[PM] Remove the preverifier and directly compute the DominatorTree for
[oota-llvm.git] / lib / ExecutionEngine / RuntimeDyld / RuntimeDyldMachO.cpp
index 24437e0f96313061d61e26e9db2c24a65f37094a..d12e7e5e4dcf2f6845dad18d27d386064134f4af 100644 (file)
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "dyld"
+#include "RuntimeDyldMachO.h"
 #include "llvm/ADT/OwningPtr.h"
-#include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/STLExtras.h"
-#include "RuntimeDyldMachO.h"
+#include "llvm/ADT/StringRef.h"
 using namespace llvm;
 using namespace llvm::object;
 
 namespace llvm {
 
-void RuntimeDyldMachO::resolveRelocation(uint8_t *LocalAddress,
-                                         uint64_t FinalAddress,
+static unsigned char *processFDE(unsigned char *P, intptr_t DeltaForText, intptr_t DeltaForEH) {
+  uint32_t Length = *((uint32_t*)P);
+  P += 4;
+  unsigned char *Ret = P + Length;
+  uint32_t Offset = *((uint32_t*)P);
+  if (Offset == 0) // is a CIE
+    return Ret;
+
+  P += 4;
+  intptr_t FDELocation = *((intptr_t*)P);
+  intptr_t NewLocation = FDELocation - DeltaForText;
+  *((intptr_t*)P) = NewLocation;
+  P += sizeof(intptr_t);
+
+  // Skip the FDE address range
+  P += sizeof(intptr_t);
+
+  uint8_t Augmentationsize = *P;
+  P += 1;
+  if (Augmentationsize != 0) {
+    intptr_t LSDA = *((intptr_t*)P);
+    intptr_t NewLSDA = LSDA - DeltaForEH;
+    *((intptr_t*)P) = NewLSDA;
+  }
+
+  return Ret;
+}
+
+static intptr_t computeDelta(SectionEntry *A, SectionEntry *B) {
+  intptr_t ObjDistance = A->ObjAddress  - B->ObjAddress;
+  intptr_t MemDistance = A->LoadAddress - B->LoadAddress;
+  return ObjDistance - MemDistance;
+}
+
+void RuntimeDyldMachO::registerEHFrames() {
+
+  if (!MemMgr)
+    return;
+  for (int i = 0, e = UnregisteredEHFrameSections.size(); i != e; ++i) {
+    EHFrameRelatedSections &SectionInfo = UnregisteredEHFrameSections[i];
+    if (SectionInfo.EHFrameSID == RTDYLD_INVALID_SECTION_ID ||
+        SectionInfo.TextSID == RTDYLD_INVALID_SECTION_ID)
+      continue;
+    SectionEntry *Text = &Sections[SectionInfo.TextSID];
+    SectionEntry *EHFrame = &Sections[SectionInfo.EHFrameSID];
+    SectionEntry *ExceptTab = NULL;
+    if (SectionInfo.ExceptTabSID != RTDYLD_INVALID_SECTION_ID)
+      ExceptTab = &Sections[SectionInfo.ExceptTabSID];
+
+    intptr_t DeltaForText = computeDelta(Text, EHFrame);
+    intptr_t DeltaForEH = 0;
+    if (ExceptTab)
+      DeltaForEH = computeDelta(ExceptTab, EHFrame);
+
+    unsigned char *P = EHFrame->Address;
+    unsigned char *End = P + EHFrame->Size;
+    do  {
+      P = processFDE(P, DeltaForText, DeltaForEH);
+    } while(P != End);
+
+    MemMgr->registerEHFrames(EHFrame->Address,
+                             EHFrame->LoadAddress,
+                             EHFrame->Size);
+  }
+  UnregisteredEHFrameSections.clear();
+}
+
+void RuntimeDyldMachO::finalizeLoad(ObjSectionToIDMap &SectionMap) {
+  unsigned EHFrameSID = RTDYLD_INVALID_SECTION_ID;
+  unsigned TextSID = RTDYLD_INVALID_SECTION_ID;
+  unsigned ExceptTabSID = RTDYLD_INVALID_SECTION_ID;
+  ObjSectionToIDMap::iterator i, e;
+  for (i = SectionMap.begin(), e = SectionMap.end(); i != e; ++i) {
+    const SectionRef &Section = i->first;
+    StringRef Name;
+    Section.getName(Name);
+    if (Name == "__eh_frame")
+      EHFrameSID = i->second;
+    else if (Name == "__text")
+      TextSID = i->second;
+    else if (Name == "__gcc_except_tab")
+      ExceptTabSID = i->second;
+  }
+  UnregisteredEHFrameSections.push_back(EHFrameRelatedSections(EHFrameSID,
+                                                               TextSID,
+                                                               ExceptTabSID));
+}
+
+// The target location for the relocation is described by RE.SectionID and
+// RE.Offset.  RE.SectionID can be used to find the SectionEntry.  Each
+// SectionEntry has three members describing its location.
+// SectionEntry::Address is the address at which the section has been loaded
+// into memory in the current (host) process.  SectionEntry::LoadAddress is the
+// address that the section will have in the target process.
+// SectionEntry::ObjAddress is the address of the bits for this section in the
+// original emitted object image (also in the current address space).
+//
+// Relocations will be applied as if the section were loaded at
+// SectionEntry::LoadAddress, but they will be applied at an address based
+// on SectionEntry::Address.  SectionEntry::ObjAddress will be used to refer to
+// Target memory contents if they are required for value calculations.
+//
+// The Value parameter here is the load address of the symbol for the
+// relocation to be applied.  For relocations which refer to symbols in the
+// current object Value will be the LoadAddress of the section in which
+// the symbol resides (RE.Addend provides additional information about the
+// symbol location).  For external symbols, Value will be the address of the
+// symbol in the target address space.
+void RuntimeDyldMachO::resolveRelocation(const RelocationEntry &RE,
+                                         uint64_t Value) {
+  const SectionEntry &Section = Sections[RE.SectionID];
+  return resolveRelocation(Section, RE.Offset, Value, RE.RelType, RE.Addend,
+                           RE.IsPCRel, RE.Size);
+}
+
+void RuntimeDyldMachO::resolveRelocation(const SectionEntry &Section,
+                                         uint64_t Offset,
                                          uint64_t Value,
                                          uint32_t Type,
-                                         int64_t Addend) {
-  bool isPCRel = (Type >> 24) & 1;
-  unsigned MachoType = (Type >> 28) & 0xf;
-  unsigned Size = 1 << ((Type >> 25) & 3);
+                                         int64_t Addend,
+                                         bool isPCRel,
+                                         unsigned LogSize) {
+  uint8_t *LocalAddress = Section.Address + Offset;
+  uint64_t FinalAddress = Section.LoadAddress + Offset;
+  unsigned MachoType = Type;
+  unsigned Size = 1 << LogSize;
 
-  DEBUG(dbgs() << "resolveRelocation LocalAddress: " << format("%p", LocalAddress)
+  DEBUG(dbgs() << "resolveRelocation LocalAddress: "
+        << format("%p", LocalAddress)
         << " FinalAddress: " << format("%p", FinalAddress)
         << " Value: " << format("%p", Value)
         << " Addend: " << Addend
@@ -42,8 +161,7 @@ void RuntimeDyldMachO::resolveRelocation(uint8_t *LocalAddress,
   // This just dispatches to the proper target specific routine.
   switch (Arch) {
   default: llvm_unreachable("Unsupported CPU type!");
-  case Triple::x86_64: // Fall through.
-  case Triple::x86:
+  case Triple::x86_64:
     resolveX86_64Relocation(LocalAddress,
                             FinalAddress,
                             (uintptr_t)Value,
@@ -52,6 +170,15 @@ void RuntimeDyldMachO::resolveRelocation(uint8_t *LocalAddress,
                             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,
@@ -65,14 +192,42 @@ void RuntimeDyldMachO::resolveRelocation(uint8_t *LocalAddress,
   }
 }
 
-bool RuntimeDyldMachO::
-resolveX86_64Relocation(uint8_t *LocalAddress,
-                        uint64_t FinalAddress,
-                        uint64_t Value,
-                        bool isPCRel,
-                        unsigned Type,
-                        unsigned Size,
-                        int64_t Addend) {
+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::GENERIC_RELOC_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::GENERIC_RELOC_SECTDIFF:
+  case MachO::GENERIC_RELOC_LOCAL_SECTDIFF:
+  case MachO::GENERIC_RELOC_PB_LA_PTR:
+    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)
@@ -83,12 +238,12 @@ resolveX86_64Relocation(uint8_t *LocalAddress,
   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: {
+  case MachO::X86_64_RELOC_SIGNED_1:
+  case MachO::X86_64_RELOC_SIGNED_2:
+  case MachO::X86_64_RELOC_SIGNED_4:
+  case MachO::X86_64_RELOC_SIGNED:
+  case MachO::X86_64_RELOC_UNSIGNED:
+  case MachO::X86_64_RELOC_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).
@@ -99,22 +254,21 @@ resolveX86_64Relocation(uint8_t *LocalAddress,
     }
     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:
+  case MachO::X86_64_RELOC_GOT_LOAD:
+  case MachO::X86_64_RELOC_GOT:
+  case MachO::X86_64_RELOC_SUBTRACTOR:
+  case MachO::X86_64_RELOC_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) {
+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) {
@@ -128,7 +282,7 @@ resolveARMRelocation(uint8_t *LocalAddress,
   switch(Type) {
   default:
     llvm_unreachable("Invalid relocation type!");
-  case macho::RIT_Vanilla: {
+  case MachO::ARM_RELOC_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;
@@ -138,7 +292,7 @@ resolveARMRelocation(uint8_t *LocalAddress,
     }
     break;
   }
-  case macho::RIT_ARM_Branch24Bit: {
+  case MachO::ARM_RELOC_BR24: {
     // 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;
@@ -154,97 +308,146 @@ resolveARMRelocation(uint8_t *LocalAddress,
     *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:
+  case MachO::ARM_THUMB_RELOC_BR22:
+  case MachO::ARM_THUMB_32BIT_BRANCH:
+  case MachO::ARM_RELOC_HALF:
+  case MachO::ARM_RELOC_HALF_SECTDIFF:
+  case MachO::ARM_RELOC_PAIR:
+  case MachO::ARM_RELOC_SECTDIFF:
+  case MachO::ARM_RELOC_LOCAL_SECTDIFF:
+  case MachO::ARM_RELOC_PB_LA_PTR:
     return Error("Relocation type not implemented yet!");
   }
   return false;
 }
 
-void RuntimeDyldMachO::
-processRelocationRef(const ObjRelocationInfo &Rel, const ObjectFile &Obj,
-                     ObjSectionToIDMap &ObjSectionToID,
-                     LocalSymbolMap &Symbols, StubMap &Stubs) {
+void RuntimeDyldMachO::processRelocationRef(unsigned SectionID,
+                                            RelocationRef RelI,
+                                            ObjectImage &Obj,
+                                            ObjSectionToIDMap &ObjSectionToID,
+                                            const SymbolTableMap &Symbols,
+                                            StubMap &Stubs) {
+  const ObjectFile *OF = Obj.getObjectFile();
+  const MachOObjectFile *MachO = static_cast<const MachOObjectFile*>(OF);
+  MachO::any_relocation_info RE= MachO->getRelocation(RelI.getRawDataRefImpl());
+
+  uint32_t RelType = MachO->getAnyRelocationType(RE);
+
+  // FIXME: Properly handle scattered relocations.
+  //        For now, optimistically skip these: they can often be ignored, as
+  //        the static linker will already have applied the relocation, and it
+  //        only needs to be reapplied if symbols move relative to one another.
+  //        Note: This will fail horribly where the relocations *do* need to be
+  //        applied, but that was already the case.
+  if (MachO->isRelocationScattered(RE))
+    return;
 
-  uint32_t RelType = (uint32_t) (Rel.Type & 0xffffffffL);
   RelocationValueRef Value;
-  SectionEntry &Section = Sections[Rel.SectionID];
-  uint8_t *Target = Section.Address + Rel.Offset;
+  SectionEntry &Section = Sections[SectionID];
+
+  bool isExtern = MachO->getPlainRelocationExternal(RE);
+  bool IsPCRel = MachO->getAnyRelocationPCRel(RE);
+  unsigned Size = MachO->getAnyRelocationLength(RE);
+  uint64_t Offset;
+  RelI.getOffset(Offset);
+  uint8_t *LocalAddress = Section.Address + Offset;
+  unsigned NumBytes = 1 << Size;
+  uint64_t Addend = 0;
+  memcpy(&Addend, LocalAddress, NumBytes);
 
-  bool isExtern = (RelType >> 27) & 1;
   if (isExtern) {
+    // Obtain the symbol name which is referenced in the relocation
+    symbol_iterator Symbol = RelI.getSymbol();
     StringRef TargetName;
-    const SymbolRef &Symbol = Rel.Symbol;
-    Symbol.getName(TargetName);
-    // First look the symbol in object file symbols.
-    LocalSymbolMap::iterator it = Symbols.find(TargetName.data());
-    if (it != Symbols.end()) {
-      Value.SectionID = it->second.first;
-      Value.Addend = it->second.second;
+    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 + Addend;
     } else {
-      // Second look the symbol in global symbol table.
-      StringMap<SymbolLoc>::iterator itS = SymbolTable.find(TargetName.data());
-      if (itS != SymbolTable.end()) {
-        Value.SectionID = itS->second.first;
-        Value.Addend = itS->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 + Addend;
+      } else {
         Value.SymbolName = TargetName.data();
+        Value.Addend = Addend;
+      }
     }
   } else {
-    error_code err;
-    uint8_t sIdx = static_cast<uint8_t>(RelType & 0xFF);
-    section_iterator sIt = Obj.begin_sections(),
-                     sItEnd = Obj.end_sections();
-    for (uint8_t i = 1; i < sIdx; i++) {
-      error_code err;
-      sIt.increment(err);
-      if (sIt == sItEnd)
-        break;
-    }
-    assert(sIt != sItEnd && "No section containing relocation!");
-    Value.SectionID = findOrEmitSection(*sIt, true, ObjSectionToID);
-    Value.Addend = *(const intptr_t *)Target;
-    if (Value.Addend) {
-      // The MachO addend is offset from the current section, we need set it
-      // as offset from destination section
-      Value.Addend += Section.ObjAddress - Sections[Value.SectionID].ObjAddress;
-    }
+    SectionRef Sec = MachO->getRelocationSection(RE);
+    Value.SectionID = findOrEmitSection(Obj, Sec, true, ObjSectionToID);
+    uint64_t Addr;
+    Sec.getAddress(Addr);
+    Value.Addend = Addend - Addr;
   }
 
-  if (Arch == Triple::arm && RelType == macho::RIT_ARM_Branch24Bit) {
+  if (Arch == Triple::x86_64 && (RelType == MachO::X86_64_RELOC_GOT ||
+                                 RelType == MachO::X86_64_RELOC_GOT_LOAD)) {
+    assert(IsPCRel);
+    assert(Size == 2);
+    StubMap::const_iterator i = Stubs.find(Value);
+    uint8_t *Addr;
+    if (i != Stubs.end()) {
+      Addr = Section.Address + i->second;
+    } else {
+      Stubs[Value] = Section.StubOffset;
+      uint8_t *GOTEntry = Section.Address + Section.StubOffset;
+      RelocationEntry RE(SectionID, Section.StubOffset,
+                         MachO::X86_64_RELOC_UNSIGNED, 0, false, 3);
+      if (Value.SymbolName)
+        addRelocationForSymbol(RE, Value.SymbolName);
+      else
+        addRelocationForSection(RE, Value.SectionID);
+      Section.StubOffset += 8;
+      Addr = GOTEntry;
+    }
+    resolveRelocation(Section, Offset, (uint64_t)Addr,
+                      MachO::X86_64_RELOC_UNSIGNED, Value.Addend, true, 2);
+  } else if (Arch == Triple::arm &&
+             (RelType & 0xf) == MachO::ARM_RELOC_BR24) {
     // This is an ARM branch relocation, need to use a stub function.
 
     //  Look up for existing stub.
-    StubMap::const_iterator stubIt = Stubs.find(Value);
-    if (stubIt != Stubs.end())
-      resolveRelocation(Target, (uint64_t)Target,
-                        (uint64_t)Section.Address + stubIt->second,
-                        RelType, 0);
+    StubMap::const_iterator i = Stubs.find(Value);
+    if (i != Stubs.end())
+      resolveRelocation(Section, Offset,
+                        (uint64_t)Section.Address + i->second,
+                        RelType, 0, IsPCRel, Size);
     else {
       // Create a new stub function.
       Stubs[Value] = Section.StubOffset;
       uint8_t *StubTargetAddr = createStubFunction(Section.Address +
                                                    Section.StubOffset);
-      AddRelocation(Value, Rel.SectionID, StubTargetAddr - Section.Address,
-                    macho::RIT_Vanilla);
-      resolveRelocation(Target, (uint64_t)Target,
+      RelocationEntry RE(SectionID, StubTargetAddr - Section.Address,
+                         MachO::GENERIC_RELOC_VANILLA, Value.Addend);
+      if (Value.SymbolName)
+        addRelocationForSymbol(RE, Value.SymbolName);
+      else
+        addRelocationForSection(RE, Value.SectionID);
+      resolveRelocation(Section, Offset,
                         (uint64_t)Section.Address + Section.StubOffset,
-                        RelType, 0);
+                        RelType, 0, IsPCRel, Size);
       Section.StubOffset += getMaxStubSize();
     }
-  } else
-    AddRelocation(Value, Rel.SectionID, Rel.Offset, RelType);
+  } else {
+    RelocationEntry RE(SectionID, Offset, RelType, Value.Addend,
+                       IsPCRel, Size);
+    if (Value.SymbolName)
+      addRelocationForSymbol(RE, Value.SymbolName);
+    else
+      addRelocationForSection(RE, Value.SectionID);
+  }
 }
 
 
-bool RuntimeDyldMachO::isCompatibleFormat(const MemoryBuffer *InputBuffer) const {
-  StringRef Magic = InputBuffer->getBuffer().slice(0, 4);
+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;
@@ -252,4 +455,9 @@ bool RuntimeDyldMachO::isCompatibleFormat(const MemoryBuffer *InputBuffer) const
   return false;
 }
 
+bool RuntimeDyldMachO::isCompatibleFile(
+        const object::ObjectFile *Obj) const {
+  return Obj->isMachO();
+}
+
 } // end namespace llvm