Teach the MC to output code/data region marker labels in MachO and ELF modes. These are used by disassemblers to provide better disassembly, particularly on targets like ARM Thumb that like to intermingle data in the TEXT segment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141135 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/MC/MCAsmInfo.h

@@ -177,6 +177,18 @@ namespace llvm {
     const char *Data32bitsDirective;         // Defaults to "\t.long\t"
     const char *Data64bitsDirective;         // Defaults to "\t.quad\t"
 
+    /// [Data|Code]Begin - These magic labels are used to marked a region as
+    /// data or code, and are used to provide additional information for
+    /// correct disassembly on targets that like to mix data and code within
+    /// a segment.  These labels will be implicitly suffixed by the streamer
+    /// to give them unique names.
+    const char *DataBegin;                   // Defaults to "$d."
+    const char *CodeBegin;                   // Defaults to "$a."
+    const char *JT8Begin;                    // Defaults to "$a."
+    const char *JT16Begin;                   // Defaults to "$a."
+    const char *JT32Begin;                   // Defaults to "$a."
+    bool SupportsDataRegions;
+
     /// GPRel32Directive - if non-null, a directive that is used to emit a word
     /// which should be relocated as a 32-bit GP-relative offset, e.g. .gpword
     /// on Mips or .gprel32 on Alpha.
@@ -371,6 +383,14 @@ namespace llvm {
     }
     const char *getGPRel32Directive() const { return GPRel32Directive; }
 
+    /// [Code|Data]Begin label name accessors.
+    const char *getCodeBeginLabelName() const { return CodeBegin; }
+    const char *getDataBeginLabelName() const { return DataBegin; }
+    const char *getJumpTable8BeginLabelName() const { return JT8Begin; }
+    const char *getJumpTable16BeginLabelName() const { return JT16Begin; }
+    const char *getJumpTable32BeginLabelName() const { return JT32Begin; }
+    bool getSupportsDataRegions() const { return SupportsDataRegions; }
+
     /// getNonexecutableStackSection - Targets can implement this method to
     /// specify a section to switch to if the translation unit doesn't have any
     /// trampolines that require an executable stack.

include/llvm/MC/MCStreamer.h

@@ -71,7 +71,22 @@ namespace llvm {
     SmallVector<std::pair<const MCSection *,
                 const MCSection *>, 4> SectionStack;
 
+    unsigned UniqueCodeBeginSuffix;
+    unsigned UniqueDataBeginSuffix;
+
   protected:
+    /// Indicator of whether the previous data-or-code indicator was for
+    /// code or not.  Used to determine when we need to emit a new indicator.
+    enum DataType {
+      Data,
+      Code,
+      JumpTable8,
+      JumpTable16,
+      JumpTable32
+    };
+    DataType RegionIndicator;
+
+
     MCStreamer(MCContext &Ctx);
 
     const MCExpr *BuildSymbolDiff(MCContext &Context, const MCSymbol *A,
@@ -224,6 +239,41 @@ namespace llvm {
     /// used in an assignment.
     virtual void EmitLabel(MCSymbol *Symbol);
 
+    /// EmitDataRegion - Emit a label that marks the beginning of a data
+    /// region.
+    /// On ELF targets, this corresponds to an assembler statement such as:
+    ///   $d.1:
+    virtual void EmitDataRegion();
+
+    /// EmitJumpTable8Region - Emit a label that marks the beginning of a
+    /// jump table composed of 8-bit offsets.
+    /// On ELF targets, this corresponds to an assembler statement such as:
+    ///   $d.1:
+    virtual void EmitJumpTable8Region();
+
+    /// EmitJumpTable16Region - Emit a label that marks the beginning of a
+    /// jump table composed of 16-bit offsets.
+    /// On ELF targets, this corresponds to an assembler statement such as:
+    ///   $d.1:
+    virtual void EmitJumpTable16Region();
+
+    /// EmitJumpTable32Region - Emit a label that marks the beginning of a
+    /// jump table composed of 32-bit offsets.
+    /// On ELF targets, this corresponds to an assembler statement such as:
+    ///   $d.1:
+    virtual void EmitJumpTable32Region();
+
+    /// EmitCodeRegion - Emit a label that marks the beginning of a code
+    /// region.
+    /// On ELF targets, this corresponds to an assembler statement such as:
+    ///   $a.1:
+    virtual void EmitCodeRegion();
+
+    /// ForceCodeRegion - Forcibly sets the current region mode to code.  Used
+    /// at function entry points.
+    void ForceCodeRegion() { RegionIndicator = Code; }
+
+
     virtual void EmitEHSymAttributes(const MCSymbol *Symbol,
                                      MCSymbol *EHSymbol);
 

lib/CodeGen/AsmPrinter/AsmPrinter.cpp

@@ -474,8 +474,10 @@ void AsmPrinter::EmitFunctionHeader() {
 void AsmPrinter::EmitFunctionEntryLabel() {
   // The function label could have already been emitted if two symbols end up
   // conflicting due to asm renaming.  Detect this and emit an error.
-  if (CurrentFnSym->isUndefined())
+  if (CurrentFnSym->isUndefined()) {
+    OutStreamer.ForceCodeRegion();
     return OutStreamer.EmitLabel(CurrentFnSym);
+  }
 
   report_fatal_error("'" + Twine(CurrentFnSym->getName()) +
                      "' label emitted multiple times to assembly file");
@@ -1058,6 +1060,15 @@ void AsmPrinter::EmitJumpTableInfo() {
 
   EmitAlignment(Log2_32(MJTI->getEntryAlignment(*TM.getTargetData())));
 
+  // If we know the form of the jump table, go ahead and tag it as such.
+  if (!JTInDiffSection) {
+    if (MJTI->getEntryKind() == MachineJumpTableInfo::EK_LabelDifference32) {
+      OutStreamer.EmitJumpTable32Region();
+    } else {
+      OutStreamer.EmitDataRegion();
+    }
+  }
+
   for (unsigned JTI = 0, e = JT.size(); JTI != e; ++JTI) {
     const std::vector<MachineBasicBlock*> &JTBBs = JT[JTI].MBBs;
 

lib/MC/MCAsmInfo.cpp

@@ -57,6 +57,12 @@ MCAsmInfo::MCAsmInfo() {
   Data16bitsDirective = "\t.short\t";
   Data32bitsDirective = "\t.long\t";
   Data64bitsDirective = "\t.quad\t";
+  DataBegin = "$d.";
+  CodeBegin = "$a.";
+  JT8Begin = "$d.";
+  JT16Begin = "$d.";
+  JT32Begin = "$d.";
+  SupportsDataRegions = true;
   SunStyleELFSectionSwitchSyntax = false;
   UsesELFSectionDirectiveForBSS = false;
   AlignDirective = "\t.align\t";

lib/MC/MCAsmInfoCOFF.cpp

@@ -35,4 +35,6 @@ MCAsmInfoCOFF::MCAsmInfoCOFF() {
   SupportsDebugInformation = true;
   DwarfSectionOffsetDirective = "\t.secrel32\t";
   HasMicrosoftFastStdCallMangling = true;
+
+  SupportsDataRegions = false;
 }

lib/MC/MCAsmInfoDarwin.cpp

@@ -42,6 +42,13 @@ MCAsmInfoDarwin::MCAsmInfoDarwin() {
   StructorOutputOrder = Structors::PriorityOrder;
   HasStaticCtorDtorReferenceInStaticMode = true;
 
+  CodeBegin = "L$start$code$";
+  DataBegin = "L$start$data$";
+  JT8Begin  = "L$start$jt8$";
+  JT16Begin = "L$start$jt16$";
+  JT32Begin = "L$start$jt32$";
+  SupportsDataRegions = true;
+
   // FIXME: Darwin 10 and newer don't need this.
   LinkerRequiresNonEmptyDwarfLines = true;
 

lib/MC/MCStreamer.cpp

@@ -16,6 +16,7 @@
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/Twine.h"
 #include <cstdlib>
 using namespace llvm;
@@ -23,7 +24,9 @@ using namespace llvm;
 MCStreamer::MCStreamer(MCContext &Ctx) : Context(Ctx), EmitEHFrame(true),
                                          EmitDebugFrame(false),
                                          CurrentW64UnwindInfo(0),
-                                         LastSymbol(0) {
+                                         LastSymbol(0),
+                                         UniqueCodeBeginSuffix(0),
+                                         UniqueDataBeginSuffix(0) {
   const MCSection *section = NULL;
   SectionStack.push_back(std::make_pair(section, section));
 }
@@ -175,6 +178,87 @@ void MCStreamer::EmitLabel(MCSymbol *Symbol) {
   LastSymbol = Symbol;
 }
 
+void MCStreamer::EmitDataRegion() {
+  if (RegionIndicator == Data) return;
+
+  MCContext &Context = getContext();
+  const MCAsmInfo &MAI = Context.getAsmInfo();
+  if (!MAI.getSupportsDataRegions()) return;
+
+  // Generate a unique symbol name.
+  MCSymbol *NewSym = Context.GetOrCreateSymbol(
+      Twine(MAI.getDataBeginLabelName()) +
+        utostr(UniqueDataBeginSuffix++));
+  EmitLabel(NewSym);
+
+  RegionIndicator = Data;
+}
+
+void MCStreamer::EmitCodeRegion() {
+  if (RegionIndicator == Code) return;
+
+  MCContext &Context = getContext();
+  const MCAsmInfo &MAI = Context.getAsmInfo();
+  if (!MAI.getSupportsDataRegions()) return;
+
+  // Generate a unique symbol name.
+  MCSymbol *NewSym = Context.GetOrCreateSymbol(
+      Twine(MAI.getCodeBeginLabelName()) +
+        utostr(UniqueCodeBeginSuffix++));
+  EmitLabel(NewSym);
+
+  RegionIndicator = Code;
+}
+
+void MCStreamer::EmitJumpTable8Region() {
+  if (RegionIndicator == JumpTable8) return;
+
+  MCContext &Context = getContext();
+  const MCAsmInfo &MAI = Context.getAsmInfo();
+  if (!MAI.getSupportsDataRegions()) return;
+
+  // Generate a unique symbol name.
+  MCSymbol *NewSym = Context.GetOrCreateSymbol(
+      Twine(MAI.getJumpTable8BeginLabelName()) +
+        utostr(UniqueDataBeginSuffix++));
+  EmitLabel(NewSym);
+
+  RegionIndicator = JumpTable8;
+}
+
+void MCStreamer::EmitJumpTable16Region() {
+  if (RegionIndicator == JumpTable16) return;
+
+  MCContext &Context = getContext();
+  const MCAsmInfo &MAI = Context.getAsmInfo();
+  if (!MAI.getSupportsDataRegions()) return;
+
+  // Generate a unique symbol name.
+  MCSymbol *NewSym = Context.GetOrCreateSymbol(
+      Twine(MAI.getJumpTable16BeginLabelName()) +
+        utostr(UniqueDataBeginSuffix++));
+  EmitLabel(NewSym);
+
+  RegionIndicator = JumpTable16;
+}
+
+
+void MCStreamer::EmitJumpTable32Region() {
+  if (RegionIndicator == JumpTable32) return;
+
+  MCContext &Context = getContext();
+  const MCAsmInfo &MAI = Context.getAsmInfo();
+  if (!MAI.getSupportsDataRegions()) return;
+
+  // Generate a unique symbol name.
+  MCSymbol *NewSym = Context.GetOrCreateSymbol(
+      Twine(MAI.getJumpTable32BeginLabelName()) +
+        utostr(UniqueDataBeginSuffix++));
+  EmitLabel(NewSym);
+
+  RegionIndicator = JumpTable32;
+}
+
 void MCStreamer::EmitCompactUnwindEncoding(uint32_t CompactUnwindEncoding) {
   EnsureValidFrame();
   MCDwarfFrameInfo *CurFrame = getCurrentFrameInfo();
@@ -206,6 +290,7 @@ void MCStreamer::EmitCFIStartProc() {
   }
 
   FrameInfos.push_back(Frame);
+  RegionIndicator = Code;
 }
 
 void MCStreamer::EmitCFIEndProc() {

lib/Target/ARM/ARMAsmPrinter.cpp

@@ -290,6 +290,8 @@ void ARMAsmPrinter::EmitDwarfRegOp(const MachineLocation &MLoc) const {
 }
 
 void ARMAsmPrinter::EmitFunctionEntryLabel() {
+  OutStreamer.ForceCodeRegion();
+
   if (AFI->isThumbFunction()) {
     OutStreamer.EmitAssemblerFlag(MCAF_Code16);
     OutStreamer.EmitThumbFunc(CurrentFnSym);
@@ -905,6 +907,9 @@ void ARMAsmPrinter::EmitJumpTable(const MachineInstr *MI) {
   const MachineOperand &MO2 = MI->getOperand(OpNum+1); // Unique Id
   unsigned JTI = MO1.getIndex();
 
+  // Tag the jump table appropriately for precise disassembly.
+  OutStreamer.EmitJumpTable32Region();
+
   // Emit a label for the jump table.
   MCSymbol *JTISymbol = GetARMJTIPICJumpTableLabel2(JTI, MO2.getImm());
   OutStreamer.EmitLabel(JTISymbol);
@@ -947,6 +952,14 @@ void ARMAsmPrinter::EmitJump2Table(const MachineInstr *MI) {
   unsigned JTI = MO1.getIndex();
 
   // Emit a label for the jump table.
+  if (MI->getOpcode() == ARM::t2TBB_JT) {
+    OutStreamer.EmitJumpTable8Region();
+  } else if (MI->getOpcode() == ARM::t2TBH_JT) {
+    OutStreamer.EmitJumpTable16Region();
+  } else {
+    OutStreamer.EmitJumpTable32Region();
+  }
+
   MCSymbol *JTISymbol = GetARMJTIPICJumpTableLabel2(JTI, MO2.getImm());
   OutStreamer.EmitLabel(JTISymbol);
 
@@ -1165,6 +1178,9 @@ extern cl::opt<bool> EnableARMEHABI;
 #include "ARMGenMCPseudoLowering.inc"
 
 void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
+  if (MI->getOpcode() != ARM::CONSTPOOL_ENTRY)
+    OutStreamer.EmitCodeRegion();
+
   // Emit unwinding stuff for frame-related instructions
   if (EnableARMEHABI && MI->getFlag(MachineInstr::FrameSetup))
     EmitUnwindingInstruction(MI);
@@ -1470,6 +1486,10 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     unsigned CPIdx   = (unsigned)MI->getOperand(1).getIndex();
 
     EmitAlignment(2);
+
+    // Mark the constant pool entry as data if we're not already in a data
+    // region.
+    OutStreamer.EmitDataRegion();
     OutStreamer.EmitLabel(GetCPISymbol(LabelId));
 
     const MachineConstantPoolEntry &MCPE = MCP->getConstants()[CPIdx];
@@ -1477,7 +1497,6 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
       EmitMachineConstantPoolValue(MCPE.Val.MachineCPVal);
     else
       EmitGlobalConstant(MCPE.Val.ConstVal);
-
     return;
   }
   case ARM::t2BR_JT: {

lib/Target/X86/X86MCInstLower.cpp

@@ -592,6 +592,8 @@ static void LowerTlsAddr(MCStreamer &OutStreamer,
 }
 
 void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
+  OutStreamer.EmitCodeRegion();
+
   X86MCInstLower MCInstLowering(Mang, *MF, *this);
   switch (MI->getOpcode()) {
   case TargetOpcode::DBG_VALUE:
@@ -608,7 +610,7 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
     if (OutStreamer.hasRawTextSupport())
       OutStreamer.EmitRawText(StringRef("\t#MEMBARRIER"));
     return;
-        
+
 
   case X86::EH_RETURN:
   case X86::EH_RETURN64: {

test/MC/ARM/elf-reloc-01.ll

@@ -62,9 +62,9 @@ declare void @exit(i32) noreturn nounwind
 
 ;; OBJ:          Relocation 1
 ;; OBJ-NEXT:     'r_offset', 
-;; OBJ-NEXT:     'r_sym', 0x000002
+;; OBJ-NEXT:     'r_sym', 0x000003
 ;; OBJ-NEXT:     'r_type', 0x2b
 
-;; OBJ:         Symbol 2
+;; OBJ:         Symbol 3
 ;; OBJ-NEXT:    '_MergedGlobals'
 ;; OBJ-NEXT:    'st_value', 0x00000010