harmony-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ilya Berezhniuk (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HARMONY-5303) [drlvm][EUT] regression in coreresources on Linux x86_64
Date Mon, 24 Dec 2007 01:18:43 GMT

    [ https://issues.apache.org/jira/browse/HARMONY-5303?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12554210

Ilya Berezhniuk commented on HARMONY-5303:

After looking at Java and native stacks and investigating VM state in GDB, it appeared that
thread 2 is trying to enter synchronized ElementTree.getChildren() method and waiting on ElementTree
monitor. The monitor is acquired by thread 5, but thread 5 stack shows that this monitor should
be released already.

I supposed that ElementTree object monitor was not released by thread 5 in some curcumstances.
One of probable causes of monitor exit miss is some error in monitor fastpath processing.
So I've prepared a patch to disable monitor enter/exit fastpath generation and started cyclic
runs in 2 different ways.

The results were the following:
1) Runs with ANT script on 16-core machine: there was 1 hangup from 10 runs - in another test
and on another native lock. Earlier I had 50% of hangups in this configuration.
2) Direct EUT runs on 4-core machine: no hangups in 80 runs (earlier every 4th..10th run leaded
to hangup exactly in Bug_134364).

So looks like the problem can be in fast path (although disabled fast path can affect results
by changed performance/locking).

I've prepared another patch to gather Java stacks from jthread_monitor_enter/jthread_monitor_exit
to analyze when ElementTree monitor should be released.
In first try I've got 156Gb file too huge to analyze :(
I'm now trying to filer debug output to get more useful trace.

> [drlvm][EUT] regression in coreresources on Linux x86_64
> --------------------------------------------------------
>                 Key: HARMONY-5303
>                 URL: https://issues.apache.org/jira/browse/HARMONY-5303
>             Project: Harmony
>          Issue Type: Bug
>          Components: DRLVM
>         Environment: Linux / x86_64
>            Reporter: Vladimir Beliaev
>         Attachments: stacks.txt
> The EUT3.3 suite - coreresources - start hanging on Linux x86_64, it  passed previouly.
> The latest passed revision is 597106 (21-Nov-2007): 
> http://people.apache.org/~smishura/r597106/Linux_x86_64/eut33/results/html/org.eclipse.core.tests.resources_linux.gtk.x86_64.html
> The next test revisions (597436) demonstrated the crash (23-Nov-2007): 
> http://people.apache.org/~smishura/r597436/Linux_x86_64/eut33/
> The initial evaluation is required to detect the guilty commit or component.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message