Uploaded image for project: 'Harmony'
  1. Harmony
  2. HARMONY-6523

[jvmti] GetTag() on the return_value of MethodExit handler returns JVMTI_ERROR_INVALID_OBJECT

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Open
    • Major
    • Resolution: Unresolved
    • 6.0M1, 5.0M13
    • None
    • DRLVM
    • None
    • Microsoft Windows Server 2008 (32bit) [Version 6.0.6001]

    Description

      GetTag(), a JVMTI function, returns JVMTI_ERROR_INVALID_OBJECT when it is invoked for the return_value of MethodExit handler. I'm sorry not to provide a short program that reproduces this problem, but here's the description. My program, so-called an agent, attached to java.exe through JVMTI listens to MethodExit (and several other) events. The MethodExit handler looks like the following:

      static void JNICALL
      handle_methodexit(jvmtiEnv *jvmti, JNIEnv *env, jthread thread, jmethodID method, jboolean was_popped_by_exception, jvalue return_value)
      {
      if (!was_popped_by_exception) {
      // analyze the signature of the "method" to figure out the type of the returning value
      if the returning value is of Object type {
      if (return_value.l)

      { dump "return_value.l" (see below) jlong tag; jvmti->GetTag(obj, &tag); // this returns JVMTI_ERROR_INVALID_OBJECT }

      }
      }
      }

      GetTag() works fine for the first event, but it returns an error at the second event. One thing I am little bit suspicious is that "return_value.l" of the second event is totally different than that of the first event.

      • 1st event: 0012F3AC
      • 2nd event: 13AA05F0

      Given that jobject is a pointer, that huge difference may mean that one of them may point a wrong address, although I'm not sure. The second event was raised when loadClass (Ljava/lang/String;Z)Ljava/lang/Class; of Ljava/lang/ClassLoader; returns. This problem occurs both in 5 and 6.

      My agent was compiled using Microsoft Visual Studio 2010 in C++. For this kind of error, I know that the first thing I need to suspect is my own program, as my program can corrupt the heap and cause DRLVM to fail. I cannot guarantee that my program is memory error-free, but it seems fine because my agent didn't introduce any memory-related errors when it was attached to Sun Java 6. Currently, I stopped using Sun Java 6 because its JVMTI has another problem.

      Attachments

        Activity

          People

            Unassigned Unassigned
            linjus Choonghwan Lee
            Votes:
            0 Vote for this issue
            Watchers:
            0 Start watching this issue

            Dates

              Created:
              Updated: