[ARM] Check all terms in emitPopInst when clearing Restored for LR. (#75527)

emitPopInst checks a single function exit MBB. If other paths also exit
the function and any of there terminators uses LR implicitly, it is not
save to clear the Restored bit.

Check all terminators for the function before clearing Restored.

This fixes a mis-compile in outlined-fn-may-clobber-lr-in-caller.ll
where the machine-outliner previously introduced BLs that clobbered LR
which in turn is used by the tail call return.

Alternative to #73553

Author: fhahn

llvm/lib/Target/ARM/ARMFrameLowering.cpp

@@ -1645,9 +1645,6 @@ void ARMFrameLowering::emitPopInst(MachineBasicBlock &MBB,
         // Fold the return instruction into the LDM.
         DeleteRet = true;
         LdmOpc = AFI->isThumbFunction() ? ARM::t2LDMIA_RET : ARM::LDMIA_RET;
-        // We 'restore' LR into PC so it is not live out of the return block:
-        // Clear Restored bit.
-        Info.setRestored(false);
       }
 
       // If NoGap is true, pop consecutive registers and then leave the rest
@@ -2785,6 +2782,33 @@ void ARMFrameLowering::determineCalleeSaves(MachineFunction &MF,
   AFI->setLRIsSpilled(SavedRegs.test(ARM::LR));
 }
 
+void ARMFrameLowering::processFunctionBeforeFrameFinalized(
+    MachineFunction &MF, RegScavenger *RS) const {
+  TargetFrameLowering::processFunctionBeforeFrameFinalized(MF, RS);
+
+  MachineFrameInfo &MFI = MF.getFrameInfo();
+  if (!MFI.isCalleeSavedInfoValid())
+    return;
+
+  // Check if all terminators do not implicitly use LR. Then we can 'restore' LR
+  // into PC so it is not live out of the return block: Clear the Restored bit
+  // in that case.
+  for (CalleeSavedInfo &Info : MFI.getCalleeSavedInfo()) {
+    if (Info.getReg() != ARM::LR)
+      continue;
+    if (all_of(MF, [](const MachineBasicBlock &MBB) {
+          return all_of(MBB.terminators(), [](const MachineInstr &Term) {
+            return !Term.isReturn() || Term.getOpcode() == ARM::LDMIA_RET ||
+                   Term.getOpcode() == ARM::t2LDMIA_RET ||
+                   Term.getOpcode() == ARM::tPOP_RET;
+          });
+        })) {
+      Info.setRestored(false);
+      break;
+    }
+  }
+}
+
 void ARMFrameLowering::getCalleeSaves(const MachineFunction &MF,
                                       BitVector &SavedRegs) const {
   TargetFrameLowering::getCalleeSaves(MF, SavedRegs);

llvm/lib/Target/ARM/ARMFrameLowering.h

@@ -59,6 +59,9 @@ class ARMFrameLowering : public TargetFrameLowering {
   void determineCalleeSaves(MachineFunction &MF, BitVector &SavedRegs,
                             RegScavenger *RS) const override;
 
+  void processFunctionBeforeFrameFinalized(
+      MachineFunction &MF, RegScavenger *RS = nullptr) const override;
+
   void adjustForSegmentedStacks(MachineFunction &MF,
                                 MachineBasicBlock &MBB) const override;
 

llvm/test/CodeGen/Thumb2/outlined-fn-may-clobber-lr-in-caller.ll

@@ -11,8 +11,6 @@ target datalayout = "e-m:o-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
 
 ; Test case to make sure calling an outlined function does not clobber LR used
 ; by a tail call in caller.
-; FIXME: Currently bl OUTLINED_FUNCTION_0 clobbers LR, which in turn is used
-;        by the later call to memcpy to return to the caller.
 define void @test(ptr nocapture noundef writeonly %arg, i32 noundef %arg1, i8 noundef zeroext %arg2) unnamed_addr #0 {
 ; CHECK-LABEL: test:
 ; CHECK:       @ %bb.0: @ %bb
@@ -22,11 +20,19 @@ define void @test(ptr nocapture noundef writeonly %arg, i32 noundef %arg1, i8 no
 ; CHECK-NEXT:    cmp r1, #1
 ; CHECK-NEXT:    bne .LBB0_5
 ; CHECK-NEXT:  @ %bb.2: @ %bb4
-; CHECK-NEXT:    bl OUTLINED_FUNCTION_0
+; CHECK-NEXT:    movs r1, #1
+; CHECK-NEXT:    strb.w r1, [r0, #36]
+; CHECK-NEXT:    movs r1, #30
+; CHECK-NEXT:    strb.w r1, [r0, #34]
+; CHECK-NEXT:    add.w r1, r2, r2, lsl #3
 ; CHECK-NEXT:    ldr r2, .LCPI0_1
 ; CHECK-NEXT:    b .LBB0_4
 ; CHECK-NEXT:  .LBB0_3: @ %bb14
-; CHECK-NEXT:    bl OUTLINED_FUNCTION_0
+; CHECK-NEXT:    movs r1, #1
+; CHECK-NEXT:    strb.w r1, [r0, #36]
+; CHECK-NEXT:    movs r1, #30
+; CHECK-NEXT:    strb.w r1, [r0, #34]
+; CHECK-NEXT:    add.w r1, r2, r2, lsl #3
 ; CHECK-NEXT:    ldr r2, .LCPI0_0
 ; CHECK-NEXT:  .LBB0_4: @ %bb4
 ; CHECK-NEXT:    add.w r1, r2, r1, lsl #2