[SFrames] reland Emit and relax FREs #158154 (#159643)

[Previously reverted due to msan failures on two uninitialized padding
bits.]

This PR emits and relaxes the FREs generated in the previous PR.

After this change llvm emits usable sframe sections that can be linked
with the gnu linker. There are a few remaining cfi directives to handle
before they are generally usable, however.

The output isn't identical with gnu-gas in every case (this code
produces fewer identical FREs in a row than gas), but I'm reasonably
sure that they are correct regardless. There are even more size
optimizations that can be done later.

Also, while working on the tests, I found a few bugs in older portions
and cleaned those up.

This is a fairly big commit, but I'm not sure how to make it smaller.

Author: Sterling-Augustine

llvm/include/llvm/BinaryFormat/SFrame.h

@@ -117,6 +117,7 @@ template <endianness E> struct FDEInfo {
     Info = ((PAuthKey & 1) << 5) | ((static_cast<uint8_t>(FDE) & 1) << 4) |
            (static_cast<uint8_t>(FRE) & 0xf);
   }
+  uint8_t getFuncInfo() const { return Info; }
 };
 
 template <endianness E> struct FuncDescEntry {
@@ -155,6 +156,7 @@ template <endianness E> struct FREInfo {
     Info = ((RA & 1) << 7) | ((static_cast<uint8_t>(Sz) & 3) << 5) |
            ((N & 0xf) << 1) | (static_cast<uint8_t>(Reg) & 1);
   }
+  uint8_t getFREInfo() const { return Info; }
 };
 
 template <typename T, endianness E> struct FrameRowEntry {

llvm/include/llvm/MC/MCAsmBackend.h

@@ -168,6 +168,7 @@ class LLVM_ABI MCAsmBackend {
   virtual bool relaxAlign(MCFragment &F, unsigned &Size) { return false; }
   virtual bool relaxDwarfLineAddr(MCFragment &) const { return false; }
   virtual bool relaxDwarfCFA(MCFragment &) const { return false; }
+  virtual bool relaxSFrameCFA(MCFragment &) const { return false; }
 
   // Defined by linker relaxation targets to possibly emit LEB128 relocations
   // and set Value at the relocated location.

llvm/include/llvm/MC/MCAssembler.h

@@ -117,6 +117,7 @@ class MCAssembler {
   void relaxBoundaryAlign(MCBoundaryAlignFragment &BF);
   void relaxDwarfLineAddr(MCFragment &F);
   void relaxDwarfCallFrameFragment(MCFragment &F);
+  void relaxSFrameFragment(MCFragment &DF);
 
 public:
   /// Construct a new assembler instance.

llvm/include/llvm/MC/MCObjectStreamer.h

@@ -150,6 +150,9 @@ class MCObjectStreamer : public MCStreamer {
                              MCSymbol *EndLabel = nullptr) override;
   void emitDwarfAdvanceFrameAddr(const MCSymbol *LastLabel,
                                  const MCSymbol *Label, SMLoc Loc);
+  void emitSFrameCalculateFuncOffset(const MCSymbol *FunCabsel,
+                                     const MCSymbol *FREBegin,
+                                     MCFragment *FDEFrag, SMLoc Loc);
   void emitCVLocDirective(unsigned FunctionId, unsigned FileNo, unsigned Line,
                           unsigned Column, bool PrologueEnd, bool IsStmt,
                           StringRef FileName, SMLoc Loc) override;

llvm/include/llvm/MC/MCSFrame.h

@@ -16,16 +16,30 @@
 #ifndef LLVM_MC_MCSFRAME_H
 #define LLVM_MC_MCSFRAME_H
 
+#include "llvm/ADT/SmallVector.h"
+#include <cstdint>
+
 namespace llvm {
 
+class MCContext;
 class MCObjectStreamer;
+class MCFragment;
 
 class MCSFrameEmitter {
 public:
   // Emit the sframe section.
   //
   // \param Streamer - Emit into this stream.
   static void emit(MCObjectStreamer &Streamer);
+
+  // Encode the FRE's function offset.
+  //
+  // \param C - Context.
+  // \param Offset - Offset to encode.
+  // \param Out - Destination of the encoding.
+  // \param FDEFrag - Frag that specifies the encoding format.
+  static void encodeFuncOffset(MCContext &C, uint64_t Offset,
+                               SmallVectorImpl<char> &Out, MCFragment *FDEFrag);
 };
 
 } // end namespace llvm

llvm/include/llvm/MC/MCSection.h

@@ -59,6 +59,7 @@ class MCFragment {
     FT_Org,
     FT_Dwarf,
     FT_DwarfFrame,
+    FT_SFrame,
     FT_BoundaryAlign,
     FT_SymbolId,
     FT_CVInlineLines,
@@ -143,6 +144,12 @@ class MCFragment {
       // .loc dwarf directives.
       int64_t LineDelta;
     } dwarf;
+    struct {
+      // This FRE describes unwind info at AddrDelta from function start.
+      const MCExpr *AddrDelta;
+      // Fragment that records how many bytes of AddrDelta to emit.
+      MCFragment *FDEFragment;
+    } sframe;
   } u{};
 
 public:
@@ -296,6 +303,24 @@ class MCFragment {
     assert(Kind == FT_Dwarf);
     u.dwarf.LineDelta = LineDelta;
   }
+
+  //== FT_SFrame functions
+  const MCExpr &getSFrameAddrDelta() const {
+    assert(Kind == FT_SFrame);
+    return *u.sframe.AddrDelta;
+  }
+  void setSFrameAddrDelta(const MCExpr *E) {
+    assert(Kind == FT_SFrame);
+    u.sframe.AddrDelta = E;
+  }
+  MCFragment *getSFrameFDE() const {
+    assert(Kind == FT_SFrame);
+    return u.sframe.FDEFragment;
+  }
+  void setSFrameFDE(MCFragment *F) {
+    assert(Kind == FT_SFrame);
+    u.sframe.FDEFragment = F;
+  }
 };
 
 // MCFragment subclasses do not use the fixed-size part or variable-size tail of

llvm/lib/MC/MCAssembler.cpp

@@ -22,6 +22,7 @@
 #include "llvm/MC/MCFixup.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCObjectWriter.h"
+#include "llvm/MC/MCSFrame.h"
 #include "llvm/MC/MCSection.h"
 #include "llvm/MC/MCSymbol.h"
 #include "llvm/MC/MCValue.h"
@@ -199,6 +200,7 @@ uint64_t MCAssembler::computeFragmentSize(const MCFragment &F) const {
   case MCFragment::FT_LEB:
   case MCFragment::FT_Dwarf:
   case MCFragment::FT_DwarfFrame:
+  case MCFragment::FT_SFrame:
   case MCFragment::FT_CVInlineLines:
   case MCFragment::FT_CVDefRange:
     return F.getSize();
@@ -399,6 +401,7 @@ static void writeFragment(raw_ostream &OS, const MCAssembler &Asm,
   case MCFragment::FT_LEB:
   case MCFragment::FT_Dwarf:
   case MCFragment::FT_DwarfFrame:
+  case MCFragment::FT_SFrame:
   case MCFragment::FT_CVInlineLines:
   case MCFragment::FT_CVDefRange: {
     if (F.getKind() == MCFragment::FT_Data)
@@ -914,6 +917,24 @@ void MCAssembler::relaxDwarfCallFrameFragment(MCFragment &F) {
   F.clearVarFixups();
 }
 
+void MCAssembler::relaxSFrameFragment(MCFragment &F) {
+  assert(F.getKind() == MCFragment::FT_SFrame);
+  MCContext &C = getContext();
+  int64_t Value;
+  bool Abs = F.getSFrameAddrDelta().evaluateAsAbsolute(Value, *this);
+  if (!Abs) {
+    C.reportError(F.getSFrameAddrDelta().getLoc(),
+                  "invalid CFI advance_loc expression in sframe");
+    F.setSFrameAddrDelta(MCConstantExpr::create(0, C));
+    return;
+  }
+
+  SmallVector<char, 4> Data;
+  MCSFrameEmitter::encodeFuncOffset(Context, Value, Data, F.getSFrameFDE());
+  F.setVarContents(Data);
+  F.clearVarFixups();
+}
+
 bool MCAssembler::relaxFragment(MCFragment &F) {
   auto Size = computeFragmentSize(F);
   switch (F.getKind()) {
@@ -932,6 +953,9 @@ bool MCAssembler::relaxFragment(MCFragment &F) {
   case MCFragment::FT_DwarfFrame:
     relaxDwarfCallFrameFragment(F);
     break;
+  case MCFragment::FT_SFrame:
+    relaxSFrameFragment(F);
+    break;
   case MCFragment::FT_BoundaryAlign:
     relaxBoundaryAlign(static_cast<MCBoundaryAlignFragment &>(F));
     break;

llvm/lib/MC/MCFragment.cpp

@@ -53,6 +53,7 @@ LLVM_DUMP_METHOD void MCFragment::dump() const {
   case MCFragment::FT_Org:           OS << "Org"; break;
   case MCFragment::FT_Dwarf:         OS << "Dwarf"; break;
   case MCFragment::FT_DwarfFrame:    OS << "DwarfCallFrame"; break;
+  case MCFragment::FT_SFrame:        OS << "SFrame"; break;
   case MCFragment::FT_LEB:           OS << "LEB"; break;
   case MCFragment::FT_BoundaryAlign: OS<<"BoundaryAlign"; break;
   case MCFragment::FT_SymbolId:      OS << "SymbolId"; break;
@@ -79,7 +80,8 @@ LLVM_DUMP_METHOD void MCFragment::dump() const {
   case MCFragment::FT_Align:
   case MCFragment::FT_LEB:
   case MCFragment::FT_Dwarf:
-  case MCFragment::FT_DwarfFrame: {
+  case MCFragment::FT_DwarfFrame:
+  case MCFragment::FT_SFrame: {
     if (isLinkerRelaxable())
       OS << " LinkerRelaxable";
     auto Fixed = getContents();
@@ -129,6 +131,7 @@ LLVM_DUMP_METHOD void MCFragment::dump() const {
       OS << " LineDelta:" << getDwarfLineDelta();
       break;
     case MCFragment::FT_DwarfFrame:
+    case MCFragment::FT_SFrame:
       OS << " AddrDelta:";
       getDwarfAddrDelta().print(OS, nullptr);
       break;

llvm/lib/MC/MCObjectStreamer.cpp

@@ -583,6 +583,19 @@ void MCObjectStreamer::emitDwarfAdvanceFrameAddr(const MCSymbol *LastLabel,
   newFragment();
 }
 
+void MCObjectStreamer::emitSFrameCalculateFuncOffset(const MCSymbol *FuncBase,
+                                                     const MCSymbol *FREBegin,
+                                                     MCFragment *FDEFrag,
+                                                     SMLoc Loc) {
+  assert(FuncBase && "No function base address");
+  assert(FREBegin && "FRE doesn't describe a location");
+  auto *F = getCurrentFragment();
+  F->Kind = MCFragment::FT_SFrame;
+  F->setSFrameAddrDelta(buildSymbolDiff(*this, FREBegin, FuncBase, Loc));
+  F->setSFrameFDE(FDEFrag);
+  newFragment();
+}
+
 void MCObjectStreamer::emitCVLocDirective(unsigned FunctionId, unsigned FileNo,
                                           unsigned Line, unsigned Column,
                                           bool PrologueEnd, bool IsStmt,