RFR(xxs): 8209495: NMethodSweeper::sweep_code_cache cause severe delays

Erik Österlund erik.osterlund at oracle.com
Wed Oct 31 15:12:23 UTC 2018


Hi Robbin,

Looks good.

Thanks,
/Erik

On 2018-10-31 15:38, Robbin Ehn wrote:
> Hi please review,
>
> Issue:
> https://bugs.openjdk.java.net/browse/JDK-8209495
>
> Code:
> diff -r 875999ffe38d src/hotspot/share/runtime/sweeper.cpp
> --- a/src/hotspot/share/runtime/sweeper.cpp    Wed Oct 31 14:07:15 
> 2018 +0100
> +++ b/src/hotspot/share/runtime/sweeper.cpp    Wed Oct 31 15:28:39 
> 2018 +0100
> @@ -390,1 +390,2 @@
> -  if (SafepointSynchronize::is_synchronizing()) {
> +  JavaThread* thread = JavaThread::current();
> +  if (SafepointMechanism::poll(thread)) {
> @@ -396,1 +396,0 @@
> -    JavaThread* thread = JavaThread::current();
>
>
> Since we only do the ThreadBlockInVM if there is a safepoint,
> this have a bad effect on handshakes.
>
> RunThese30M works with much lower timeout, tenths of millis.
>
> Thanks, Robbin



More information about the hotspot-compiler-dev mailing list