RFR(sh/jdk8): [backport] 8232992: Shenandoah: Implement self-fixing interpreter LRB

[Roman Kennke]

Am 01.12.20 um 11:53 schrieb Aleksey Shipilev:
> On 11/26/20 9:00 PM, Roman Kennke wrote:
>> This backports the issue:
>> https://bugs.openjdk.java.net/browse/JDK-8257180
> 
> More like
>    https://bugs.openjdk.java.net/browse/JDK-8232992
> 
>> http://cr.openjdk.java.net/~rkennke/JDK-8232992-jdk8/webrev.01/
> 
> *) src/cpu/x86/vm/macroAssembler_x86.cpp
> 
> Why do you need these?
> 
> 5790   Register result_dst = dst;
> 
> 5810   Register result_dst = dst;

We don't need them. This is a leftover from when I had all the 
Shenandoah code crammed in MA::load_heap_oop(). Good catch.

> *) src/cpu/x86/vm/shenandoahBarrierSetAssembler_x86.cpp
> 
> save/restore_xmm_registers are cherry-picked from some other changeset?

Yes, as is basically the whole backport-implementation.

> I ask because it would break x86_32 without UseSSE=2 (which I am fixing 
> upstream now, and would backport). It is important to record that we 
> backported this change, so that we known it affects 8-shenandoah after 
> this lands.

It was from this original change:
https://bugs.openjdk.java.net/browse/JDK-8250588

Roman