Hi Serguei,
Thank you for raising the bug.
I had a chat with one of my colleagues who could recreate it, and
it's probably related to the handshaking that is done in the
particular scenario. So with the JCK harness:
com.sun.jck.lib.ExecJCKTestOtherJVMCmd
LD_LIBRARY_PATH=/_javatest_/_lib_/_jck_/jck8b/natives/linux_x86-64
/projects/_jck_/jdwp/j2sdk-image/bin/java -Xdump:system:none
-Xdump:system:events=_gpf_+abort+_traceassert_+_corruptcache_
-Xdump:snap:none
-Xdump:snap:events=_gpf_+abort+_traceassert_+_corruptcache_
-Xdump:java:none
-Xdump:java:events=_gpf_+abort+_traceassert_+_corruptcache_
-Xdump:heap:none
-Xdump:heap:events=_gpf_+abort+_traceassert_+_corruptcache_
-_Xfuture_
-agentlib:jdwp=server=y,transport=dt_socket,address=_localhost_:35000,suspend=y
-_classpath_ /_javatest_/_lib_/_jck_/JCK8b-b03/JCK-runtime-8b/classes
-Djava.security.policy=/_javatest_/_lib_/_jck_/JCK8b-b03/JCK-runtime-8b/_lib_/jck.policy
javasoft.sqe.jck.lib.jpda.jdwp.DebuggeeLoader -_waittime_=600
-msgSwitch=ub1604x64vm10:38636
-componentName=*ArrayReference.GetValues.getvalues002*
Note that the JCK test harness starts the target process, attaches to
it, and sends the resume command
in a very short time with no handshaking.
That may not help..but hopefully helps explain things a bit? It's the
timing of the resume command during the test that is crucial,
resuming before the VM initialization is complete will trigger it.
Thanks
Andrew
Andrew Leonard
Java Runtimes Development
IBM Hursley
IBM United Kingdom Ltd
Phone internal: 245913, external: 01962 815913
internet email: andrew_m_leon...@uk.ibm.com
From: "serguei.spit...@oracle.com" <serguei.spit...@oracle.com>
To: Andrew Leonard <andrew_m_leon...@uk.ibm.com>
Cc: serviceability-dev@openjdk.java.net
Date: 11/04/2018 09:57
Subject: Re: RFR: Fix race condition in jdwp
------------------------------------------------------------------------
Hi Andrew,
I've filed the bug:
_https://bugs.openjdk.java.net/browse/JDK-8201409_
<https://urldefense.proofpoint.com/v2/url?u=https-3A__bugs.openjdk.java.net_browse_JDK-2D8201409&d=DwMDaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=NaV8Iy8Ld-vjpXZFDdTbgGlRTghGHnwM75wUPd5_NUQ&m=nYdDNT1pdP7gQK1JviPJ6zIxLWcwe7R-azh_NnvI0Ok&s=z0_3fj5wG3cofrwXExX0jmnKvUeBGxjAOsYE1Gz6xCg&e=>
Also, this is a webrev with your patch:
_http://cr.openjdk.java.net/~sspitsyn/webrevs/2018/8201409-jdwp-initsync.ibm.1/_
<https://urldefense.proofpoint.com/v2/url?u=http-3A__cr.openjdk.java.net_-7Esspitsyn_webrevs_2018_8201409-2Djdwp-2Dinitsync.ibm.1_&d=DwMDaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=NaV8Iy8Ld-vjpXZFDdTbgGlRTghGHnwM75wUPd5_NUQ&m=nYdDNT1pdP7gQK1JviPJ6zIxLWcwe7R-azh_NnvI0Ok&s=8LH2-PYUqJbVkawhCgmuD5106lOuFKI-jAVhCpB_tYY&e=>
I agree that creating a standalone test is tricky here.
I've added usleep(10000) into the eventHelper_reportVMInit()
and ran the JTreg com/sun/jdi tests with my JDK build.
However, none of the tests failed with the failure mode you described.
So that I'm puzzled a little bit.
I suspect that some specific debugLoop commands were used in your
scenario.
It is still possible that I've missed something here.
Will try to double check everything.
Thanks,
Serguei
On 4/11/18 01:29, Andrew Leonard wrote:
Thanks Serguei,
I terms of a standalone testcase it is quite tricky, as due to the
nature of the issue which took a lot of investigation to solve it's
very timing dependent and will only occur randomly. It can be forced
as I indicated below by adding a "sleep" in the VMInit report code
but that's not a testcase, however the issue was originally found in
our JCK testing for IBMJava8, testcase test.jck8b.runtime.vm.jdwp,
but again only happened intermittently. Sort of like "performance"
type issues we're not always going to be able to create a testcase
that will always "fail" if the fix is not present.
Your thoughts?
Cheers
Andrew
Andrew Leonard
Java Runtimes Development
IBM Hursley
IBM United Kingdom Ltd
Phone internal: 245913, external: 01962 815913
internet email: _andrew_m_leon...@uk.ibm.com_
<mailto:andrew_m_leon...@uk.ibm.com>
From: _"serguei.spit...@oracle.com"_
<mailto:serguei.spit...@oracle.com>_<serguei.spit...@oracle.com>_
<mailto:serguei.spit...@oracle.com>
To: Andrew Leonard _<andrew_m_leon...@uk.ibm.com>_
<mailto:andrew_m_leon...@uk.ibm.com>
Cc: _serviceability-...@openjdk.java.net_
<mailto:serviceability-dev@openjdk.java.net>
Date: 11/04/2018 01:02
Subject: Re: RFR: Fix race condition in jdwp
------------------------------------------------------------------------
Hi Andrew,
Okay, I'll file a bug on this topic.
But do you have a standalone test demonstrating this issue?
Thanks,
Serguei
On 4/10/18 06:23, Andrew Leonard wrote:
Hi Serguei,
I don't have access to the bug database to raise one, are you able to
please?
Summary: JDWP debugger initialization hangs intermittently
Description: If during the JDWP setup initialization the VM
initialization takes slightly longer than the main debug
initialization thread a "hang" situation can occur. This has been
seen in testcase test.jck8b.runtime.vm.jdwp and can also be recreated
easily by adding a 10 second sleep to the beginning of the
src/jdk.jdwp.agent/share/native/libjdwp/eventHelper.c method
eventHelper_reportVMInit() .
First seen: JDK8
Recreated: JDK11
Thanks
Andrew
Andrew Leonard
Java Runtimes Development
IBM Hursley
IBM United Kingdom Ltd
Phone internal: 245913, external: 01962 815913
internet email: _andrew_m_leon...@uk.ibm.com_
<mailto:andrew_m_leon...@uk.ibm.com>
From: _"serguei.spit...@oracle.com"_
<mailto:serguei.spit...@oracle.com>_<serguei.spit...@oracle.com>_
<mailto:serguei.spit...@oracle.com>
To: Andrew Leonard _<andrew_m_leon...@uk.ibm.com>_
<mailto:andrew_m_leon...@uk.ibm.com>,
_serviceability-...@openjdk.java.net_
<mailto:serviceability-dev@openjdk.java.net>
Date: 09/04/2018 23:03
Subject: Re: RFR: Fix race condition in jdwp
------------------------------------------------------------------------
Hi Andrew,
The patch itself looks reasonable.
However, in order to proceed with it, a bug report with a standalone
test case demonstrating the issue is needed.
Thanks,
Serguei
On 4/9/18 09:07, Andrew Leonard wrote:
> Hi,
> We discovered in our testing with OpenJ9 that a race condition can
> occur in the jdwp under certain circumstances, and we were able to
> force the same issue with Hotspot. Normally, the event helper thread
> suspends all threads, then the debug loop in the listener thread
> receives a command to resume. The debugger may deadlock if the debug
> loop in the listener thread starts processing commands (e.g. resume
> threads) before the event helper completes the initialization (and
> suspends threads).
>
> This patch adds synchronization to ensure the event helper completes
> the initialization sequence before debugger commands are processed.
>
> Please can I find a sponsor for this contribution? Patch below..
>
> Many thanks
>
> Andrew
>
>
>
> diff --git a/src/jdk.jdwp.agent/share/native/libjdwp/debugInit.c
> b/src/jdk.jdwp.agent/share/native/libjdwp/debugInit.c
> --- a/src/jdk.jdwp.agent/share/native/libjdwp/debugInit.c
> +++ b/src/jdk.jdwp.agent/share/native/libjdwp/debugInit.c
> @@ -1,5 +1,5 @@
> /*
> - * Copyright (c) 1998, 2017, Oracle and/or its affiliates. All rights
> reserved.
> + * Copyright (c) 1998, 2018, Oracle and/or its affiliates. All rights
> reserved.
> * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
> *
> * This code is free software; you can redistribute it and/or
modify it
> @@ -58,6 +58,7 @@
> static jboolean vmInitialized;
> static jrawMonitorID initMonitor;
> static jboolean initComplete;
> +static jboolean VMInitComplete;
> static jbyte currentSessionID;
>
> /*
> @@ -617,6 +618,35 @@
> debugMonitorExit(initMonitor);
> }
>
> +/*
> + * Signal VM initialization is complete.
> + */
> +void
> +signalVMInitComplete(void)
> +{
> + /*
> + * VM Initialization is complete
> + */
> + LOG_MISC(("signal VM initialization complete"));
> + debugMonitorEnter(initMonitor);
> + VMInitComplete = JNI_TRUE;
> + debugMonitorNotifyAll(initMonitor);
> + debugMonitorExit(initMonitor);
> +}
> +
> +/*
> + * Wait for VM initialization to complete.
> + */
> +void
> +debugInit_waitVMInitComplete(void)
> +{
> + debugMonitorEnter(initMonitor);
> + while (!VMInitComplete) {
> + debugMonitorWait(initMonitor);
> + }
> + debugMonitorExit(initMonitor);
> +}
> +
> /* All process exit() calls come from here */
> void
> forceExit(int exit_code)
> @@ -672,6 +702,7 @@
> LOG_MISC(("Begin initialize()"));
> currentSessionID = 0;
> initComplete = JNI_FALSE;
> + VMInitComplete = JNI_FALSE;
>
> if ( gdata->vmDead ) {
> EXIT_ERROR(AGENT_ERROR_INTERNAL,"VM dead at initialize() time");
> diff --git a/src/jdk.jdwp.agent/share/native/libjdwp/debugInit.h
> b/src/jdk.jdwp.agent/share/native/libjdwp/debugInit.h
> --- a/src/jdk.jdwp.agent/share/native/libjdwp/debugInit.h
> +++ b/src/jdk.jdwp.agent/share/native/libjdwp/debugInit.h
> @@ -1,5 +1,5 @@
> /*
> - * Copyright (c) 1998, 2015, Oracle and/or its affiliates. All rights
> reserved.
> + * Copyright (c) 1998, 2018, Oracle and/or its affiliates. All rights
> reserved.
> * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
> *
> * This code is free software; you can redistribute it and/or
modify it
> @@ -39,4 +39,7 @@
> void debugInit_exit(jvmtiError, const char *);
> void forceExit(int);
>
> +void debugInit_waitVMInitComplete(void);
> +void signalVMInitComplete(void);
> +
> #endif
> diff --git a/src/jdk.jdwp.agent/share/native/libjdwp/debugLoop.c
> b/src/jdk.jdwp.agent/share/native/libjdwp/debugLoop.c
> --- a/src/jdk.jdwp.agent/share/native/libjdwp/debugLoop.c
> +++ b/src/jdk.jdwp.agent/share/native/libjdwp/debugLoop.c
> @@ -1,5 +1,5 @@
> /*
> - * Copyright (c) 1998, 2017, Oracle and/or its affiliates. All rights
> reserved.
> + * Copyright (c) 1998, 2018, Oracle and/or its affiliates. All rights
> reserved.
> * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
> *
> * This code is free software; you can redistribute it and/or
modify it
> @@ -98,6 +98,7 @@
> standardHandlers_onConnect();
> threadControl_onConnect();
>
> + debugInit_waitVMInitComplete();
> /* Okay, start reading cmds! */
> while (shouldListen) {
> if (!dequeue(&p)) {
> diff --git a/src/jdk.jdwp.agent/share/native/libjdwp/eventHelper.c
> b/src/jdk.jdwp.agent/share/native/libjdwp/eventHelper.c
> --- a/src/jdk.jdwp.agent/share/native/libjdwp/eventHelper.c
> +++ b/src/jdk.jdwp.agent/share/native/libjdwp/eventHelper.c
> @@ -1,5 +1,5 @@
> /*
> - * Copyright (c) 1998, 2017, Oracle and/or its affiliates. All rights
> reserved.
> + * Copyright (c) 1998, 2018, Oracle and/or its affiliates. All rights
> reserved.
> * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
> *
> * This code is free software; you can redistribute it and/or
modify it
> @@ -580,6 +580,7 @@
> (void)threadControl_suspendThread(command->thread, JNI_FALSE);
> }
>
> + signalVMInitComplete();
> outStream_initCommand(&out, uniqueID(), 0x0,
> JDWP_COMMAND_SET(Event),
> JDWP_COMMAND(Event, Composite));
>
>
>
> Andrew Leonard
> Java Runtimes Development
> IBM Hursley
> IBM United Kingdom Ltd
> Phone internal: 245913, external: 01962 815913
> internet email: _andrew_m_leon...@uk.ibm.com_
<mailto:andrew_m_leon...@uk.ibm.com>
>
>
> Unless stated otherwise above:
> IBM United Kingdom Limited - Registered in England and Wales with
> number 741598.
> Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire
PO6 3AU
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with
number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire
PO6 3AU
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with
number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire
PO6 3AU
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with
number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire
PO6 3AU