[AArch64] Add check that prologue insertion doesn't clobber live regs. (llvm#71826)

This patch extends AArch64FrameLowering::emitProglogue to check if the
inserted prologue clobbers live registers.

It updates `llvm/test/CodeGen/AArch64/framelayout-scavengingslot.mir`
with an extra load to make x9 live before the store, preserving the
original test.

It uses the original
`llvm/test/CodeGen/AArch64/framelayout-scavengingslot.mir` as
`llvm/test/CodeGen/AArch64/emit-prologue-clobber-verification.mir`,
because there x9 is marked as live on entry, but used as scratch reg as
it is not callee saved.

The new assertion catches a mis-compile in
`store-swift-async-context-clobber-live-reg.ll` on
https://github.com/apple/llvm-project/tree/next

Author: fhahn

llvm/lib/Target/AArch64/AArch64FrameLowering.cpp

@@ -1395,6 +1395,18 @@ static void emitDefineCFAWithFP(MachineFunction &MF, MachineBasicBlock &MBB,
       .setMIFlags(MachineInstr::FrameSetup);
 }
 
+/// Collect live registers from the end of \p MI's parent up to (including) \p
+/// MI in \p LiveRegs.
+static void getLivePhysRegsUpTo(MachineInstr &MI, const TargetRegisterInfo &TRI,
+                                LivePhysRegs &LiveRegs) {
+
+  MachineBasicBlock &MBB = *MI.getParent();
+  LiveRegs.addLiveOuts(MBB);
+  for (const MachineInstr &MI :
+       reverse(make_range(MI.getIterator(), MBB.instr_end())))
+    LiveRegs.stepBackward(MI);
+}
+
 void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
                                         MachineBasicBlock &MBB) const {
   MachineBasicBlock::iterator MBBI = MBB.begin();
@@ -1403,6 +1415,8 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
   const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>();
   const AArch64RegisterInfo *RegInfo = Subtarget.getRegisterInfo();
   const TargetInstrInfo *TII = Subtarget.getInstrInfo();
+  const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
+
   MachineModuleInfo &MMI = MF.getMMI();
   AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
   bool EmitCFI = AFI->needsDwarfUnwindInfo(MF);
@@ -1412,6 +1426,39 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
   bool HasWinCFI = false;
   auto Cleanup = make_scope_exit([&]() { MF.setHasWinCFI(HasWinCFI); });
 
+  MachineBasicBlock::iterator End = MBB.end();
+#ifndef NDEBUG
+  // Collect live register from the end of MBB up to the start of the existing
+  // frame setup instructions.
+  MachineBasicBlock::iterator NonFrameStart = MBB.begin();
+  while (NonFrameStart != End &&
+         NonFrameStart->getFlag(MachineInstr::FrameSetup))
+    ++NonFrameStart;
+
+  LivePhysRegs LiveRegs(*TRI);
+  if (NonFrameStart != MBB.end()) {
+    getLivePhysRegsUpTo(*NonFrameStart, *TRI, LiveRegs);
+    // Ignore registers used for stack management for now.
+    LiveRegs.removeReg(AArch64::SP);
+    LiveRegs.removeReg(AArch64::X19);
+    LiveRegs.removeReg(AArch64::FP);
+    LiveRegs.removeReg(AArch64::LR);
+  }
+
+  auto VerifyClobberOnExit = make_scope_exit([&]() {
+    if (NonFrameStart == MBB.end())
+      return;
+    // Check if any of the newly instructions clobber any of the live registers.
+    for (MachineInstr &MI :
+         make_range(MBB.instr_begin(), NonFrameStart->getIterator())) {
+      for (auto &Op : MI.operands())
+        if (Op.isReg() && Op.isDef())
+          assert(!LiveRegs.contains(Op.getReg()) &&
+                 "live register clobbered by inserted prologue instructions");
+    }
+  });
+#endif
+
   bool IsFunclet = MBB.isEHFuncletEntry();
 
   // At this point, we're going to decide whether or not the function uses a
@@ -1610,7 +1657,6 @@ void AArch64FrameLowering::emitPrologue(MachineFunction &MF,
   // Move past the saves of the callee-saved registers, fixing up the offsets
   // and pre-inc if we decided to combine the callee-save and local stack
   // pointer bump above.
-  MachineBasicBlock::iterator End = MBB.end();
   while (MBBI != End && MBBI->getFlag(MachineInstr::FrameSetup) &&
          !IsSVECalleeSave(MBBI)) {
     if (CombineSPBump)

llvm/test/CodeGen/AArch64/emit-prologue-clobber-verification.mir

@@ -0,0 +1,24 @@
+# RUN: not --crash llc -mtriple=aarch64-none-linux-gnu -run-pass=prologepilog %s -o -
+#
+# REQUIRES: asserts
+#
+---
+# x9 is marked as live on function entry, but it will be used as scratch
+# register for prologue computations at the beginning of the prologue.
+# Use this to check we catch that the prologue clobbers $x9.
+name: x9_clobbered_on_fn_entry
+tracksRegLiveness: true
+frameInfo:
+  isFrameAddressTaken: true
+stack:
+  - { id: 0, size:    16, alignment: 16 }
+  - { id: 1, size: 32768, alignment: 32 }
+body: |
+  bb.0:
+    liveins: $x0, $x1, $x2, $x3, $x4, $x5, $x6, $x7, $x8, $x9, $x10, $x11, $x12, $x13, $x14, $x15, $x16, $x17, $x18, $x19, $x20, $x21, $x22, $x23, $x24, $x25, $x26, $x27, $x28, $lr
+    STRXui $x0, %stack.0, 0
+    B %bb.1
+  bb.1:
+    liveins: $x1, $x2, $x3, $x4, $x5, $x6, $x7, $x8, $x9, $x10, $x11, $x12, $x13, $x14, $x15, $x16, $x17, $x18, $x19, $x20, $x21, $x22, $x23, $x24, $x25, $x26, $x27, $x28, $lr
+    RET_ReallyLR implicit $lr
+...

llvm/test/CodeGen/AArch64/framelayout-scavengingslot.mir

@@ -19,6 +19,7 @@ stack:
 body: |
   bb.0:
     liveins: $x0, $x8
+    $x9 = LDRXui $x0, 0 :: (load (s64))
     STRXui $x0, %stack.0, 0
     B %bb.1
   bb.1:

llvm/test/CodeGen/AArch64/framelayout-sve-calleesaves-fix.mir

@@ -30,6 +30,7 @@
   ; CHECK-NEXT:    ret
 ...
 name: fix_restorepoint_p4
+tracksRegLiveness: true
 stack:
   - { id: 0, stack-id: scalable-vector, size: 16, alignment: 16 }
 body:             |

llvm/test/CodeGen/AArch64/store-swift-async-context-clobber-live-reg.ll

@@ -0,0 +1,91 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -o - -mtriple=arm64e-apple-macosx %s | FileCheck %s
+
+target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
+
+define swifttailcc void @test_async_with_jumptable(ptr %src, ptr swiftasync %as) #0 {
+; CHECK-LABEL: test_async_with_jumptable:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    orr x29, x29, #0x1000000000000000
+; CHECK-NEXT:    str x19, [sp, #-32]! ; 8-byte Folded Spill
+; CHECK-NEXT:    stp x29, x30, [sp, #16] ; 16-byte Folded Spill
+; CHECK-NEXT:    add x16, sp, #8
+; CHECK-NEXT:    movk x16, #49946, lsl #48
+; CHECK-NEXT:    mov x17, x22
+; CHECK-NEXT:    pacdb x17, x16
+; CHECK-NEXT:    str x17, [sp, #8]
+; CHECK-NEXT:    add x29, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa w29, 16
+; CHECK-NEXT:    .cfi_offset w30, -8
+; CHECK-NEXT:    .cfi_offset w29, -16
+; CHECK-NEXT:    .cfi_offset w19, -32
+; CHECK-NEXT:    ldr x16, [x0]
+; CHECK-NEXT:    mov x20, x22
+; CHECK-NEXT:    mov x22, x0
+; CHECK-NEXT:    mov x19, x20
+; CHECK-NEXT:    cmp x16, #3
+; CHECK-NEXT:    csel x16, x16, xzr, ls
+; CHECK-NEXT:  Lloh0:
+; CHECK-NEXT:    adrp x17, LJTI0_0@PAGE
+; CHECK-NEXT:  Lloh1:
+; CHECK-NEXT:    add x17, x17, LJTI0_0@PAGEOFF
+; CHECK-NEXT:    ldrsw x16, [x17, x16, lsl #2]
+; CHECK-NEXT:  Ltmp0:
+; CHECK-NEXT:    adr x17, Ltmp0
+; CHECK-NEXT:    add x16, x17, x16
+; CHECK-NEXT:    br x16
+; CHECK-NEXT:  LBB0_1: ; %then.2
+; CHECK-NEXT:    mov x19, #0 ; =0x0
+; CHECK-NEXT:    b LBB0_3
+; CHECK-NEXT:  LBB0_2: ; %then.3
+; CHECK-NEXT:    mov x19, x22
+; CHECK-NEXT:  LBB0_3: ; %exit
+; CHECK-NEXT:    bl _foo
+; CHECK-NEXT:    mov x2, x0
+; CHECK-NEXT:    mov x0, x19
+; CHECK-NEXT:    mov x1, x20
+; CHECK-NEXT:    ldp x29, x30, [sp, #16] ; 16-byte Folded Reload
+; CHECK-NEXT:    ldr x19, [sp], #32 ; 8-byte Folded Reload
+; CHECK-NEXT:    and x29, x29, #0xefffffffffffffff
+; CHECK-NEXT:    br x2
+; CHECK-NEXT:    .loh AdrpAdd Lloh0, Lloh1
+; CHECK-NEXT:    .cfi_endproc
+; CHECK-NEXT:    .section __TEXT,__const
+; CHECK-NEXT:    .p2align 2, 0x0
+; CHECK-NEXT:  LJTI0_0:
+; CHECK-NEXT:    .long LBB0_3-Ltmp0
+; CHECK-NEXT:    .long LBB0_1-Ltmp0
+; CHECK-NEXT:    .long LBB0_1-Ltmp0
+; CHECK-NEXT:    .long LBB0_2-Ltmp0
+entry:
+  %l = load i64, ptr %src, align 8
+  switch i64 %l, label %dead [
+    i64 0, label %exit
+    i64 1, label %then.1
+    i64 2, label %then.2
+    i64 3, label %then.3
+  ]
+
+then.1:
+  br label %exit
+
+then.2:
+  br label %exit
+
+then.3:
+  br label %exit
+
+dead:                                                ; preds = %entryresume.5
+  unreachable
+
+exit:
+  %p = phi ptr [ %src, %then.3 ], [ null, %then.2 ], [ %as, %entry ], [ null, %then.1 ]
+  %r = call i64 @foo()
+  %fn = inttoptr i64 %r to ptr
+  musttail call swifttailcc void %fn(ptr swiftasync %src, ptr %p, ptr %as)
+  ret void
+}
+
+declare i64 @foo()
+
+attributes #0 = { "frame-pointer"="non-leaf" }