[8u] RFR 8191227: Unsafe handle resolution in ConstantOopWriteValue::write_on() / print_on() and LIR_Assembler::jobject2reg()

[Aleksey Shipilev]

On 2/20/20 5:22 AM, Andrew Hughes wrote:
> On 19/02/2020 11:51, Aleksey Shipilev wrote:
>> Original fix:
>>   https://bugs.openjdk.java.net/browse/JDK-8191227
>>   https://hg.openjdk.java.net/jdk/jdk/rev/bb957f109a1f
>>
>> The patch applies after reshuffling and fuzz. We ship parts of this fix in shenandoah/jdk8u:
>>   https://mail.openjdk.java.net/pipermail/shenandoah-dev/2020-February/011434.html
>>
>> I would appreciate reviews w.r.t. its bug tail. It seems to be followed up by JDK-8193699, which
>> introduced numerous failures in itself. I don't think this fix individually causes those troubles.
>>
>> 8u webrev:
>>   http://cr.openjdk.java.net/~shade/8191227/webrev.8u.01/
> 
> Patch looks fine, just some context differences.
> 
> I assume the bug tail is more about JDK-8167372: "Add code to check for
> getting oops while thread is in native" (JDK-8193699 "AArch64: Fails to
> build after 8167372" is a response to that, and for AArch64, which
> currently isn't supported in 8u). I can see the merits in backporting
> that, but I'd wait until the 8u262 cycle so we maximise the time to
> shake out any issues.

Yes. I wanted someone to arrive to the same conclusion.

> This patch seems pretty safe, given the only product build change is to
> introduce ThreadInVMfromUnknown tiv in
> ConstantOopWriteValue::print_on(outputStream* st). It's effectively an
> instance of the bug JDK-8167372 aims to make easier to catch.

Thanks! I added the tags.

-- 
Thanks,
-Aleksey