[10] RFR(M): 8176506: C2: loop unswitching and unsafe accesses cause crash

[Volker Simonis]

Hi,

the latest webrev was fine for ppc64/s390 (and also Sparc). No changes are
needed for  these platforms because they all produced SIGILL already.

So thumbs up from my side!

Regards,
Volker

Roland Westrelin <rwestrel at redhat.com> schrieb am Fr. 2. Juni 2017 um 21:25:

>
> > Our internal testing of webrev.04 version of changes passed.
>
> Thanks.
>
> > http://cr.openjdk.java.net/~roland/8176506/webrev.04/
> >
> > One thing. I don't see changes for ppc64/s390x which Volker promised to
> > provide. Do we need them in this latest version? What about SPARC?
>
> Volker said:
>
> > I've also looked at your change and I think it works fine on
> > ppc64/s390x because we've always implemented the HaltNode by issuing a
> > trap instructions. So with your change (and your modified test) we
> > will just crash in the generates C2 method which I think is fine:
> >
> > #  SIGTRAP (0x5) at pc=0x00003fff47b12bd8, pid=19883, tid=19964
> >
> > J 215 c2 TestMaybeNullUnsafeAccess.test2(Ljava/lang/Object;)I (34
> > bytes) @ 0x00003fff47b12bd8 [0x00003fff47b12b80+0x0000000000000058]
> > j  TestMaybeNullUnsafeAccess.main([Ljava/lang/String;)V+49
> > v  ~StubRoutines::call_stub
>
> So my understanding is that nothing needs to be done on ppc.
>
> On sparc, doesn't the current implementation of Halt:
>
> __ illtrap(0);
>
> cause a SIGILL as well?
>
> Roland.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20170602/43d85caa/attachment.html>