[mte] fix compiler crash with musttail.

see D118852 for matching hwasan fix.

Reviewed By: eugenis

Differential Revision: https://reviews.llvm.org/D118854

llvm/lib/Target/AArch64/AArch64StackTagging.cpp

@@ -579,8 +579,9 @@ bool AArch64StackTagging::runOnFunction(Function &Fn) {
           Allocas[AI].LifetimeEnd.push_back(II);
       }
 
-      if (isa<ReturnInst, ResumeInst, CleanupReturnInst>(&I))
-        RetVec.push_back(&I);
+      Instruction *ExitUntag = getUntagLocationIfFunctionExit(I);
+      if (ExitUntag)
+        RetVec.push_back(ExitUntag);
     }
   }
 

llvm/test/CodeGen/AArch64/stack-tagging-musttail.ll

@@ -0,0 +1,29 @@
+; Regression test for a compiler bug that caused a crash when instrumenting code
+; using musttail.
+
+; RUN: opt -S -aarch64-stack-tagging -stack-tagging-use-stack-safety=0 %s -o - | FileCheck %s
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-arm-unknown-eabi"
+
+define dso_local noundef i32 @_Z3bari(i32 noundef %0) sanitize_memtag {
+  %2 = alloca i32, align 4
+  store i32 %0, i32* %2, align 4
+  %3 = load i32, i32* %2, align 4
+  ret i32 %3
+}
+
+define dso_local noundef i32 @_Z3fooi(i32 noundef %0) sanitize_memtag {
+  %2 = alloca i32, align 4
+  %3 = alloca i32, align 4
+  store i32 %0, i32* %2, align 4
+  store volatile i32 5, i32* %3, align 4
+  %4 = load i32, i32* %2, align 4
+  %5 = load volatile i32, i32* %3, align 4
+  %6 = add nsw i32 %4, %5
+  ; CHECK: call void @llvm.aarch64.settag
+  ; CHECK: musttail call
+  ; CHECK-NOT: call void @llvm.aarch64.settag
+  %7 = musttail call noundef i32 @_Z3bari(i32 noundef %6)
+  ret i32 %7
+}