6791815: Fix for 6471657 can cause deadlock on non-Solaris platforms when initializing direct buffer support
Summary: Place the state transition inside the loop so that the VMThread could proceed for safepoint
Reviewed-by: dholmes, never, acorn
--- a/hotspot/src/share/vm/prims/jni.cpp Thu Jan 08 16:27:28 2009 -0500
+++ b/hotspot/src/share/vm/prims/jni.cpp Tue Jan 13 12:08:43 2009 -0800
@@ -2691,8 +2691,13 @@
directBufferSupportInitializeEnded = 1;
} else {
- ThreadInVMfromNative tivn(thread); // set state as yield_all can call os:sleep
while (!directBufferSupportInitializeEnded && !directBufferSupportInitializeFailed) {
+ // Set state as yield_all can call os:sleep. On Solaris, yield_all calls
+ // os::sleep which requires the VM state transition. On other platforms, it
+ // is not necessary. The following call to change the VM state is purposely
+ // put inside the loop to avoid potential deadlock when multiple threads
+ // try to call this method. See 6791815 for more details.
+ ThreadInVMfromNative tivn(thread);
os::yield_all();
}
}