Make it so we can read the .llvm_bb_addr_map section from memory at runtime.

We make the section loadable and emit symbols to help us find the data at
runtime.

LLVM streams the BB data for each function (independently) out to the binary on
the fly. This makes it hard to provide the runtime with a simple `(address,
length)` pair that it needs to make a Rust slice of the data.

We work around this like so: for each function we emit a start and end symbol
marking the bbaddrmap data for the function. The name of the symbol encodes the
function for which the BB data corresponds. The runtime knows what functions
are available (from the LTO LLVM IR) and can therefore enumerate the symbols to
find all of the BB data.

Author: vext01

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

@@ -133,6 +133,8 @@ static cl::opt<bool>
                       cl::desc("Embed final IR as bitcode after all "
                                "optimisations and transformations have run."));
 
+extern bool YkAllocLLVMBBAddrMapSection;
+
 const char DWARFGroupName[] = "dwarf";
 const char DWARFGroupDescription[] = "DWARF Emission";
 const char DbgTimerName[] = "emit";
@@ -1324,6 +1326,20 @@ static unsigned getBBAddrMapMetadata(const MachineBasicBlock &MBB) {
          (const_cast<MachineBasicBlock &>(MBB).canFallThrough() << 3);
 }
 
+void emitYkBBAddrMapSymbol(const MachineFunction &MF, MCStreamer &OutStreamer,
+                           bool Start) {
+  std::string SymName("ykllvm.bbaddrmap.");
+  SymName.append(MF.getName().str());
+  if (Start)
+    SymName.append(".start");
+  else
+    SymName.append(".end");
+
+  MCSymbol *Sym = MF.getContext().getOrCreateSymbol(SymName);
+  OutStreamer.emitSymbolAttribute(Sym, llvm::MCSA_Global);
+  OutStreamer.emitLabel(Sym);
+}
+
 void AsmPrinter::emitBBAddrMapSection(const MachineFunction &MF) {
   MCSection *BBAddrMapSection =
       getObjFileLowering().getBBAddrMapSection(*MF.getSection());
@@ -1333,6 +1349,11 @@ void AsmPrinter::emitBBAddrMapSection(const MachineFunction &MF) {
 
   OutStreamer->pushSection();
   OutStreamer->switchSection(BBAddrMapSection);
+
+  // Add the `ykllvm.bbaddrmap.<func>.start` symbol.
+  if (YkAllocLLVMBBAddrMapSection)
+    emitYkBBAddrMapSymbol(MF, *OutStreamer, true);
+
   OutStreamer->AddComment("version");
   OutStreamer->emitInt8(OutStreamer->getContext().getBBAddrMapVersion());
   OutStreamer->AddComment("feature");
@@ -1421,6 +1442,11 @@ void AsmPrinter::emitBBAddrMapSection(const MachineFunction &MF) {
       OutStreamer->emitULEB128IntValue(I);
     }
   }
+
+  // Add the `ykllvm.bbaddrmap.<func>.end` symbol.
+  if (YkAllocLLVMBBAddrMapSection)
+    emitYkBBAddrMapSymbol(MF, *OutStreamer, false);
+
   OutStreamer->popSection();
 }
 

llvm/lib/MC/MCObjectFileInfo.cpp

@@ -1120,6 +1120,8 @@ MCObjectFileInfo::getStackSizesSection(const MCSection &TextSec) const {
                             cast<MCSymbolELF>(TextSec.getBeginSymbol()));
 }
 
+extern bool YkAllocLLVMBBAddrMapSection;
+
 MCSection *
 MCObjectFileInfo::getBBAddrMapSection(const MCSection &TextSec) const {
   if (Ctx->getObjectFileType() != MCContext::IsELF)
@@ -1133,6 +1135,9 @@ MCObjectFileInfo::getBBAddrMapSection(const MCSection &TextSec) const {
     Flags |= ELF::SHF_GROUP;
   }
 
+  if (YkAllocLLVMBBAddrMapSection)
+    Flags |= ELF::SHF_ALLOC;
+
   // Use the text section's begin symbol and unique ID to create a separate
   // .llvm_bb_addr_map section associated with every unique text section.
   return Ctx->getELFSection(".llvm_bb_addr_map", ELF::SHT_LLVM_BB_ADDR_MAP,

llvm/lib/Support/Yk.cpp

@@ -6,3 +6,9 @@ bool YkAllocLLVMBCSection;
 static cl::opt<bool, true> YkAllocLLVMBCSectionParser(
     "yk-alloc-llvmbc-section", cl::desc("Make the `.llvmbc` section loadable"),
     cl::NotHidden, cl::location(YkAllocLLVMBCSection));
+
+bool YkAllocLLVMBBAddrMapSection;
+static cl::opt<bool, true> YkAllocLLVMBBAddrMapSectionParser(
+    "yk-alloc-llvmbbaddrmap-section",
+    cl::desc("Make the `.llvmbbaddrmap` section loadable"), cl::NotHidden,
+    cl::location(YkAllocLLVMBBAddrMapSection));