RFR(S): 8231949: [PPC64, s390]: Make async profiling more reliable

[Doerr, Martin]

Hi Richard,

>     I wonder if that complexity and risk of crashing for interpreter frames is
> necessary at all...
>     Well, maybe if we've got some hot, but not compilable method.
>     Personally I'd prefer a safe solution. Profiling should _never_ crash the vm.
This sounds like a more fundamental change which includes shared code changes, right?

I'd like to backport the fix for crashes in pd_get_top_frame_for_profiling, so maybe we should just avoid accessing memory outside of the expected data structures in this backportable fix and make the more fundamental change in jdk14/15?

Best regards,
Martin


> -----Original Message-----
> From: Haug, Gunter <gunter.haug at sap.com>
> Sent: Freitag, 11. Oktober 2019 14:14
> To: Reingruber, Richard <richard.reingruber at sap.com>; Doerr, Martin
> <martin.doerr at sap.com>; hotspot-runtime-dev at openjdk.java.net;
> Schmidt, Lutz <lutz.schmidt at sap.com>; Stuefe, Thomas
> <thomas.stuefe at sap.com>
> Subject: Re: RFR(S): 8231949: [PPC64, s390]: Make async profiling more
> reliable
> 
> Hi Richard,
> 
> That's a really good idea, not only for PPC but even more so for s390. On PPC
> we always have a valid SP, which I think doesn't hold on s390.
> Otherwise +1 for Martin's change.
> 
> Gunter // not a Reviewer
> 
> On 11.10.19, 14:07, "Reingruber, Richard" <richard.reingruber at sap.com>
> wrote:
> 
>     Hi Martin,
> 
>     your changes look like an improvement to me.
> 
>     I wonder if that complexity and risk of crashing for interpreter frames is
> necessary at all...
>     Well, maybe if we've got some hot, but not compilable method.
>     Personally I'd prefer a safe solution. Profiling should _never_ crash the vm.
> 
>     For instance we could write the currently interpreted method and a recent
> bcp to the thread.
> 
>     Richard.  // not yet Reviewer
> 
>     -----Original Message-----
>     From: hotspot-runtime-dev <hotspot-runtime-dev-
> bounces at openjdk.java.net> On Behalf Of Doerr, Martin
>     Sent: Montag, 7. Oktober 2019 16:57
>     To: hotspot-runtime-dev at openjdk.java.net; Haug, Gunter
> <gunter.haug at sap.com>; Schmidt, Lutz <lutz.schmidt at sap.com>; Stuefe,
> Thomas <thomas.stuefe at sap.com>
>     Subject: [CAUTION] RFR(S): 8231949: [PPC64, s390]: Make async profiling
> more reliable
> 
>     Hi,
> 
>     I'd like to make pd_get_top_frame_for_profiling more reliable on PPC64
> and s390 as we have seen crashes in this function.
> 
>     See bug for more details:
>     https://bugs.openjdk.java.net/browse/JDK-8231949
> 
>     Webrev:
> 
> http://cr.openjdk.java.net/~mdoerr/8231949_get_top_frame_for_profiling/
> webrev.00/
> 
>     Please review.
> 
>     Best regards,
>     Martin
> 
>