SIGSEGV on GCTaskThread in jdk 6u18

[Hiroshi Yamauchi]

Ramki,

Thanks for your advice.

> If you have to use an older version (i.e. hs16 etc.),
> i'd suggest using the workaround below, and of
> running with heap verification enabled to see
> what may be causing the issue. Yes, missing card-marks
> (generated code, interpreter, or GC) or bad oop-maps
> for example can also cause such crashes. Heap verification
> would help narrow that down some for further debugging
> to proceed.

As stated, -ReduceInitialCardMarks didn't make it go away. I haven't
tried +UseSerialGC, but would try it as a last resort because the
performance difference could be significant. Also, the heap
verification thing requires a fastdebug build, I suppose, which isn't
sometimes an option in a production environment.

>
> If you have a support contract, feel free to file a
> detailed bug and/or escalate. At that time please
> include a test case if you have one.

As far as I know, we don't have a support contract.

>
> oh, and please do check if you can reproduce with the
> most recent bits...

That's a good idea, except that trying recent bits isn't
straightforward as it can be. I'll try.

Thanks a lot. I may come back to this issue if none of the workarounds
work. I kind of hoped that we forgot to incorporate certain patches
that would fix this problem. But it sounds like that's not the case.

Hiroshi


> On 07/02/10 11:04, Hiroshi Yamauchi wrote:
>>
>> Hi folks,
>>
>> I dig up this old thread. I thought this issue has been resolved. But
>> I've been still seeing a similar crash in the UseParallelGC collector
>> with a version of HS16 (and HS11, too.)
>>
>> Here's the stack trace at the crash point:
>>
>> PSPromotionManager::copy_to_survivor_space(oopDesc*, bool)
>> PSPromotionManager::drain_stacks_depth(bool)
>> CardTableExtension::scavenge_contents_parallel(ObjectStartArray*,
>> MutableSpace*, HeapWord*, PSPromotionManager*, unsigned int)
>> OldToYoungRootsTask::do_it(GCTaskManager*, unsigned int)
>> GCTaskThread::run()
>> java_start(Thread*)
>>
>> which is similar to the one from
>> http://bugs.sun.com/view_bug.do?bug_id=6896647 . This bug is marked
>> 'fix delivered'. We made sure we have already had the patch
>> http://hg.openjdk.java.net/jdk7/hotspot-gc/hotspot/rev/7b0e9cba0307
>> incorporated in our build. Also, we tried the
>> -XX:-ReduceInitialCardMarks flag. But the crash keeps coming back
>> (though the flag seems to reduce the frequency of the crash). This is
>> on Linux/x86_32.
>>
>> A potentially related bug might be
>> http://bugs.sun.com/view_bug.do?bug_id=6850723
>>
>> Is there some other patch that we should use or some other workarounds
>> or diagnosis we can try?
>>
>> Thanks,
>> Hiroshi
>>
>>
>> On Thu, Jan 21, 2010 at 8:53 AM, Y. Srinivas Ramakrishna
>> <Y.S.Ramakrishna at sun.com> wrote:
>>>
>>> Janda Martin wrote:
>>>>
>>>> Thank you for fast response. I supposed that I use Serial GC so this
>>>> problem is different.
>>>
>>> Unless you specify it explicitly via -XX:+UseSerialGC, the GC ergonomics
>>> might pick a different collector, and in this case it picked parallel gc
>>> because yr machine was considered a "server class" machine (or some such)
>>> because you had 8 cores on yr box.
>>>
>>>>  I will try ReduceInitialCardMarks
>>>
>>> Yes either that, or explicitly specify serial gc.
>>>
>>> In your case, since you have 8 cores, you should use -ReduceInitialCardMarks
>>> and continue using the parallel collector you are getting by default.
>>> Indeed you might want to specify -XX:+UseParallelOldGC which i think you
>>> aren't getting by default (i wonder why; i thought that was the parallel
>>> collector default now, but haven't checked recently).
>>>
>>> -- ramki
>>>
>>>>  Martin
>>>>
>>>> ----- Original Message -----
>>>> From: "Vladimir Kozlov" <Vladimir.Kozlov at Sun.COM>
>>>> To: hotspot-dev at openjdk.java.net
>>>> Sent: Thursday, January 21, 2010 10:03:01 AM GMT +01:00 Amsterdam / Berlin
>>>> / Bern / Rome / Stockholm / Vienna
>>>> Subject: Re: SIGSEGV on GCTaskThread in jdk 6u18
>>>>
>>>> Try -XX:-ReduceInitialCardMarks
>>>>
>>>> Please, read release notes for 6u18.
>>>>
>>>> Vladimir
>>>>
>>>> On 1/21/10 12:24 AM, Andreas Kohn wrote:
>>>>>
>>>>> On Thu, 2010-01-21 at 08:45 +0100, Janda Martin wrote:
>>>>>>
>>>>>> Please can you help me. I experienced several crashes on GCTaskThread in
>>>>>> latest Java6u18 when running IntelliJ IDEA (not only). 64bit linux, 32bit
>>>>>> jvm, 4 core processor
>>>>>>  I create new bug report with Review ID: 1699960.
>>>>>>
>>>>>>  Is there any -XX options to help diagnose this problem. It looks like
>>>>>> regression from 6u17 (maybe) or 6u16 (sure).
>>>>>
>>>>> We observed a very similar 6u18 crash (once so far), 64bit linux, 64bit
>>>>> jvm.
>>>>>
>>>>> I have attached our hs_err log as well.
>>>>>
>>>>> Regards,
>>>>> --
>>>>> Andreas
>>>>>
>>>