Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-6321448

self-suspension with JVM-internal monitors and mutexes may not be needed



    • Bug
    • Status: Closed
    • P3
    • Resolution: Fixed
    • 6
    • 6
    • hotspot
    • b77
    • generic
    • generic


      <This causes JDI jtreg test
      to hang. See Comments for more info>

      I'm opening this bug to track an e-mail thread between Jim, Ken
      and I concerning a self-suspension deadlock. The original e-mail
      message is here in the description. The various replies are in
      the comments.

      Hello Dan. I am looking (again) at netbeans problem with
      doing an invokeMethod in a JDI bkpt handler.
      Note that they said today that they get a hang even if they
      do the invoke in a different thread. Huh? That should work ok.

      So I tried it, and sure enough, I got a hang. But not the
      same hang as discussed before when two threads are both hitting
      bkpts. In this case, the debuggee has a main that
      starts a thread and then does a thread.join.
      The started thread hits a bkpt and in the debugger, the bkpt handler
      starts up a new thread to do the invokeMethod.

      In the debuggee, this thread is hung doing a println:

      ----------------- t@13 -----------------
      0xff31f760 ___lwp_mutex_lock + 0x8
      0xfe89d0b4 void Mutex::lock() + 0x148
      0xfed429c4 HeapWord*CompactingPermGen::mem_allocate(unsigned) + 0x24
      0xfeaf37cc oopDesc*CollectedHeap::permanent_obj_allocate(KlassHandle,int,Thread*) + 0x48
      0xfe8e30cc instanceOopDesc*instanceKlass::allocate_permanent_instance(Thread*) + 0x64
      0xfe8e7de8 Handle java_lang_String::create_tenured_from_unicode(unsigned short*,int,Thread*) + 0x2c
      0xfe9a4b10 oopDesc*StringTable::basic_add(int,Handle,unsigned short*,int,unsigned,Thread*) + 0x6c
      0xfed8d92c oopDesc*StringTable::intern(Handle,unsigned short*,int,Thread*) + 0x108
      0xfe8e4a08 oopDesc*StringTable::intern(symbolOopDesc*,Thread*) + 0x40
      0xfe8e3284 oopDesc*constantPoolOopDesc::string_at_impl(constantPoolHandle,int,Thread*) + 0x80
      0xfeb68d6c void InterpreterRuntime::ldc(JavaThread*,bool) + 0x168
      0xfc414df4 * InvokeBreakpointTarg.invokee() bci:3 line:45 (Interpreted frame)

      It is waiting for the Heap_lock.
      The main thread is hung here doing the Thread.join:

      0xff31f7d8 ___lwp_cond_wait + 0x4
      0xfed18720 bool Monitor::wait(bool,long) + 0x13c
      0xfed29890 int os::pd_self_suspend_thread(Thread*) + 0xc
      0xfed9d598 int JavaThread::java_suspend_self() + 0xc4
      0xfed18d48 bool Monitor::wait(bool,long) + 0x764
      0xfe9984ac void VMThread::execute(VM_Operation*) + 0x1b0
      0xfeaa2d70 BiasedLocking::Condition BiasedLocking::revoke_and_rebias(Handle,bool,Thread*) + 0xa8
      0xfe99e2c4 void ObjectSynchronizer::wait(Handle,long long,Thread*) + 0x50
      0xfe99e048 JVM_MonitorWait + 0x294
      0xfc4134fc * java.lang.Object.wait(long) bci:134425908 (Interpreted frame)
      0xfc4134a8 * java.lang.Object.wait(long) bci:0 (Interpreted frame)
      0xfc4057e8 * java.lang.Thread.join(long) bci:38 line:1111 (Interpreted frame)
      0xfc4057e8 * java.lang.Thread.join() bci:2 line:1164 (Interpreted frame)
      0xfc4057e8 * InvokeBreakpointTarg.main(java.lang.String[]) bci:43 line:39 (Interpreted frame)
      0xfc40021c <StubRoutines>
      0xfe8f78a0 void JavaCalls::call_helper(JavaValue*,methodHandle*,JavaCallArguments*,Thread*) + 0x1e0
      0xfe9b74e8 jni_CallStaticVoidMethod + 0x500
      0x00013220 main + 0x1d10

      It owns the Heap_lock and is trying to do a VM op to do the revoke.

      Recall that this is the same intermittent hang we saw for
      jtreg InstancesTest.java for a few nightlys.

      The VMThread::execute is calling wait (I don't know why) and wait, in
      mutex_solaris.cpp is doing this:

            // were we externally suspended while we were waiting?
            if (jt->handle_special_suspend_equivalent_condition()) {
      // Our event wait has finished and we own the _Lock_Event, but
      // while we were waiting another thread suspended us. We don't
      // want to hold the _Lock_Event while suspended because that
      // would surprise the thread that suspended us.

      which Ken says this is a no-no and should be deleted (and ditto for
      similar code in the other pd files).
      Should we open a bug on this and have you fix it?

      - jjh


        Issue Links



              dcubed Daniel Daugherty
              dcubed Daniel Daugherty
              0 Vote for this issue
              5 Start watching this issue