Re: [rvm-research] Find holes in a block of SegregatedFreeListSpace

Hi Dingwen,

You might find the MMTk Test Harness
(http://www.jikesrvm.org/The+MMTk+Test+Harness) useful in debugging this
problem.

Regards,
Robin

Dingwen Yuan wrote:
> Hallo Steve,
>
> Thank you for the reply. Some context is like this. I added some code
> to scan the mature space of GenMS for holes in the available list.
>
> GenMS.collectionPhase
>   public final void collectionPhase(short phaseId) {
>
>     if (traceFullHeap()) {
>       ...........
>       if (phaseId == RELEASE) {
>           if (Options.verbose.getValue() > 0)
>               Log.writeln("main release");
>         matureTrace.release();
>
>         msSpace.release();
>
>         holes = 0;
>         msSpace.linearScan();
>         Log.write("holes: ");
>         Log.writeln(holes);
>
>         super.collectionPhase(phaseId);
>         return;
>       }
>     }
>   }
>
> SegregatedFreeListSpace.linearScan
>
>   public final void linearScan(boolean print) {
>     for (int sizeClass = 0; sizeClass < sizeClassCount(); sizeClass++) {
>
>         if (VM.VERIFY_ASSERTIONS) {
>
> VM.assertions._assert(this.flushedBlockHead.get(sizeClass).isZero());
>
> VM.assertions._assert(this.consumedBlockHead.get(sizeClass).isZero());
>         }
>
>         Address block = this.availableBlockHead.get(sizeClass);
>         Extent blockSize =
> Extent.fromIntSignExtend(BlockAllocator.blockSize(blockSizeClass[sizeClass]));
>
>         while (!block.isZero()) {
>             Address next = BlockAllocator.getNext(block);
>             this.auditBlock(block, sizeClass, blockSize);
>             block = next;
>           }
>       }
>     }
>   }
>
> SegregatedFreeListSpace.auditBlock
>    @Inline
>    private void auditBlock (Address block, int sizeClass,
>             Extent blockSize) {
>
>     Address end = block.plus(blockSize);
>
>     Address cursor = block.plus(blockHeaderSize[sizeClass]);
>     Extent cellExtent = Extent.fromIntSignExtend(cellSize[sizeClass]);
>
>     while (cursor.LT(end)) {
>       ObjectReference current =
> VM.objectModel.getObjectFromStartAddress(cursor);
>       if (current.isNull() || !isCellLive(current)) {
>           holes++;
>       }
>       cursor = cursor.plus(cellExtent);
>     }
>   }
>
> And I also did a very simple experiment on it. The experiment code
> added 20 Integer to a static arrayList. Of course, all these objects
> will be promoted to mature space after gc. At the end of the code I
> called 10 System.gc() in a sequence and I also set
> -X:gc:fullHeapSystemGC=true so that full gc was be triggered. The
> holes of these 10 gcs are
> holes: 0
> holes: 1367
> holes: 348
> holes: 1359
> holes: 1011
> holes: 1359
> holes: 348
> holes: 1359
> holes: 1011
> holes: 1359
>
> This doesn't conform to my expectation. I supposed that the second to
> the tenth "holes" should be 0 when the first is already 0. And in
> addition, there seems to be some pattern in the numbers -- 1359 =
> 1011+348. My guess is that this may relate to the markState of
> MarkSweepSpace which changes from gc to gc.
>
> BTW, the version I used is 2.9.2
>
> cheers,
>
> Dingwen
>
>
> 2009/6/11 Steve Blackburn <Steve.Blackburn@...

> <mailto:Steve.Blackburn@...>>

>
>     Hi Dingwen,
>
>     Sorry I have not replied earlier.
>
>     I'm afraid the specifics of the surrounding code are not at the front
>     of my mind right now and it's pretty tough to debug your code with so
>     little context.  Can you be a bit more specific about how it is
>     failing?
>
>     For what it's worth, I strongly recommend the use of assertions,
>     particularly in GC code---figure out what your invariants and
>     preconditions are and then thrown assertions in liberally to ensure
>     your assumptions are holding true.   The assertions cost nothing in a
>     production build (they are turned off and removed statically---no
>     runtime checks).
>
>     --Steve
>
>     On 10/06/2009, at 6:32 PM, Dingwen Yuan wrote:
>
>     > Sorry, I hit a space and the above mail was mistakenly sent. Again,
>     > the code is like this:
>     >
>     > @Inline
>     > private void auditBlock (Address block, int sizeClass,
>     >             Extent blockSize) {
>     >
>     >       Extent cellBytes = Extent.fromIntSignExtend(
>     > cellSize[sizeClass]);
>     >       Address cursor = block.plus(blockHeaderSize[sizeClass]);
>     >       Address end = block.plus(blockSize);
>     >
>     >       while(cursor.LT(end)) {
>     >         ObjectReference current =
>     > VM.objectModel.getObjectFromStartAddress(cursor);
>     >         if (current == null || !this.isCellLive(current)) {
>     >                holes++;
>     >         }
>     >         cursor = cursor.plus(cellBytes);
>     >      }
>     >   }
>     >
>     > I call this function to scan blocks in "available list" at the
>     > RELEASE phase of GC directly after space.release() was called. But I
>     > found there should be some bug in it, but cannot tell where. Could
>     > anyone help me with this?
>     >
>     > Thanks in advance.
>     >
>     > Dingwen
>     >
>     > TU Darmstadt
>     >
>     ------------------------------------------------------------------------------
>     > Crystal Reports - New Free Runtime and 30 Day Trial
>     > Check out the new simplified licensing option that enables unlimited
>     > royalty-free distribution of the report engine for externally facing
>     > server and web deployment.
>     >
>     http://p.sf.net/sfu/businessobjects_______________________________________________
>     > Jikesrvm-researchers mailing list
>     > Jikesrvm-researchers@...

>     <mailto:Jikesrvm-researchers@...>

>     > https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>
>
>     ------------------------------------------------------------------------------
>     Crystal Reports - New Free Runtime and 30 Day Trial
>     Check out the new simplified licensing option that enables unlimited
>     royalty-free distribution of the report engine for externally facing
>     server and web deployment.
>     http://p.sf.net/sfu/businessobjects
>     _______________________________________________
>     Jikesrvm-researchers mailing list
>     Jikesrvm-researchers@...

>     <mailto:Jikesrvm-researchers@...>

>     https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>
>
> ------------------------------------------------------------------------
>
> ------------------------------------------------------------------------------
> Crystal Reports - New Free Runtime and 30 Day Trial
> Check out the new simplified licensing option that enables unlimited
> royalty-free distribution of the report engine for externally facing
> server and web deployment.
> http://p.sf.net/sfu/businessobjects
> ------------------------------------------------------------------------
>
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jikesrvm-researchers@...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>


------------------------------------------------------------------------------
Crystal Reports - New Free Runtime and 30 Day Trial
Check out the new simplified licensing option that enables unlimited
royalty-free distribution of the report engine for externally facing
server and web deployment.
http://p.sf.net/sfu/businessobjects
_______________________________________________
Jikesrvm-researchers mailing list
Jikesrvm-researchers@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers