On Tue, 3 Sep 2024 04:52:08 GMT, Amit Kumar <amitku...@openjdk.org> wrote:
>> As discussed in the JBS issue:
>>
>> FFM upcall stubs embed a `Method*` of the target method in the stub. This
>> `Method*` is read from the `LambdaForm::vmentry` field associated with the
>> target method handle at the time when the upcall stub is generated. The MH
>> instance itself is stashed in a global JNI ref. So, there should be a
>> reachability chain to the holder class: `MH (receiver) -> LF (form) ->
>> MemberName (vmentry) -> ResolvedMethodName (method) -> Class<?> (vmholder)`
>>
>> However, it appears that, due to multiple threads racing to initialize the
>> `vmentry` field of the `LambdaForm` of the target method handle of an upcall
>> stub, it is possible that the `vmentry` is updated _after_ we embed the
>> corresponding `Method`* into an upcall stub (or rather, the latest update is
>> not visible to the thread generating the upcall stub). Technically, it is
>> fine to keep using a 'stale' `vmentry`, but the problem is that now the
>> reachability chain is broken, since the upcall stub only extracts the target
>> `Method*`, and doesn't keep the stale `vmentry` reachable. The holder class
>> can then be unloaded, resulting in a crash.
>>
>> The fix I've chosen for this is to mimic what we already do in
>> `MethodHandles::jump_to_lambda_form`, and re-load the `vmentry` field from
>> the target method handle each time. Luckily, this does not really seem to
>> impact performance.
>>
>> <details>
>> <summary>Performance numbers</summary>
>> x64:
>>
>> before:
>>
>> Benchmark Mode Cnt Score Error Units
>> Upcalls.panama_blank avgt 30 69.216 ± 1.791 ns/op
>>
>>
>> after:
>>
>> Benchmark Mode Cnt Score Error Units
>> Upcalls.panama_blank avgt 30 67.787 ± 0.684 ns/op
>>
>>
>> aarch64:
>>
>> before:
>>
>> Benchmark Mode Cnt Score Error Units
>> Upcalls.panama_blank avgt 30 61.574 ± 0.801 ns/op
>>
>>
>> after:
>>
>> Benchmark Mode Cnt Score Error Units
>> Upcalls.panama_blank avgt 30 61.218 ± 0.554 ns/op
>>
>> </details>
>>
>> As for the added TestUpcallStress test, it takes about 800 seconds to run
>> this test on the dev machine I'm using, so I've set the timeout quite high.
>> Since it runs for so long, I've dropped it from the default `jdk_foreign`
>> test suite, which runs in tier2. Instead the new test will run in tier4.
>>
>> Testing: tier 1-4
>
> RuntimeAddress will not work for s390x. @JornVernee would you please apply
> these changes.
>
> Thanks @TheRealMDoerr for pointing it out. I will create a JBS issue for
> implementing `resolve_global_jobject`.
>
>
> diff --git a/src/hotspot/cpu/s390/upcallLinker_s390.cpp
> b/src/hotspot/cpu/s390/upcallLinker_s390.cpp
> index 1b07522858f..8baad40a519 100644
> --- a/src/hotspot/cpu/s390/upcallLinker_s390.cpp
> +++ b/src/hotspot/cpu/s390/upcallLinker_s390.cpp
> @@ -216,10 +216,11 @@ address UpcallLinker::make_upcall_stub(jobject
> receiver, Symbol* signature,
> arg_shuffle.generate(_masm, shuffle_reg, abi._shadow_space_bytes,
> frame::z_jit_out_preserve_size);
> __ block_comment("} argument_shuffle");
>
> - __ block_comment("load target {");
> + __ block_comment("load_target {");
> __ load_const_optimized(Z_ARG1, (intptr_t)receiver);
> - __ call(RuntimeAddress(StubRoutines::upcall_stub_load_target())); // load
> taget Method* into Z_method
> - __ block_comment("} load target");
> + __ load_const_optimized(call_target_address,
> StubRoutines::upcall_stub_load_target());
> + __ call(call_target_address); // load taget Method* into Z_method
> + __ block_comment("} load_target");
>
> __ z_lg(call_target_address, Address(Z_method,
> in_bytes(Method::from_compiled_offset())));
> __ call(call_target_address);
@offamitkumar thanks, I've added those changes.
-------------
PR Comment: https://git.openjdk.org/jdk/pull/20479#issuecomment-2326422537
