[aarch64-port-dev ] RFR(XS): 8u fails on assert on jtreg compiler/uncommontrap/TestDeoptOOM.java

[Felix Yang]

Hi,

    +#ifdef ASSERT
    +    // If there is exception we go to interpreter before code below clear
    +    // last_sp of current frame
    +    __ str(zr, Address(rfp,
frame::interpreter_frame_last_sp_offset * wordSize));
    +#endif


    This works for a debug build, should we do the same thing for a release
build?
    And it looks that the error message is not accurate: call_VM_leaf_base
should be changed into call_VM_base.
    Maybe we should correct that in your patch.

Thanks,
Felix


On 12 August 2017 at 02:37, Anton Kozlov <akozlov at azul.com> wrote:

>
>
> On 11.08.2017 20:52, Andrew Haley wrote:
> > I'm baffled by how we missed it and why it's only a problem for JDK 8.
>
> It breaks "NULL in last_sp" invariant of top frame in non-fatal way,
> making it visible only in debug builds.
> And this part is rewritten in jdk9 when JVMCI introduced. Starging with 9,
> everything restored prior checking for exception. 8 is trying to restore as
> little as possible before checking for exception and going to handler.
> The 8's non-exceptional path of is OK.
>
> PS. I have no push rights, so I ask someone to push
>
> Thanks,
> Anton
>