RFR: 8225344: Avoid VM_EnableBiasedLocking VM operation during bootstrap if BiasedLockingStartupDelay is 0

David Holmes david.holmes at oracle.com
Wed Jun 26 16:01:33 UTC 2019


Hi Claes,

On 26/06/2019 9:10 am, Claes Redestad wrote:
> Hi,
> 
> please review this cleanup to avoid a safepointing VM operation during
> bootstrap to enable biased locking if the configured delay is zero
> (current default).
> 
> Webrev: http://cr.openjdk.java.net/~redestad/8225344/open.00/
> Bug:    https://bugs.openjdk.java.net/browse/JDK-8225344

I don't understand the change here:

   bool BiasedLocking::enabled() {
!   assert(UseBiasedLocking, "precondition");
!   return _biased_locking_enabled || BiasedLockingStartupDelay == 0;
   }

This only makes a difference is someone can call this before 
BiasedLocking::init is called, but the existing code has the same 
constraint AFAICS.

Thanks,
David

> While this avoids a safepoint during bootstrap it's effectively
> performance neutral on all startup and throughput tests I've done.
> Still, it seems like a good cleanup to reduce the amount of visible
> (logged) work happening during a typical bootstrap.
> 
> Testing: tier1-4
> 
> Thanks!
> 
> /Claes


More information about the hotspot-runtime-dev mailing list