RFR(S): JDK-8139868: CMSScavengeBeforeRemark broken after JDK-8134953

[Jesper Wilhelmsson]

Looks good!
/Jesper

Den 19/10/15 kl. 10:24, skrev Bengt Rutisson:
>
> Hi everyone,
>
> Could I have a couple of reviews for this fix for a GC id issue?
>
> http://cr.openjdk.java.net/~brutisso/8139868/webrev.00/index.html
> https://bugs.openjdk.java.net/browse/JDK-8139868
>
>  From the CR:
>
> Running with -XX:+CMSScavengeBeforeRemark will do an extra call to
> GenCollectedHeap::do_collection() inside CMSCollector::checkpointRootsFinal().
>
> Inside CMSCollector::checkpointRootsFinal() we already have a valid GC id set
> up, but the call to GenCollectedHeap::do_collection() will create a new one
> using a GCIdMark, which means that when it returns the GC id is set to undefined.
>
> It is correct that GenCollectedHeap::do_collection() creates a new GC id since
> we are doing one extra GC there. But it should not reset the GC id to undefined
> when it is done.
>
>
> The fix is to use a GCIdMarkAndRestore inside GenCollectedHeap::do_collection()
> instead. This will preserve the "outer" GC id for the remark. But since the two
> other calls to GenCollectedHeap::do_collection() are for distinct GCs we can not
> assume that there always is an "outer" GC id to preserve. Thus, the constructor
> for GCIdMarkAndRestore has to use current_raw() instead of current() to store
> the previous GC id.
>
> Thanks,
> Bengt