diff --git a/clang/lib/CIR/Dialect/Transforms/LoweringPrepareItaniumCXXABI.h b/clang/lib/CIR/Dialect/Transforms/LoweringPrepareItaniumCXXABI.h
index 1dbef0d24ddd..df96fa73f9f3 100644
--- a/clang/lib/CIR/Dialect/Transforms/LoweringPrepareItaniumCXXABI.h
+++ b/clang/lib/CIR/Dialect/Transforms/LoweringPrepareItaniumCXXABI.h
@@ -20,5 +20,8 @@ class LoweringPrepareItaniumCXXABI : public cir::LoweringPrepareCXXABI {
                                clang::ASTContext &astCtx,
                                mlir::cir::DynamicCastOp op) override;
   mlir::Value lowerVAArg(cir::CIRBaseBuilderTy &builder, mlir::cir::VAArgOp op,
-                         const cir::CIRDataLayout &datalayout) override;
+                         const cir::CIRDataLayout &datalayout) override {
+    // Itanium C++ ABI has nothing to do with VAArg.
+    return {};
+  }
 };
diff --git a/clang/lib/CIR/Dialect/Transforms/TargetLowering/Targets/LoweringPrepareItaniumCXXABI.cpp b/clang/lib/CIR/Dialect/Transforms/TargetLowering/Targets/LoweringPrepareItaniumCXXABI.cpp
index 9d79fb7ccb43..11d66f8e66db 100644
--- a/clang/lib/CIR/Dialect/Transforms/TargetLowering/Targets/LoweringPrepareItaniumCXXABI.cpp
+++ b/clang/lib/CIR/Dialect/Transforms/TargetLowering/Targets/LoweringPrepareItaniumCXXABI.cpp
@@ -163,11 +163,3 @@ LoweringPrepareItaniumCXXABI::lowerDynamicCast(CIRBaseBuilderTy &builder,
           })
       .getResult();
 }
-
-mlir::Value LoweringPrepareItaniumCXXABI::lowerVAArg(
-    CIRBaseBuilderTy &builder, mlir::cir::VAArgOp op,
-    const ::cir::CIRDataLayout &datalayout) {
-  // There is no generic cir lowering for var_arg, here we fail
-  // so to prevent attempt of calling lowerVAArg for ItaniumCXXABI
-  llvm_unreachable("NYI");
-}
diff --git a/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp b/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
index b38c4608f875..d98fda81fd69 100644
--- a/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
+++ b/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
@@ -1603,7 +1603,12 @@ class CIRVAArgLowering : public mlir::OpConversionPattern<mlir::cir::VAArgOp> {
   mlir::LogicalResult
   matchAndRewrite(mlir::cir::VAArgOp op, OpAdaptor adaptor,
                   mlir::ConversionPatternRewriter &rewriter) const override {
-    return op.emitError("cir.vaarg lowering is NYI");
+    auto vaArgResTy = getTypeConverter()->convertType(op.getType());
+    auto opaquePtr = mlir::LLVM::LLVMPointerType::get(getContext());
+    auto va_list = rewriter.create<mlir::LLVM::BitcastOp>(
+        op.getLoc(), opaquePtr, adaptor.getOperands().front());
+    rewriter.replaceOpWithNewOp<mlir::LLVM::VaArgOp>(op, vaArgResTy, va_list);
+    return mlir::success();
   }
 };
 
diff --git a/clang/test/CIR/CodeGen/var-arg.c b/clang/test/CIR/CodeGen/var-arg.c
index e689adefebe5..fc9a7fe99139 100644
--- a/clang/test/CIR/CodeGen/var-arg.c
+++ b/clang/test/CIR/CodeGen/var-arg.c
@@ -2,6 +2,8 @@
 // RUN: %clang_cc1 -triple aarch64-none-linux-android21 -fclangir -emit-cir -mmlir --mlir-print-ir-after=cir-lowering-prepare %s -o %t.cir 2>&1 | FileCheck %s -check-prefix=AFTER
 // RUN: %clang_cc1 -triple aarch64-none-linux-android21 -fclangir -emit-llvm %s -o %t.ll
 // RUN: FileCheck --input-file=%t.ll %s -check-prefix=LLVM
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t.ll
+// RUN: FileCheck --input-file=%t.ll %s -check-prefix=LLVM-X86
 #include <stdarg.h>
 
 int f1(int n, ...) {
@@ -118,3 +120,9 @@ int f1(int n, ...) {
 // LLVM: store i32 [[RES]], ptr [[RETP]], align 4,
 // LLVM: [[RETV:%.*]] = load i32, ptr [[RETP]], align 4,
 // LLVM-NEXT: ret i32 [[RETV]],
+
+// LLVM-X86: %struct.__va_list_tag = type { i32, i32, ptr, ptr }
+// LLVM-X86: define {{.*}} @f1
+// LLVM-X86: call void @llvm.va_start.p0(
+// LLVM-X86: va_arg ptr %{{.*}}, i32
+// LLVM-X86: call void @llvm.va_end.p0(
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
index fb3b8e77c492..860c1948e376 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
@@ -597,6 +597,26 @@ def LLVM_LandingpadOp : LLVM_Op<"landingpad"> {
   let hasVerifier = 1;
 }
 
+//===----------------------------------------------------------------------===//
+// VAArgOp
+//===----------------------------------------------------------------------===//
+
+def LLVM_VaArgOp : LLVM_Op<"va_arg"> {
+  let arguments = (ins LLVM_AnyPointer:$arg);
+  let results = (outs LLVM_Type:$res);
+
+  let builders = [LLVM_OneResultOpBuilder];
+
+  let llvmBuilder = " $res = builder.CreateVAArg($arg, $_resultType); ";
+  let assemblyFormat = "$arg attr-dict `:` `(` type($arg) `)` `->` type($res)";
+
+  string llvmInstName = "VAArg";
+  string mlirBuilder = [{
+    $res = $_builder.create<mlir::LLVM::VaArgOp>(
+      $_location, $_resultType, $arg);
+  }];
+}
+
 //===----------------------------------------------------------------------===//
 // CallOp
 //===----------------------------------------------------------------------===//
diff --git a/mlir/test/Target/LLVMIR/llvmir.mlir b/mlir/test/Target/LLVMIR/llvmir.mlir
index 7116f3b64d7f..89db1b6ffe75 100644
--- a/mlir/test/Target/LLVMIR/llvmir.mlir
+++ b/mlir/test/Target/LLVMIR/llvmir.mlir
@@ -2244,8 +2244,8 @@ llvm.func @qux(f32)
 
 // CHECK: %struct.va_list = type { ptr }
 
-// CHECK: define void @vararg_function(i32 %{{.*}}, ...)
-llvm.func @vararg_function(%arg0: i32, ...) {
+// CHECK: define i32 @vararg_function(i32 %{{.*}}, ...)
+llvm.func @vararg_function(%arg0: i32, ...) -> i32 {
   %0 = llvm.mlir.constant(1 : i32) : i32
   %1 = llvm.mlir.constant(1 : i32) : i32
   // CHECK: %[[ALLOCA0:.+]] = alloca %struct.va_list, align 8
@@ -2256,12 +2256,14 @@ llvm.func @vararg_function(%arg0: i32, ...) {
   %4 = llvm.alloca %0 x !llvm.ptr {alignment = 8 : i64} : (i32) -> !llvm.ptr
   // CHECK: call void @llvm.va_copy.p0(ptr %[[ALLOCA1]], ptr %[[ALLOCA0]])
   llvm.intr.vacopy %2 to %4 : !llvm.ptr, !llvm.ptr
+  // CHECK: %[[RET:.+]] = va_arg ptr %[[ALLOCA1]], i32
+  %ret = llvm.va_arg %4 : (!llvm.ptr) -> i32
   // CHECK: call void @llvm.va_end.p0(ptr %[[ALLOCA1]])
   // CHECK: call void @llvm.va_end.p0(ptr %[[ALLOCA0]])
   llvm.intr.vaend %4 : !llvm.ptr
   llvm.intr.vaend %2 : !llvm.ptr
-  // CHECK: ret void
-  llvm.return
+  // CHECK: ret i32 %[[RET]]
+  llvm.return %ret : i32
 }
 
 // -----