RFR: JDK-8220671: Initialization race for non-JavaThread PtrQueues
kim.barrett at oracle.com
Tue Mar 19 23:12:26 UTC 2019
> On Mar 19, 2019, at 3:28 AM, Roman Kennke <rkennke at redhat.com> wrote:
> Am 19. März 2019 08:16:49 MEZ schrieb Kim Barrett <kim.barrett at oracle.com>:
>> Again, don’t care about safepoints per se, it’s the interaction between
>> on_thread_attach and changing the global SATB active state that
> Right! Zhengyu suggested to use GlobalCounters multi-read/single-write protection around the relevant operations. I might try that today. What do you think?
For the record (we’re long past here I think), I don’t think that’s a good approach.
It puts a blocking operation inside a critical section, which is often a path to deadlock.
I’m not certain there’s an actual deadlock case, but I’m not certain there’s not.
More information about the hotspot-gc-dev