8218401: WRONG_PHASE: vmTestbase/nsk/jvmti test crash

[serguei.spitsyn at oracle.com]

Hi Daniil,

No chance for it if you lock a raw monitor in the event callbacks.

Thanks,
Serguei

On 3/20/19 14:58, Daniil Titov wrote:
> Hi Serguei,
>
> I think that just disabling event notifications inside VMDeath callback still leaves a small window for VMDeath callback being called after the classload callback is called (or about being called)  but before it enters a raw monitor. Thus I decided to follow your original suggestion  and restore volatile modifier for callbacksEnabled.  Please review a new version of the patch.
>
> Webrev: http://cr.openjdk.java.net/~dtitov/8218401/webrev.03/
> Bug : https://bugs.openjdk.java.net/browse/JDK-8218401
>
> Thanks!
>
> Best regards,
> Daniil
>
>
>
> From: <serguei.spitsyn at oracle.com>
> Organization: Oracle Corporation
> Date: Tuesday, March 19, 2019 at 7:17 PM
> To: Daniil Titov <daniil.x.titov at oracle.com>, OpenJDK Serviceability <serviceability-dev at openjdk.java.net>, Jean Christophe Beyler <jcbeyler at google.com>
> Subject: Re: 8218401: WRONG_PHASE: vmTestbase/nsk/jvmti test crash
>
> Hi Daniil,
>
> I'd keep the volatile modifier for callbacksEnabled to disable compiler optimizations.
> Otherwise, looks good to me.
>
>
> Another approach could be to disable event notifications in VMDeath callback with:
>    SetEventNotificationMode(JVMTI_DISABLE, JVMTI_EVENT_CLASS_LOAD, NULL);
>    SetEventNotificationMode(JVMTI_DISABLE, JVMTI_EVENT_BREAKPOINT, NULL);
>    . . .
>
> Thanks,
> Serguei
>
> On 3/18/19 6:58 PM, Daniil Titov wrote:
> Hi Serguei and JC,
>
> Please review a new version of the fix that locks a monitor across the callbacks, as Serguei suggested.
>
> Webrev: http://cr.openjdk.java.net/~dtitov/8218401/webrev.02/
> Bug: https://bugs.openjdk.java.net/browse/JDK-8218401
>
> Thanks!
> --Daniil
>
>
> On 3/18/19, 9:47 AM, mailto:serguei.spitsyn at oracle.com mailto:serguei.spitsyn at oracle.com wrote:
>
>      Hi Daniil,
>      
>      The JVMTI phase can change in the middle of callback work after the
>      check you added.
>      I'd suggest to lock a raw monitor across the callbacks to make them atomic.
>      
>      Thank you for taking care about this issue!
>      
>      Thanks,
>      Serguei
>      
>      
>      
>      On 3/15/19 16:08, Daniil Titov wrote:
>      > Please review the change that fixes 3 tests that intermittently fail with JVMTI_ERROR_WRONG_PHASE error.
>      >
>      > The problem here is that the callbacks these tests enable keep processing events and perform JVMTI calls after VM is terminated. The fix makes these test listen for VMDeath event and  quick return from the callbacks after VMDeath event is received.
>      >
>      > Webrev: http://cr.openjdk.java.net/~dtitov/8218401/webrev.01/
>      > Bug: https://bugs.openjdk.java.net/browse/JDK-8218401
>      >
>      > Thanks!
>      > -Daniil
>      >
>      >
>      
>      
>
>
>
>
>
>