RFR: 8263896: Remove not_suspended parameter from ObjectMonitor::exit()

David Holmes dholmes at openjdk.java.net
Tue Mar 23 03:08:40 UTC 2021


On Sat, 20 Mar 2021 00:24:43 GMT, Patricio Chilano Mateo <pchilanomate at openjdk.org> wrote:

> Hi,
> 
> Please review the following small patch. The boolean parameter not_suspended is used to detect if we need to set the current JavaThread exiting the monitor as the previous owner (_previous_owner_tid). If not_suspended is true then we set _previous_owner_tid, otherwise we skip the write (modulo the JFR checks). This parameter is always true except when we call exit() from inside enter(). This happens when the JavaThread acquires the monitor but notices that it was suspended while being in the _thread_blocked state. Since in that case the JT was never really "the owner" we skip setting _previous_owner_tid.
> 
> This behaviour of releasing the monitor is just an implementation detail of ObjectMonitor::enter() which doesn't need to be exposed in the exit() API. We can identify the same scenario just by checking _current_pending_monitor instead.
> 
> Thanks,
> Patricio

Hi Patricio,

I'm in two-minds about using incidental state to infer whether to store the last_owner_tid here. Someone might reasonably argue that we should actually be clearing the pending-monitor field whilst in the self-suspend loop.

Is this going to feed into Robbin's thread suspension changes? Otherwise I'm not really seeing the motivation.

Thanks,
David

-------------

PR: https://git.openjdk.java.net/jdk/pull/3101


More information about the hotspot-runtime-dev mailing list