RFR(M): 8202080: Introduce ordering semantics for Atomic::add

Robbin Ehn robbin.ehn at oracle.com
Fri Apr 20 19:47:50 UTC 2018


Hi Martin,

Thanks for tackling this, just had a peak at it.

/Robbin

On 2018-04-20 18:04, Doerr, Martin wrote:
> Hi,
> 
> JDK-8195099 "Concurrent safe-memory-reclamation mechanism" introduced a usage of Atomic::add for which we're not using the correct memory barriers on PPC64 and s390.
> Some more description can be found in the bug: https://bugs.openjdk.java.net/browse/JDK-8202080
> 
> We can fix this by the following change:
> http://cr.openjdk.java.net/~mdoerr/8202080_atomic_add/webrev.00/
> 
> Please note that I only implemented the platform code for linux x86, PPC64 and s390 so far. The remaining ones should be trivial and I'll take care of them after I got some feedback.
> I can also do the same change for other atomic functions like sub, inc and dec if you like.
> 
> Intention of this change is only to fix this new code and possibly other future enhancements, not to change the behavior of any older usages of Atomic::add.
> 
> Feedback is welcome.
> 
> Best regards,
> Martin
> 


More information about the hotspot-runtime-dev mailing list