RFR: [8u, 9, 10] Cherry-pick SATB/KAL fixes for G1 and Shenandoah

[Roman Kennke]

Am 13.04.2018 um 16:55 schrieb Aleksey Shipilev:
> I did the sweep for missing SATB/Keepalive barriers in Shenandoah backports, basically enumerating
> all the places where current JDK 11 have them, and cross-referencing if they are missing in the
> backports.
> 
> These are the fixes:
> 
> === sh/jdk10:
> 
> http://cr.openjdk.java.net/~shade/shenandoah/audit-satb/10/webrev.01/
> 
> 6b510cb0f14f: Cherry pick JDK-8187577: JVM crash during gc doing concurrent marking
> cds-unsharable: Fix assert in ConstantPool::restore_unshareable_info (see also JDK-8194741)
> 
> I am going to request the backport for JDK-8187577 to JDK 10 a bit later. The assert modifies the
> code coming from JDK-8194741 to enable Shenandoah -- it is not required in sh/jdk, because Access
> API does the right thing.
> 
> === sh/jdk9:
> 
> http://cr.openjdk.java.net/~shade/shenandoah/audit-satb/9/webrev.01/
> 
> 6b510cb0f14f: Cherry pick JDK-8187577: JVM crash during gc doing concurrent marking
> 
> There seem to be no chance for JDK-8187577 to come to upstream JDK 9, because it is EOLed.
> 
> === sh/jdk8u:
> 
> http://cr.openjdk.java.net/~shade/shenandoah/audit-satb/8u/webrev.01/
> 
> 6c280bbde146: Cherry-pick JDK-8173013: JVMTI tagged object access needs G1 pre-barrier
> a696583f5ddb: Cherry-pick JDK-8165489: Missing G1 barrier in Unsafe_GetObjectVolatile
> 6b510cb0f14f: Cherry-pick JDK-8187577: JVM crash during gc doing concurrent marking
> 
> JDK-8187577 is going to come with the 8u merge at some point. I am going to request backports for
> JDK-8173013 and JDK-8165489 to 8u-dev a bit later, so they also come via the regular channel.
> 
> All patches pass hotspot_gc_shenandoah {fastdebug|release}.

Patches look good.

I wonder if we want to backport our (interim) keep_alive_barrier() API
in BarrierSet to JDK8?

Roman