[aarch64-port-dev ] RFR: Bulk integration from Shenandoah

[Roman Kennke]

Hi Andrew,

> On 06/03/2020 17:49, Aleksey Shipilev wrote:
>> On 3/6/20 6:22 PM, Roman Kennke wrote:
>>> Changeset:
>>> http://cr.openjdk.java.net/~rkennke/bulk-integration-2020-03-06/changesets.01
>>>
>>> Full webrev:
>>> http://cr.openjdk.java.net/~rkennke/bulk-integration-2020-03-06/webrev.01/
>>
>> This looks good to me.
>>
>> Andrews need to concur before we can push :)
>>
> 
> These would be easier to review if the cumulative effect of the changes
> could be more easily seen. I was going to suggest including the merge
> changeset, but this doesn't appear to be a merge.

Well yeah. The cumulative effect of the patches is exactly what you see
here :-)

> Why are we not just
> pushing these to aarch64/shenandoah-jdk8u as they are developed, rather
> than waiting to do them all in bulk? Particularly as it's relatively
> late in the development process for the April CPU.

I was thinking why are we integrating this to aarch64/jdk8u-shenandoah
at all? We could also use shenandoah/jdk8 as integration forest.
shenandoah/jdk8 has aarch64/jdk8 as upstream, which in turn has jdk8u as
upstream. Dragging everything into aarch64/jdk8u-shenandoah is just a
somewhat unnecessary extra step.

> Most of these seem to be Shenandoah-specific. The one that catches my
> eye is:
> 
> 8229919: Support JNI Critical functions in object pinning API on x86_32
> platforms
> 
> This seems like it will introduce a behavioural deviation for those who
> are not using Shenandoah, in comparison to vanilla 8u. Can you assure me
> this is safe?

Yes. This code is only entered when
Universe::heap()->supports_object_pinning() returns true, and Shenandoah
is currently the only implementation that does this. In the x86_32
counterparts I actually put assert(UseShenandoahGC) in a couple of
relevant places, I forgot this in x86_32. Sorry.

> I do wonder if patches like this should be backported to upstream 8u &
> 11u, ahead of any proposed Shenandoah merge into those upstreams.

I was considering this. But it would be dead code upstream. So no.

The integration has already been pushed. I hope that's ok.

Roman