RFR(S): 8032011 nsk/stress/jck60/jck60022 crashes in src\share\vm\runtime\synchronizer.cpp:239

[Igor Veresov]

Looks good to me.

igor

On Feb 10, 2014, at 3:07 PM, Roland Westrelin <roland.westrelin at oracle.com> wrote:

> With the code sketched below:
> 
> m1() {
>  synchronized(monitor) { // eliminated by c2
>    m2(); // deoptimized here
>  }
> }
> 
> m2() {
>  synchronized(monitor) { // not eliminated but biased to current thread
>    // uncommon trap
>  }
> }
> 
> 1) In m1(), the monitor is eliminated
> 2) m2() calls m1(), the monitor is acquired biased to current thread
> 3) uncommon trap happens in m2() which causes the bias on monitor to be revoked. The monitor is erroneoulsy locked using a basic lock in m1()
> 4) execution returns to m1() which is mark for deoptimization. Deoptimization tries to relock the eliminated monitor but find it already locked by current thread which should be impossible.
> 
> Skipping all eliminated monitors in get_or_compute_monitor_info() makes sure revoke_bias does not lock the monitor using a basic lock in the wrong frame.
> 
> http://cr.openjdk.java.net/~roland/8032011/webrev.00/
> 
> Roland.