Re: [rvm-research] Possible race hazard in RVMClass

Thanks for the bug reports.

The second stack trace is also due to a concurrency bug arising from an
unsynchronized use of a WeakHashMap.    I should be pushing fixes for both
shortly.

--dave


UGAWA Tomoharu <[email protected]> wrote on 03/18/2013 11:56:32 AM:

> From: UGAWA Tomoharu <[email protected]>
> To: [email protected],
> Cc: Jones Richard <[email protected]>, "Ritson C.G."
> <[email protected]>
> Date: 03/18/2013 12:55 PM
> Subject: [rvm-research] Possible race hazard in RVMClass
>
> Hi
>
> I am afraid that we found a possible bug.
>
> There may be a race in org.jikesrvm.classloader.RVMClass.
> The addSubClass method can be used by multiple threads simultaneously
> causing corruption of the subClasses array.
>
>   /**
>    * Add to list of classes that derive from this one.
>    */
>   private void addSubClass(RVMClass sub) {
>     int n = subClasses.length;
>     RVMClass[] tmp = new RVMClass[n + 1];
>
>     for (int i = 0; i < n; ++i) {
>       tmp[i] = subClasses[i];
>     }
>     tmp[n] = sub;
>
>     subClasses = tmp;
>   }
>
> This bug manifests when testing our concurrent collector with DaCapo
> 2006 jython.
> However it can be reproduced in Jikes tip using the attached example
program.
> The example program uses a script to create a large number of class
> files which
> are then loaded by separate threads using reflection.
>
> We tested on Linux:
> > Linux enju 3.2.0-23-generic #36-Ubuntu SMP Tue Apr 10 20:39:51 UTC
> 2012 x86_64 x86_64 x86_64 GNU/Linux
> with Jikes version:
> > e6fab50149e5 tip
> The configuration was:
> > FastAdaptiveGenImmix_x86_64-linux
>
> This occasionally produces two different error messages:
>
> Exception in thread "Thread-3"
> java.lang.ArrayIndexOutOfBoundsException: Array index out of range: 2466
>    at org.jikesrvm.classloader.RVMClass.addSubClass(RVMClass.java:1646)
>    at org.jikesrvm.classloader.RVMClass.<init>(RVMClass.java:1011)
>    at org.jikesrvm.classloader.RVMClass.createReflectionClass
> (RVMClass.java:1935)
>    at org.jikesrvm.classloader.RVMMethod.getInvoker(RVMMethod.java:881)
>    at java.lang.reflect.VMMethod.<init>(VMMethod.java:42)
>    at java.lang.reflect.JikesRVMSupport.createMethod
(JikesRVMSupport.java:27)
>    at java.lang.Class.getMethods(Class.java:1018)
>    at Ref.run(Unknown Source:0)
>    at java.lang.Thread.run(Thread.java:745)
>
> Exception in thread "Thread-2" java.lang.InternalError: WeakHashMap
> in incosistent state
>    at java.lang.Error.<init>(Error.java:81)
>    at java.lang.VirtualMachineError.<init>(VirtualMachineError.java:71)
>    at java.lang.InternalError.<init>(InternalError.java:70)
>    at java.util.WeakHashMap.internalRemove(WeakHashMap.java:706)
>    at java.util.WeakHashMap.cleanQueue(WeakHashMap.java:595)
>    at java.util.WeakHashMap.put(WeakHashMap.java:776)
>    at org.jikesrvm.classloader.MethodVector.finish(MethodVector.java:74)
>    at org.jikesrvm.classloader.RVMClass.resolve(RVMClass.java:1200)
>    at
> org.jikesrvm.runtime.RuntimeEntrypoints.initializeClassForDynamicLink
> (RuntimeEntrypoints.java:608)
>    at org.jikesrvm.classloader.RVMClass.createReflectionClass
> (RVMClass.java:1942)
>    at org.jikesrvm.classloader.RVMMethod.getInvoker(RVMMethod.java:881)
>    at java.lang.reflect.VMMethod.<init>(VMMethod.java:42)
>    at java.lang.reflect.JikesRVMSupport.createMethod
(JikesRVMSupport.java:27)
>    at java.lang.Class.getMethods(Class.java:1018)
>    at Ref.run(Unknown Source:0)
>    at java.lang.Thread.run(Thread.java:745)
>
> The first of these relates to the addSubClass method and can be
> fixed by making the
> method synchronized.
>
> diff -r fb5276512eae rvm/src/org/jikesrvm/classloader/RVMClass.java
> --- a/rvm/src/org/jikesrvm/classloader/RVMClass.java   Fri Feb 15
> 23:30:49 2013 +0100
> +++ b/rvm/src/org/jikesrvm/classloader/RVMClass.java   Tue Feb 19
> 07:16:34 2013 +0000
> @@ -1638,7 +1638,7 @@
>    /**
>     * Add to list of classes that derive from this one.
>     */
> -  private void addSubClass(RVMClass sub) {
> +  private synchronized void addSubClass(RVMClass sub) {
>      int n = subClasses.length;
>      RVMClass[] tmp = new RVMClass[n + 1];
>
> Attached is a PDF containing graphs of the first run performance for
> DaCapo 2006 without patch (prod-tip) and with patch (prod-ascp).
> These show this patch does not cause a performance regression.
>
> As for the second message, we have not yet located the cause.
>
> Kind regards
> Tomoharu
>
> [attachment "reflection.tar.gz" deleted by David P Grove/Watson/IBM]
> [attachment "ascp-cryo.pdf" deleted by David P Grove/Watson/IBM]
>
------------------------------------------------------------------------------

> Everyone hates slow websites. So do we.
> Make your web apps faster with AppDynamics
> Download AppDynamics Lite for free today:
> http://p.sf.net/sfu/appdyn_d2d_mar
> _______________________________________________
> Jikesrvm-researchers mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers

------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_mar

_______________________________________________
Jikesrvm-researchers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers