On 2/18/19 7:36 AM, Nick Gasson (Arm Technology China) wrote:
> http://cr.openjdk.java.net/~ngasson/8209413/webrev.03/
>
> It contains your suggested change below, as well as an extra check in
> the AARCH64Frame two-argument constructor that the PC from SP[-1] is
> valid. Also changed SharedRuntime::generate_native_wrapper and
> TemplateInterpreterGenerator::generate_native_entry to save an accurate
> return PC in the frame anchor.
That looks right, thanks. OK.
> There's a FIXME in set_last_Java_frame like this:
>
> } else {
> // FIXME: This is almost never correct. We should delete all
> // cases of set_last_Java_frame with last_java_pc=NULL and use the
> // correct return address instead.
> adr(scratch, pc());
> }
>
> Should we do this now? There's only two places I can see:
> StubGenerator::generate_throw_exception and the set_last_Java_frame
> overload that takes a Label.
I hate cruft like this, but it's not actually a bug, and any change in this
delicate area has a non-zero probability of breaking something. On the other
hand, it would be a very low-risk change.
--
Andrew Haley
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671
