[13] RFR(S) 8223794: applications/kitchensink/Kitchensink.java crash bad oop with Graal

[Vladimir Kozlov]

Thank you, Dean

Vladimir K

On 6/19/19 4:16 PM, dean.long at oracle.com wrote:
> Looks good.
> 
> dl
> 
> On 6/19/19 2:53 PM, Vladimir Kozlov wrote:
>> http://cr.openjdk.java.net/~kvn/8223794/webrev.03/
>> https://bugs.openjdk.java.net/browse/JDK-8223794
>>
>> Big thanks to Erik O. for pointing the possible cause of this problem and fix suggestion:
>>
>> "The JVMCINMethodData::get_nmethod_mirror() function in JVMCI is used to read the nmethod mirror from the nmethod oop 
>> section. However, the nmethod oops are weak, and not kept alive by GC. Yet strong references are created to the 
>> returned oop in places, without going through a keep-alive barrier. This could result in creating edges to dead oops, 
>> resulting in a crash eventually."
>>
>> I added nmethod::oop_at_phantom() method to notify GC that oop should be kept alive. It is used in JVMCI where it 
>> needs strong reference. I originally named new method and argument 'strong' but Erik pointed that:
>>
>> "Maybe s/strong/phantom/g for the new method name and parameters. Otherwise it is the other way around, i.e. when 
>> strong is true in the current webrev01, we actually read it as a phantom oop ref, and conversely when strong is false, 
>> we do read it as a strong oop ref. That is a bit confusing."
>>
>> Running Kitchensink without fix triggers the failure intermittently. With the fix I did not see the failure.
>>
>> Thanks,
>> Vladimir
>