[8] RFR: Shenandoah: JNI weak roots are always cleared before Full GC mark

[Zhengyu Gu]

Hi Kornel,

On 9/7/20 1:08 PM, Kornel Pal wrote:
> Hi Zhengyu,
> 
> Thank you for getting back to me and preparing the change this quickly.
> 
> I'm looking forward to see the change in the nightly builds.
> 

Thanks for reporting.

The fix is reviewed and ready for pushing.

However, current shenandoah/jdk8 repo is closed for upcoming 
integration. I will push the change once the repo is reopen.

-Zhengyu

> Thank you,
> Kornel
> 
> On 2020-08-28. 14:36, Zhengyu Gu wrote:
>> Hi Kornel,
>>
>> Yes, I will RFR to backout this workaround.
>>
>> Thanks,
>>
>> -Zhengyu
>>
>> On 8/28/20 12:52 AM, Kornel Pal wrote:
>>> Hi Aleksey,
>>>
>>> Thank you very much for all the dedicated work on Shenandoah GC, it 
>>> really is useful.
>>>
>>> I wonder whether this workaround could safely be removed, now that 
>>> JDK-8248041 was backported to 8u.
>>>
>>> Thank you,
>>> Kornel
>>>
>>> On 2020-06-22 19:39, Aleksey Shipilev wrote:
>>>> Amir found the crash with Cassandra that started to happen recently 
>>>> with Shenandoah 8u. It
>>>> reproduces more or less reliably with starting Cassandra 3.11.4 with 
>>>> fastdebug build and
>>>> -XX:ShenandoahGCMode=passive -XX:-ShenandoahDegeneratedGC.
>>>>
>>>> This is a symptom of a larger problem (see JDK-8248041), but 8u it 
>>>> is catastrophic due to clearing
>>>> of the weak roots.
>>>>
>>>> Fix:
>>>> https://cr.openjdk.java.net/~shade/shenandoah/8u-fullgc-jni-weak/webrev.01/ 
>>>>
>>>>
>>>> Testing: hotspot_gc_shenandoah; Cassandra tests
>>>>
>>>
>>
>