Re: [patch] Real-Time Preemption, -RT-2.6.10-rc2-mm1-V0.7.28-1

[hui]

On Mon, Nov 22, 2004 at 10:31:00AM +0100, Christian Meder wrote:
> Hi Ingo,
> 
> after two evenings of experimenting this is the current status
> (everything based on 0.7.29-0, will try 0.7.30-x during the day):
> 
> * the lockup can't be triggered from the console or using a remote
> session and I really tried to torture the box ;-)
> * the real trigger is mouse activity in X
> * the other important factor is running the jvm in profiling mode,
> running without jvm or with the jvm in non-profiling mode leaves the box
> stable
> * I couldn't yet figure out the pattern of java program which is
> triggering. Not every java program is triggering but at least I found
> several public available ones. I wrote some small test programs doing
> simple multithreading but they didn't trigger.
> 
> So the simplest setup I found til now is the following: 
> 
> chris@blue:~$ java -version
> java version "1.4.1"
> Java(TM) 2 Runtime Environment, Standard Edition (build Blackdown-1.4.1-01)
> Java HotSpot(TM) Client VM (build Blackdown-1.4.1-01, mixed mode)
> chris@blue:~$ JAVA_OPTIONS=-Xrunhprof:cpu=samples,file=crap.log,depth=3 jython 
> Jython 2.1 on java1.4.1 (JIT: null)
> Type "copyright", "credits" or "license" for more information.
> >>>
> 
> Now moving the mouse around in X will make the box lockup in less than
> 10 seconds.

HotSpot is pretty heavy on VM faulting as well as signal handling, SIGUSR1,
during per thread GC safepointing operations to get the current thread
ucontext for GC traversal roots. Debugging the HotSpot VM is nearly impossible
without heavy unit testing and even that isn't going to push through that
almost pure voodoo code easily. I'm a former HotSpot tweeker for the BSDs so
I know a thing or two about that particular VM.

Try a small Java app to see if this triggers the same lock ups. Are you 
pushing it using Swing ?

I'd try pushing an incremental load on it, maybe some rapid object creation
and destruction with increasing number of threads.

Also, CC the Sun/Blackdown folks about this. It could very well be some
kind of NPTL glue problem triggering.

bill

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/