Re: Application (GC) stuck in "_Jv_MarkArray" when doing "klass->getName();"

[Martin Egholm Nielsen <martin at egholm-nielsen dot dk>]

Hi,

> You somehow want to move the getName call out of the collector.  Calling
> it from GC_dump() or a partial clone, and then calling that from a
> debugger or the client code should work.  Just printing klass and then
> using a debugger to perform the getName() call should also work.  As
> should dumping the klass values into a static buffer, and then later
> traversing it and performing the getName call.
I made a terrible hack, caching the array when gc'ing and then dumping 
the stuff in the next call to "_Jv_GCTotalMemory".
But that reveal much, since it was just an array of "java.lang.Object".

The next thing is to invoke "toString()" on the elements, so I may get 
an idea of what they consist of...

> The problem is that the collector isn't designed to allocate
> garbage-collected memory while you're in the middle of the garbage
> collector, particularly in non-incremental mode.
Makes sense...

// Martin

> > -----Original Message-----
> > From: java-owner@gcc.gnu.org [mailto:java-owner@gcc.gnu.org] 
> > On Behalf Of Martin Egholm Nielsen
> > Sent: Monday, March 20, 2006 4:21 AM
> > To: java@gcc.gnu.org
> > Subject: Re: Application (GC) stuck in "_Jv_MarkArray" when 
> > doing "klass->getName();"
> >
> >
> > > > I haven't checked, but I'd be suspicious that getName allocates 
> > > > memory, which acquires the GC lock, which is already held 
> >
> > by the same thread.
> >
> > > > (This is a non-reentrant lock.  Which is good, because 
> >
> > things would 
> >
> > > > fail in more complex ways if it weren't.)
> > >
> > > Well, going into getName seems to trace furhter to 
> >
> > "_Jv_AllocString" 
> >
> > > using "_Jv_AllocPtrFreeObj" or "_Jv_AllocPtrFreeObject" 
> >
> > depending on 
> >
> > > the state of "ENABLE_JVMPI". However, now things get deep 
> >
> > for me - at 
> >
> > > least at my current state of knowledge...
> > > So I don't really know where that leaves the question regarding the 
> > > non-reentrant lock... :-)
> >
> > But if this is indeed a matter of locking, what other means 
> > do I have at that point (_Jv_MarkArray) to identify what the 
> > array is actually about?
> >
> > // Martin
> >
> >
> > > > On Sat, 18 Mar 2006, Martin Egholm Nielsen wrote:
> > > >
> > > >
> > > > > Hi there,
> > > > >
> > > > > In my long and tiresome (for the rest of you) attempts to 
> >
> > understand 
> >
> > > > > the doubling of my GC-times, I've now added some output 
> >
> > to "_Jv_MarkArray"
> >
> > > > > to get an idea of what and why there is suddenly an array 
> >
> > given of 
> >
> > > > > length 170.000.
> > > > > (This attempt is a consequence of calling "gcj_describe_type_fn" 
> > > > > from the GC, but it just halted.) So I added parts of the 
> >
> > the code 
> >
> > > > > from "gcj_describe_type_fn" into
> > > > > _Jv_MarkArray:
> > > > >
> > > > >  if ( klass ) {
> > > > >      printf( "MEN: 1\n" );
> > > > >      jstring name = klass -> getName();
> > > > >      printf( "MEN: 2\n" );
> > > > >  } // if
> > > > >
> > > > > However, when the GC gets to it, "1" is printed, "2" is 
> >
> > not, and the 
> >
> > > > > GC is stuck in between.
> > > > > Anyone can tell me why this is not "allowed"/fails?
> > > > >
> > > > > BR,
> > > > > Martin Egholm