x86_64: Fix calls to __morestack under the large code model.

[pcc]

Summary

Under the large code model, we cannot assume that morestack lives within
2^31 bytes of the call site, so we cannot use pc-relative addressing. We
cannot perform the call via a temporary register, as the rax register may
be used to store the static chain, and all other suitable registers may be
either callee-save or used for parameter passing. We cannot use the stack
at this point either because morestack manipulates the stack directly.

To avoid these issues, perform an indirect call via a read-only memory
location containing the address.

This solution is not perfect, as it assumes that the .rodata section
is laid out within 2^31 bytes of each function body, but this seems to
be sufficient for JIT.

[reames]

LGTM w/comments addressed.

include/llvm/CodeGen/MachineModuleInfo.h

164 ↗

(On Diff #17657)

Add a comment about what this means and why it's needed.

lib/CodeGen/AsmPrinter/AsmPrinter.cpp

998 ↗

(On Diff #17657)

I'm concerned about how this interacts with the code just before it w.r.t. changing sections. Can it be moved right above the trampoline check?

lib/Target/X86/X86FrameLowering.cpp

1644 ↗

(On Diff #17657)

Can you be slightly more verbose here? In particular, your comments about not having a free register in general is important. Also, document the assumption w.r.t. section layout.

llvm/trunk/include/llvm/CodeGen/MachineModuleInfo.h

class MachineModuleInfo : public ImmutablePass {
   /// in this module.
   bool DbgInfoAvailable;
 
   /// UsesVAFloatArgument - True if this module calls VarArg function with
   /// floating-point arguments.  This is used to emit an undefined reference
   /// to _fltused on Windows targets.
   bool UsesVAFloatArgument;
 
+  /// UsesMorestackAddr - True if the module calls the __morestack function
+  /// indirectly, as is required under the large code model on x86. This is used
+  /// to emit a definition of a symbol, __morestack_addr, containing the
+  /// address. See comments in lib/Target/X86/X86FrameLowering.cpp for more
+  /// details.
+  bool UsesMorestackAddr;
+
 public:
   static char ID; // Pass identification, replacement for typeid
 
   struct VariableDbgInfo {
     TrackingMDNodeRef Var;
     TrackingMDNodeRef Expr;
     unsigned Slot;
     DebugLoc Loc;
public:
   bool usesVAFloatArgument() const {
     return UsesVAFloatArgument;
   }
 
   void setUsesVAFloatArgument(bool b) {
     UsesVAFloatArgument = b;
   }
 
+  bool usesMorestackAddr() const {
+    return UsesMorestackAddr;
+  }
+
+  void setUsesMorestackAddr(bool b) {
+    UsesMorestackAddr = b;
+  }
+
   /// \brief Returns a reference to a list of cfi instructions in the current
   /// function's prologue.  Used to construct frame maps for debug and exception
   /// handling comsumers.
   const std::vector<MCCFIInstruction> &getFrameInstructions() const {
     return FrameInstructions;
   }
 
   unsigned LLVM_ATTRIBUTE_UNUSED_RESULT

llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

bool AsmPrinter::doFinalization(Module &M) {
   assert(MI && "AsmPrinter didn't require GCModuleInfo?");
   for (GCModuleInfo::iterator I = MI->end(), E = MI->begin(); I != E; )
     if (GCMetadataPrinter *MP = GetOrCreateGCPrinter(**--I))
       MP->finishAssembly(M, *MI, *this);
 
   // Emit llvm.ident metadata in an '.ident' directive.
   EmitModuleIdents(M);
 
+  // Emit __morestack address if needed for indirect calls.
+  if (MMI->usesMorestackAddr()) {
+    const MCSection *ReadOnlySection =
+        getObjFileLowering().getSectionForConstant(SectionKind::getReadOnly(),
+                                                   /*C=*/nullptr);
+    OutStreamer.SwitchSection(ReadOnlySection);
+
+    MCSymbol *AddrSymbol =
+        OutContext.GetOrCreateSymbol(StringRef("__morestack_addr"));
+    OutStreamer.EmitLabel(AddrSymbol);
+
+    const DataLayout &DL = *TM.getSubtargetImpl()->getDataLayout();
+    unsigned PtrSize = DL.getPointerSize(0);
+    OutStreamer.EmitSymbolValue(GetExternalSymbolSymbol("__morestack"),
+                                PtrSize);
+  }
+
   // If we don't have any trampolines, then we don't require stack memory
   // to be executable. Some targets have a directive to declare this.
   Function *InitTrampolineIntrinsic = M.getFunction("llvm.init.trampoline");
   if (!InitTrampolineIntrinsic || InitTrampolineIntrinsic->use_empty())
     if (const MCSection *S = MAI->getNonexecutableStackSection(OutContext))
       OutStreamer.SwitchSection(S);
 
   // Allow the target to emit any magic that it wants at the end of the file,

llvm/trunk/lib/CodeGen/MachineModuleInfo.cpp

 }
 
 bool MachineModuleInfo::doInitialization(Module &M) {
 
   ObjFileMMI = nullptr;
   CurCallSite = 0;
   CallsEHReturn = 0;
   CallsUnwindInit = 0;
-  DbgInfoAvailable = UsesVAFloatArgument = false; 
+  DbgInfoAvailable = UsesVAFloatArgument = UsesMorestackAddr = false;
   // Always emit some info, by default "no personality" info.
   Personalities.push_back(nullptr);
   AddrLabelSymbols = nullptr;
   TheModule = nullptr;
 
   return false;
 }
 

llvm/trunk/lib/Target/X86/X86FrameLowering.cpp

X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
   } else {
     BuildMI(allocMBB, DL, TII.get(X86::PUSHi32))
       .addImm(X86FI->getArgumentStackSize());
     BuildMI(allocMBB, DL, TII.get(X86::PUSHi32))
       .addImm(StackSize);
   }
 
   // __morestack is in libgcc
+  if (Is64Bit && MF.getTarget().getCodeModel() == CodeModel::Large) {
+    // Under the large code model, we cannot assume that __morestack lives
+    // within 2^31 bytes of the call site, so we cannot use pc-relative
+    // addressing. We cannot perform the call via a temporary register,
+    // as the rax register may be used to store the static chain, and all
+    // other suitable registers may be either callee-save or used for
+    // parameter passing. We cannot use the stack at this point either
+    // because __morestack manipulates the stack directly.
+    //
+    // To avoid these issues, perform an indirect call via a read-only memory
+    // location containing the address.
+    //
+    // This solution is not perfect, as it assumes that the .rodata section
+    // is laid out within 2^31 bytes of each function body, but this seems
+    // to be sufficient for JIT.
+    BuildMI(allocMBB, DL, TII.get(X86::CALL64m))
+        .addReg(X86::RIP)
+        .addImm(0)
+        .addReg(0)
+        .addExternalSymbol("__morestack_addr")
+        .addReg(0);
+    MF.getMMI().setUsesMorestackAddr(true);
+  } else {
     if (Is64Bit)
       BuildMI(allocMBB, DL, TII.get(X86::CALL64pcrel32))
         .addExternalSymbol("__morestack");
     else
       BuildMI(allocMBB, DL, TII.get(X86::CALLpcrel32))
         .addExternalSymbol("__morestack");
+  }
 
   if (IsNested)
     BuildMI(allocMBB, DL, TII.get(X86::MORESTACK_RET_RESTORE_R10));
   else
     BuildMI(allocMBB, DL, TII.get(X86::MORESTACK_RET));
 
   allocMBB->addSuccessor(&prologueMBB);
 

llvm/trunk/test/CodeGen/X86/segmented-stacks.ll

 ; RUN: llc < %s -mcpu=generic -mtriple=i686-linux -verify-machineinstrs | FileCheck %s -check-prefix=X32-Linux
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux  -verify-machineinstrs | FileCheck %s -check-prefix=X64-Linux
+; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux -code-model=large -verify-machineinstrs | FileCheck %s -check-prefix=X64-Linux-Large
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-gnux32 -verify-machineinstrs | FileCheck %s -check-prefix=X32ABI
 ; RUN: llc < %s -mcpu=generic -mtriple=i686-darwin -verify-machineinstrs | FileCheck %s -check-prefix=X32-Darwin
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-darwin -verify-machineinstrs | FileCheck %s -check-prefix=X64-Darwin
 ; RUN: llc < %s -mcpu=generic -mtriple=i686-mingw32 -verify-machineinstrs | FileCheck %s -check-prefix=X32-MinGW
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-freebsd -verify-machineinstrs | FileCheck %s -check-prefix=X64-FreeBSD
 ; RUN: llc < %s -mcpu=generic -mtriple=i686-dragonfly -verify-machineinstrs | FileCheck %s -check-prefix=X32-DFlyBSD
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-dragonfly -verify-machineinstrs | FileCheck %s -check-prefix=X64-DFlyBSD
 ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-mingw32 -verify-machineinstrs | FileCheck %s -check-prefix=X64-MinGW

 ; X64-Linux:       cmpq %fs:112, %rsp
 ; X64-Linux-NEXT:  ja      .LBB0_2
 
 ; X64-Linux:       movabsq $40, %r10
 ; X64-Linux-NEXT:  movabsq $0, %r11
 ; X64-Linux-NEXT:  callq __morestack
 ; X64-Linux-NEXT:  ret
 
+; X64-Linux-Large-LABEL:       test_basic:
+
+; X64-Linux-Large:       cmpq %fs:112, %rsp
+; X64-Linux-Large-NEXT:  ja      .LBB0_2
+
+; X64-Linux-Large:       movabsq $40, %r10
+; X64-Linux-Large-NEXT:  movabsq $0, %r11
+; X64-Linux-Large-NEXT:  callq *__morestack_addr(%rip)
+; X64-Linux-Large-NEXT:  ret
+
 ; X32ABI-LABEL:       test_basic:
 
 ; X32ABI:       cmpl %fs:64, %esp
 ; X32ABI-NEXT:  ja      .LBB0_2
 
 ; X32ABI:       movl $40, %r10d
 ; X32ABI-NEXT:  movl $0, %r11d
 ; X32ABI-NEXT:  callq __morestack

 ; X32-DFlyBSD-LABEL: test_nostack:
 ; X32-DFlyBSD-NOT:   calll __morestack
 
 ; X64-DFlyBSD-LABEL: test_nostack:
 ; X64-DFlyBSD-NOT:   callq __morestack
 }
 
 attributes #0 = { "split-stack" }
+
+; X64-Linux-Large: .rodata
+; X64-Linux-Large-NEXT: __morestack_addr:
+; X64-Linux-Large-NEXT: .quad __morestack