[X86] Insure split-stack prolog check is retained for sibling calls

[thanm]

Summary

If a routine with no stack frame makes a sibling call, we need to
preserve the stack space check even if the local stack frame is empty,
since the call target could be a "no-split" function (in which case
the linker needs to be able to fix up the prolog sequence in order to
switch to a larger stack).

This fixes PR37807.

[cherry]

LGTM. Thanks.

[thanm]

If there are no other comments or concerns on this patch, I plan to submit it tomorrow. Thanks -NM.

[thanm]

Update commit message to reflect that this is for both ARM and x86.

llvm/trunk/lib/Target/ARM/ARMFrameLowering.cpp

void ARMFrameLowering::adjustForSegmentedStacks(
   const MCRegisterInfo *MRI = Context.getRegisterInfo();
   const ARMBaseInstrInfo &TII =
       *static_cast<const ARMBaseInstrInfo *>(MF.getSubtarget().getInstrInfo());
   ARMFunctionInfo *ARMFI = MF.getInfo<ARMFunctionInfo>();
   DebugLoc DL;
 
   uint64_t StackSize = MFI.getStackSize();
 
-  // Do not generate a prologue for functions with a stack of size zero
-  if (StackSize == 0)
+  // Do not generate a prologue for leaf functions with a stack of size zero.
+  // For non-leaf functions we have to allow for the possibility that the
+  // call is to a non-split function, as in PR37807.
+  if (StackSize == 0 && !MFI.hasTailCall())
     return;
 
   // Use R4 and R5 as scratch registers.
   // We save R4 and R5 before use and restore them before leaving the function.
   unsigned ScratchReg0 = ARM::R4;
   unsigned ScratchReg1 = ARM::R5;
   uint64_t AlignedStackSize;
 

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

if (!STI.isTargetLinux() && !STI.isTargetDarwin() && !STI.isTargetWin32() &&
       !STI.isTargetDragonFly())
     report_fatal_error("Segmented stacks not supported on this platform.");
 
   // Eventually StackSize will be calculated by a link-time pass; which will
   // also decide whether checking code needs to be injected into this particular
   // prologue.
   StackSize = MFI.getStackSize();
 
-  // Do not generate a prologue for functions with a stack of size zero
-  if (StackSize == 0)
+  // Do not generate a prologue for leaf functions with a stack of size zero.
+  // For non-leaf functions we have to allow for the possibility that the
+  // call is to a non-split function, as in PR37807.
+  if (StackSize == 0 && !MFI.hasTailCall())
     return;
 
   MachineBasicBlock *allocMBB = MF.CreateMachineBasicBlock();
   MachineBasicBlock *checkMBB = MF.CreateMachineBasicBlock();
   X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
   bool IsNested = false;
 
   // We need to know if the function has a nest argument only in 64 bit mode.

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

 
 ; ARM-linux-LABEL: test_nostack:
 ; ARM-linux-NOT:   bl __morestack
 
 ; ARM-android-LABEL: test_nostack:
 ; ARM-android-NOT:   bl __morestack
 }
 
+; Test to make sure that a morestack call is generated if there is a
+; sibling call, even if the function in question has no stack frame
+; (PR37807).
+
+declare i32 @callee(i32)
+
+define i32 @test_sibling_call_empty_frame(i32 %x) #0 {
+  %call = tail call i32 @callee(i32 %x) #0
+  ret i32 %call
+
+; ARM-linux:      test_sibling_call_empty_frame:
+; ARM-linux:      bl      __morestack
+
+; ARM-android:      test_sibling_call_empty_frame:
+; ARM-android:      bl      __morestack
+
+}
+
 attributes #0 = { "split-stack" }

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

 ; X64-DFlyBSD-LABEL: test_nostack:
 ; X64-DFlyBSD-NOT:   callq __morestack
 }
 
 define void @test_nosplitstck() {
   ret void
 }
 
+; Test to make sure that a morestack call is generated if there is a
+; sibling call, even if the function in question has no stack frame
+; (PR37807).
+
+declare i32 @callee(i32)
+
+define i32 @test_sibling_call_empty_frame(i32 %x) #0 {
+  %call = tail call i32 @callee(i32 %x) #0
+  ret i32 %call
+
+; X32-Linux-LABEL:       test_sibling_call_empty_frame:
+; X32-Linux:  calll __morestack
+
+; X64-Linux-LABEL:       test_sibling_call_empty_frame:
+; X64-Linux:  callq __morestack
+
+; X64-Linux-Large-LABEL:       test_sibling_call_empty_frame:
+; X64-Linux-Large:  callq *__morestack_addr(%rip)
+
+; X32ABI-LABEL:       test_sibling_call_empty_frame:
+; X32ABI:  callq __morestack
+
+; X32-Darwin-LABEL:      test_sibling_call_empty_frame:
+; X32-Darwin: calll ___morestack
+
+; X64-Darwin-LABEL:      test_sibling_call_empty_frame:
+; X64-Darwin: callq ___morestack
+
+; X32-MinGW-LABEL:       test_sibling_call_empty_frame:
+; X32-MinGW:  calll ___morestack
+
+; X64-MinGW-LABEL:       test_sibling_call_empty_frame:
+; X64-MinGW:  callq __morestack
+
+; X64-FreeBSD-LABEL:       test_sibling_call_empty_frame:
+; X64-FreeBSD:  callq __morestack
+
+; X32-DFlyBSD-LABEL:       test_sibling_call_empty_frame:
+; X32-DFlyBSD:  calll __morestack
+; X32-DFlyBSD-NEXT:  ret
+
+; X64-DFlyBSD-LABEL:       test_sibling_call_empty_frame:
+; X64-DFlyBSD:  callq __morestack
+
+}
+
 attributes #0 = { "split-stack" }
 
 ; X64-Linux-Large: .rodata
 ; X64-Linux-Large-NEXT: __morestack_addr:
 ; X64-Linux-Large-NEXT: .quad __morestack
 
 ; X32-Linux: .section ".note.GNU-split-stack","",@progbits
 ; X32-Linux: .section ".note.GNU-no-split-stack","",@progbits