URGENT (XS) RFR: 8149697: Fix for 8145725 is broken

David Holmes david.holmes at oracle.com
Thu Feb 11 20:37:32 UTC 2016


bug: https://bugs.openjdk.java.net/browse/JDK-8149697

webrev: http://cr.openjdk.java.net/~dholmes/8149697/webrev/

In the refactoring for 8145725 a check of _cur_index was moved outside 
the mutex block, but this was wrong as the value could be changed by a 
parking thread. As per the existing comment we needed to check the saved 
'index' instead:

--- old/src/os/linux/vm/os_linux.cpp	2016-02-11 15:23:59.696802390 -0500
+++ new/src/os/linux/vm/os_linux.cpp	2016-02-11 15:23:57.440675175 -0500
@@ -5756,7 +5756,7 @@
    int index = _cur_index;
    status = pthread_mutex_unlock(_mutex);
    assert_status(status == 0, status, "invariant");
-  if (s < 1 && _cur_index != -1) {
+  if (s < 1 && index != -1) {
      // thread is definitely parked
      status = pthread_cond_signal(&_cond[index]);
      assert_status(status == 0, status, "invariant");

This could introduce spurious wakeups for a parking thread.

Thanks,
David


More information about the hotspot-runtime-dev mailing list