RFR: [8u] JDK-8191393: Random crashes during cfree+0x1c

[Sergey Nazarkin]

Hi!

I’m a bit confused but is gcLogFileStream a singleton object? Shouldn’t new Mutex be destroyed at gcLogFileStream destructor?

/Sergey




> On Apr 6, 2020, at 13:43, Liu, Xin <xxinliu at amazon.com> wrote:
> 
> Hi,  Reviewers, 
> 
> May I request review this bugfix? 
> Bug: https://bugs.openjdk.java.net/browse/JDK-8191393
> Webrev: https://cr.openjdk.java.net/~xliu/8191393/webrev/
> 
> I believe that root cause  is the race condition between ConcurrentMarkThread and the VMThread.
> We can’t directly backport from upstream because upstream is based on unified logging framework. further, TIP doesn’t have reentry problem while jdk8u has. 
> Instead, this patch uses a mutex to protect FILE* fileStream::_file and a Boolean flag to avoid reentry.  
> 
> Of course, a mutex  is not free. Therefore, I didn't use it if users don't emit log to file. 
> 
> Test: 
> I ran the stress test over 24 hours and no problem is identified. 
> I passed tier1 test of jtreg. 
> 
> Thanks,
> --lx
> 
> 
>