Review Request: 7196045 Possible JVM deadlock in ThreadTimesClosure

[David Holmes]

Hi Kevin,

On 5/09/2012 8:35 AM, Kevin Walls wrote:
> I'd like to get this reviewed, it's a deadlock that can happen due to
> Java objects being allocated while holding Threads_lock.
>
> We need to collect just the characters of the thread names while holding
> the lock, and create the String objects when it's released. You do need
> to hit the (non-public) HotspotInternal MBean very rapidly to trigger
> this reliably, but there's a small chance the deadlock could happen when
> it's called by the tool that is meant to call it.
>
> http://cr.openjdk.java.net/~kevinw/7196045/webrev.01/

The only thing I don't like here is that the use of NEW_C_HEAP_ARRAY 
will cause an abort on out-of-memory and I'm not sure how big this array 
might be. If you are running out of C-Heap then you're on the edge of 
the cliff anyway and likely to fall at any minute, but I still dislike 
seeing more cases added to the code. I don't see an alternative though. :(

David


> Thanks!
> Kevin
>