synchronization of JVMTI phase notifications [Fwd: Data visibility between threads in Hotspot]

[Hiroshi Yamauchi]

Dan,

I don't have a small test for this, unfortunately, It happens about
once in 100k runs of some javac job. It's not easy to reproduce, but
reliably happens in a certain setup. It's a race condition bug :(

Hiroshi

On Thu, Feb 19, 2009 at 12:16 PM, Daniel D. Daugherty
<Daniel.Daugherty at sun.com> wrote:
> Hiroshi,
>
> Do you have a test program for this problem?
>
> That would help with my analysis of the code paths...
>
> Dan
>
>
> Hiroshi Yamauchi wrote:
>
> There are two bugs:
>
> 1. A CompiledMethodLoad event is sent after the VM death event.
> 2. SetEventCallbacks and SetEventNotificationMode calls may not be
> respected because of the race condition/visibility issue.
>
> Delaying the posting of the VM death event will probably fix bug 1 and
> we can probably avoid using mutex locks in event posting. But I
> suspect that fixing bug 2 will need mutex locks after all. Is there a
> way around mutex locks here?
>
> Thanks,
> Hiroshi
>