diff --git a/mlir/lib/Conversion/GPUToCUDA/ConvertKernelFuncToCubin.cpp b/mlir/lib/Conversion/GPUToCUDA/ConvertKernelFuncToCubin.cpp
--- a/mlir/lib/Conversion/GPUToCUDA/ConvertKernelFuncToCubin.cpp
+++ b/mlir/lib/Conversion/GPUToCUDA/ConvertKernelFuncToCubin.cpp
@@ -31,6 +31,7 @@
 #include "llvm/IR/LegacyPassManager.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Support/Error.h"
+#include "llvm/Support/Mutex.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/TargetSelect.h"
 #include "llvm/Target/TargetMachine.h"
@@ -57,6 +58,12 @@
   void runOnOperation() override {
     gpu::GPUModuleOp module = getOperation();
 
+    // Lock access to the llvm context.
+    llvm::sys::SmartScopedLock<true> scopedLock(
+        module.getContext()
+            ->getRegisteredDialect<LLVM::LLVMDialect>()
+            ->getLLVMContextMutex());
+
     // Make sure the NVPTX target is initialized.
     LLVMInitializeNVPTXTarget();
     LLVMInitializeNVPTXTargetInfo();