On Fri, 20 Jun 2025 20:47:37 GMT, Coleen Phillimore <cole...@openjdk.org> wrote:
> I think this should prevent this race but I'd have to think about it some
> more.
Aside from the fact the `checked` method is not used much, the problem is that
if the caller does not have something keeping the class alive, then the
resolution of the jMethodID can succeed and we will proceed with trying to call
the method. In the meantime the fact the class is unreferenced could be noticed
and the class then unloaded. Now that can only happen at safepoints, so it then
depends on the details of the code that tries to invoke the method e.g. in
jni.cpp
static void jni_invoke_static(JNIEnv *env, JavaValue* result, jobject receiver,
JNICallType call_type, jmethodID method_id, JNI_ArgumentPusher *args, TRAPS) {
methodHandle method(THREAD, Method::resolve_jmethod_id(method_id));
// Create object to hold arguments for the JavaCall, and associate it with
// the jni parser
ResourceMark rm(THREAD);
int number_of_parameters = method->size_of_parameters();
JavaCallArguments java_args(number_of_parameters);
assert(method->is_static(), "method should be static");
// Fill out JavaCallArguments object
args->push_arguments_on(&java_args);
// Initialize result type
result->set_type(args->return_type());
// Invoke the method. Result is returned as oop.
JavaCalls::call(result, method, &java_args, CHECK);
Can we hit safepoint checks anywhere on the path to the actual invocation of
the method? If not, what is guaranteeing that?
-------------
PR Comment: https://git.openjdk.org/jdk/pull/25267#issuecomment-2994989682
