RFR (L) 8154580: Save mirror in interpreter frame to enable cleanups of CLDClosure

[Coleen Phillimore]

Chris, thank you for reviewing this.

On 4/19/16 4:35 PM, Christian Thalinger wrote:
>> On Apr 19, 2016, at 9:50 AM, Coleen Phillimore <coleen.phillimore at oracle.com> wrote:
>>
>> Summary: GC walks the mirror using OopClosure rather than using CLDClosure in oops_interpreted_do()
>>
>> See bug for more description and justification.  The changes are large but very redundant.  The main change is in TemplateInterpreterGenerator::generate_fixed_frame().
> +  // Save oop Mirror (with padding)
> +  __ load_mirror(rscratch1, rmethod);
>
> +  // get mirror and store it in the frame so that this Method* is never
> +  // reclaimed while it's running.
> +  Register mirror = LcpoolCache;
> +  __ load_mirror(mirror, Method);
>
> +  // Push the mirror so this method isn't collected
> +  __ load_mirror(rdx, rbi);
>
> Please use the same comment on all platforms.

Yes, that's inconsistent.  I changed all platforms to have this comment:

   // Get mirror and store it in the frame as GC root for this Method*

>> open webrev at http://cr.openjdk.java.net/~coleenp/8154580.01/webrev
>> bug link https://bugs.openjdk.java.net/browse/JDK-8154580
> src/share/vm/runtime/frame.cpp
>
>       // The method pointer in the frame might be the only path to the method's
>       // klass, and the klass needs to be kept alive while executing. The GCs
>       // don't trace through method pointers, so typically in similar situations
>       // the mirror or the class loader of the klass are installed as a GC root.
>
> -    // To minimize the overhead of doing that here, we ask the GC to pass down a
> -    // closure that knows how to keep klasses alive given a ClassLoaderData.
> -    cld_f->do_cld(m->method_holder()->class_loader_data());
> -  }
> -
> -  if (m->is_native() PPC32_ONLY(&& m->is_static())) {
> -    f->do_oop(interpreter_frame_temp_oop_addr());
> -  }
> +  // And it is here too.
> +  f->do_oop(interpreter_frame_mirror_addr());
>
> That comment is kinda funny now.  It still hints at the old-way of doing things but “it is here too”.

I reworded it as:

   // The method pointer in the frame might be the only path to the method's
   // klass, and the klass needs to be kept alive while executing. The GCs
   // don't trace through method pointers, so the mirror of the method's 
klass
   // is installed as a GC root.
   f->do_oop(interpreter_frame_mirror_addr());

>> Tested with hotspot-runtime-nightly and gc-nightly tests.
>>
>> Need testing with ppc and aarch64 open code.  I implemented the changes but I can't test them.
> One obvious bug is that you copied the __ as well:
>
> +void MacroAssembler::load_mirror(Register dst, Register method) {
> +  const int mirror_offset = in_bytes(Klass::java_mirror_offset());
> +  __ ldr(dst, Address(rmethod, Method::const_offset()));
> +  __ ldr(dst, Address(dst, ConstMethod::constants_offset()));
> +  __ ldr(dst, Address(dst, ConstantPool::pool_holder_offset_in_bytes()));
> +  __ ldr(dst, Address(dst, mirror_offset));
> +}
>
> Other than that it looks fine to me.  Nothing obvious stood out.

Oh, yes, this was wrong on both ppc and aarch64.   I hope someone that 
has access to this platforms can test this patch.

Thank you!
Coleen

>
>> Thanks,
>> Coleen