[aarch64-port-dev ] Question: JNI critical native support is not fully implemented for AArch64?

[Robbin Ehn]

Hi,

The lazy part of this, not taking the GC locker, instead letting the VM thread
do it while safepointing causes some troubles. The GC lock can be locked _after_
the safepoint have started, GC must thus abort the safepoint.
It relies on the _suspend_flags which we are trying to remove (in favor of
handshakes), so we need to do something about it.
If this is something the vm needs I would rather see it in Panama with a
different solution.
I need to test the performance difference, but if it's small I would suggest
removing it completely.

So before anyone starts implementing/spreading the use of this can we have some 
time to evaluate?

Thanks, Robbin

On 12/19/19 11:10 AM, Andrew Haley wrote:
> On 12/19/19 7:42 AM, Pengfei Li wrote:
>> While I also see someone saying that this feature is designed for internal use only and there is no public specification for it[2]. And I don't know if there is any real workload currently using it. So could anyone tell me if "Critical Natives" is a sort of "public feature" of OpenJDK? And is there much value to complement the implementation for AArch64?
> 
> For anything to be fixed we'd need to write some code that uses critical natives,
> at least to test it. It'd be interesting to see what this could be used for. It
> might be useful.
>