = Section.getContents(data))) report_fatal_error(err.message());
+  if ((bool)(err = Section.getAlignment(Alignment64)))
+    report_fatal_error(err.message());
+
+  unsigned Alignment = (unsigned)Alignment64 & 0xffffffffL;
+  unsigned DataSize = data.size();
+  unsigned Allocate = DataSize + StubBufSize;
+  unsigned SectionID = Sections.size();
+  const char *pData = data.data();
+  uint8_t *Addr = IsCode
+    ? MemMgr->allocateCodeSection(Allocate, Alignment, SectionID)
+    : MemMgr->allocateDataSection(Allocate, Alignment, SectionID);
+
+  memcpy(Addr, pData, DataSize);
+  DEBUG(dbgs() << "emitSection SectionID: " << SectionID
+               << " obj addr: " << format("%p", pData)
+               << " new addr: " << format("%p", Addr)
+               << " DataSize: " << DataSize
+               << " StubBufSize: " << StubBufSize
+               << " Allocate: " << Allocate
+               << "\n");
+  Sections.push_back(SectionEntry(Addr, Allocate, DataSize,(uintptr_t)pData));
+  return SectionID;
+}
+
+unsigned RuntimeDyldImpl::
+findOrEmitSection(const SectionRef &Section, bool IsCode,
+                  ObjSectionToIDMap &LocalSections) {
+
+  unsigned SectionID = 0;
+  ObjSectionToIDMap::iterator sIDIt = LocalSections.find(Section);
+  if (sIDIt != LocalSections.end())
+    SectionID = sIDIt->second;
+  else {
+    SectionID = emitSection(Section, IsCode);
+    LocalSections[Section] = SectionID;
+  }
+  return SectionID;
+}
+
+void RuntimeDyldImpl::AddRelocation(const RelocationValueRef &Value,
+                                   unsigned SectionID, uintptr_t Offset,
+                                   uint32_t RelType) {
+  DEBUG(dbgs() << "AddRelocation SymNamePtr: " << format("%p", Value.SymbolName)
+               << " SID: " << Value.SectionID
+               << " Addend: " << format("%p", Value.Addend)
+               << " Offset: " << format("%p", Offset)
+               << " RelType: " << format("%x", RelType)
+               << "\n");
+
+  if (Value.SymbolName == 0) {
+    Relocations[Value.SectionID].push_back(RelocationEntry(
+      SectionID,
+      Offset,
+      RelType,
+      Value.Addend));
+  } else
+    SymbolRelocations[Value.SymbolName].push_back(RelocationEntry(
+      SectionID,
+      Offset,
+      RelType,
+      Value.Addend));
+}
+
+uint8_t *RuntimeDyldImpl::createStubFunction(uint8_t *Addr) {
+  // TODO: There is only ARM far stub now. We should add the Thumb stub,
+  // and stubs for branches Thumb - ARM and ARM - Thumb.
+  if (Arch == Triple::arm) {
+    uint32_t *StubAddr = (uint32_t*)Addr;
+    *StubAddr = 0xe51ff004; // ldr pc,<label>
+    return (uint8_t*)++StubAddr;
+  }
+  else
+    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. Re-apply any
+  // relocations referencing this section with the given address.
+  //
+  // Addr is a uint64_t because we can't assume the pointer width
+  // of the target is the same as that of the host. Just use a generic
+  // "big enough" type.
+  Sections[SectionID].LoadAddress = Addr;
+  DEBUG(dbgs() << "Resolving relocations Section #" << SectionID
+          << "\t" << format("%p", (uint8_t *)Addr)
+          << "\n");
+  resolveRelocationList(Relocations[SectionID], Addr);
+}
+
+void RuntimeDyldImpl::resolveRelocationEntry(const RelocationEntry &RE,
+                                             uint64_t Value) {
+    uint8_t *Target = Sections[RE.SectionID].Address + RE.Offset;
+    DEBUG(dbgs() << "\tSectionID: " << RE.SectionID
+          << " + " << RE.Offset << " (" << format("%p", Target) << ")"
+          << " Data: " << RE.Data
+          << " Addend: " << RE.Addend
+          << "\n");
+
+    resolveRelocation(Target, Sections[RE.SectionID].LoadAddress + RE.Offset,
+                      Value, RE.Data, 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);
+  }
+}
+
+// resolveSymbols - Resolve any relocations to the specified symbols if
+// we know where it lives.
+void RuntimeDyldImpl::resolveSymbols() {
+  StringMap<RelocationList>::iterator it = SymbolRelocations.begin(),
+                                      itEnd = SymbolRelocations.end();
+  for (; it != itEnd; it++) {
+    StringRef Name = it->first();
+    RelocationList &Relocs = it->second;
+    StringMap<SymbolLoc>::const_iterator Loc = SymbolTable.find(Name);
+    if (Loc == SymbolTable.end()) {
+      // This is an external symbol, try to get it address from
+      // MemoryManager.
+      uint8_t *Addr = (uint8_t*) MemMgr->getPointerToNamedFunction(Name.data(),
+                                                                   true);
+      DEBUG(dbgs() << "Resolving relocations Name: " << Name
+              << "\t" << format("%p", Addr)
+              << "\n");
+      resolveRelocationList(Relocs, (uintptr_t)Addr);
+    } else {
+      // Change the relocation to be section relative rather than symbol
+      // relative and move it to the resolved relocation list.
+      DEBUG(dbgs() << "Resolving symbol '" << Name << "'\n");
+      for (int i = 0, e = Relocs.size(); i != e; ++i) {
+        RelocationEntry Entry = Relocs[i];
+        Entry.Addend += Loc->second.second;
+        Relocations[Loc->second.first].push_back(Entry);
+      }
+      Relocs.clear();
+    }
+  }
+}
+
+
 //===----------------------------------------------------------------------===//
 // RuntimeDyld class implementation
 RuntimeDyld::RuntimeDyld(RTDyldMemoryManager *mm) {
diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp
index e15b200c5e..9351b6c361 100644
--- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp
+++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp
@@ -25,222 +25,58 @@ using namespace llvm::object;
 
 namespace llvm {
 
-namespace {
 
-// FIXME: this function should probably not live here...
-//
-// Returns the name and address of an unrelocated symbol in an ELF section
-void getSymbolInfo(symbol_iterator Sym, uint64_t &Addr, StringRef &Name) {
-  //FIXME: error checking here required to catch corrupt ELF objects...
-  error_code Err = Sym->getName(Name);
-
-  uint64_t AddrInSection;
-  Err = Sym->getAddress(AddrInSection);
-
-  SectionRef empty_section;
-  section_iterator Section(empty_section);
-  Err = Sym->getSection(Section);
-
-  StringRef SectionContents;
-  Section->getContents(SectionContents);
-
-  Addr = reinterpret_cast<uint64_t>(SectionContents.data()) + AddrInSection;
-}
-
-}
-
-bool RuntimeDyldELF::loadObject(MemoryBuffer *InputBuffer) {
-  if (!isCompatibleFormat(InputBuffer))
-    return true;
-
-  OwningPtr<ObjectFile> Obj(ObjectFile::createELFObjectFile(InputBuffer));
-
-  Arch = Obj->getArch();
-
-  // Map address in the Object file image to function names
-  IntervalMap<uint64_t, StringRef>::Allocator A;
-  IntervalMap<uint64_t, StringRef> FuncMap(A);
-
-  // This is a bit of a hack.  The ObjectFile we've just loaded reports
-  // section addresses as 0 and doesn't provide access to the section
-  // offset (from which we could calculate the address.  Instead,
-  // we're storing the address when it comes up in the ST_Debug case
-  // below.
-  //
-  StringMap<uint64_t> DebugSymbolMap;
-
-  symbol_iterator SymEnd = Obj->end_symbols();
-  error_code Err;
-  for (symbol_iterator Sym = Obj->begin_symbols();
-       Sym != SymEnd; Sym.increment(Err)) {
-    SymbolRef::Type Type;
-    Sym->getType(Type);
-    if (Type == SymbolRef::ST_Function) {
-      StringRef Name;
-      uint64_t Addr;
-      getSymbolInfo(Sym, Addr, Name);
-
-      uint64_t Size;
-      Err = Sym->getSize(Size);
-
-      uint8_t *Start;
-      uint8_t *End;
-      Start = reinterpret_cast<uint8_t*>(Addr);
-      End   = reinterpret_cast<uint8_t*>(Addr + Size - 1);
-
-      extractFunction(Name, Start, End);
-      FuncMap.insert(Addr, Addr + Size - 1, Name);
-    } else if (Type == SymbolRef::ST_Debug) {
-      // This case helps us find section addresses
-      StringRef Name;
-      uint64_t Addr;
-      getSymbolInfo(Sym, Addr, Name);
-      DebugSymbolMap[Name] = Addr;
-    }
-  }
-
-  // Iterate through the relocations for this object
-  section_iterator SecEnd = Obj->end_sections();
-  for (section_iterator Sec = Obj->begin_sections();
-       Sec != SecEnd; Sec.increment(Err)) {
-    StringRef SecName;
-    uint64_t  SecAddr;
-    Sec->getName(SecName);
-    // Ignore sections that aren't in our map
-    if (DebugSymbolMap.find(SecName) == DebugSymbolMap.end()) {
-      continue;
-    }
-    SecAddr = DebugSymbolMap[SecName];
-    relocation_iterator RelEnd = Sec->end_relocations();
-    for (relocation_iterator Rel = Sec->begin_relocations();
-         Rel != RelEnd; Rel.increment(Err)) {
-      uint64_t RelOffset;
-      uint64_t RelType;
-      int64_t RelAddend;
-      SymbolRef RelSym;
-      StringRef SymName;
-      uint64_t SymAddr;
-      uint64_t SymOffset;
-
-      Rel->getAddress(RelOffset);
-      Rel->getType(RelType);
-      Rel->getAdditionalInfo(RelAddend);
-      Rel->getSymbol(RelSym);
-      RelSym.getName(SymName);
-      RelSym.getAddress(SymAddr);
-      RelSym.getFileOffset(SymOffset);
-
-      // If this relocation is inside a function, we want to store the
-      // function name and a function-relative offset
-      IntervalMap<uint64_t, StringRef>::iterator ContainingFunc
-        = FuncM