<html>
<head>
<meta content="text/html; charset=windows-1252"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<div class="moz-cite-prefix">On 5/22/16 10:01 PM, Jon Masamitsu
wrote:<br>
</div>
<blockquote
cite="mid:d573e400-f72a-3845-9209-a3d70b583572@oracle.com"
type="cite">
<br>
<br>
On 5/20/2016 2:24 PM, Kim Barrett wrote:
<br>
<blockquote type="cite">Please review this fix for a serious
performance bug in non-product
<br>
verification code in CMS precleaning.
<br>
<br>
When precleaning an object, the mark bits for the object are
verified,
<br>
including scanning for invalid set mark bits interior to the
object.
<br>
If there are multiple dirty ranges applicable to the object,
<br>
processing each range leads to mark bit verification. If the
object
<br>
is large and there are many applicable dirty ranges, a lot of
time may
<br>
be spent on this verification.
<br>
<br>
For example, in a test that creates many large objArrays (O(64)
of
<br>
O(16M) elements) and randomly scribbles on them, we've seen
preclean
<br>
times of 20+ minutes because of this.
<br>
This verification isn't precise, since it is performed when
the object
<br>
is visited during precleaning. Not all live objects will
require
<br>
precleaning, and this won't catch post-visit mark corruption.
This
<br>
verification is also redundant with that done later, during the
sweep
<br>
phase, where it *is* precise. So the only benefit of the
problematic
<br>
verification is that it might catch a problem earlier.
<br>
</blockquote>
<br>
Instead of eliminate the verification can you leave it in for
non-object arrays?
<br>
You might be able to do the verification for the object array once
if you did it
<br>
only when an object header was in the dirty range but do it for
the length of
<br>
the entire object array.
<br>
<br>
If the verification were to consistently fail here and never fail
when done during
<br>
the sweep, I think that bounds where in the code we need to look
for a problem.
<br>
<br>
Jon
<br>
<br>
</blockquote>
Alternatively, if we still wanted to verify this early (I have no
strong opinion), could we simply verify the mark bits corresponding
to the MemRegion passed in, instead of the entire object? Something
like <br>
<meta http-equiv="content-type" content="text/html;
charset=windows-1252">
<br>
<meta http-equiv="content-type" content="text/html;
charset=windows-1252">
<pre>HeapWord* obj_rng_start = <span class="removed">addr+2</span>;</pre>
<pre>HeapWord* obj_rng_end = (<span class="removed">_bitMap->isMarked(addr+1)</span>) ? <span class="removed">addr+size-1</span> : <span class="removed">addr+size;</span></pre>
<pre><span class="removed"></span><span class="removed">_bitMap->verifyNoOneBitsInRange(MAX2(</span><span class="removed">obj_rng_start, mr->start()),
</span><span class="removed"><span class="removed"> MIN2(</span><span class="removed">obj_rng_end, mr->end()));
</span></span></pre>
<span class="removed"><span class="removed"> - Derek</span> </span><br>
<span class="removed"></span><br>
<br>
<blockquote
cite="mid:d573e400-f72a-3845-9209-a3d70b583572@oracle.com"
type="cite">
<blockquote type="cite">
<br>
Since the problematic verification is redundant with later
checks, we
<br>
simply remove it.
<br>
<br>
CR:
<br>
<a class="moz-txt-link-freetext" href="https://bugs.openjdk.java.net/browse/JDK-8156548">https://bugs.openjdk.java.net/browse/JDK-8156548</a>
<br>
<br>
Webrev:
<br>
<a class="moz-txt-link-freetext" href="http://cr.openjdk.java.net/~kbarrett/8156548/webrev.00/">http://cr.openjdk.java.net/~kbarrett/8156548/webrev.00/</a>
<br>
<br>
Testing:
<br>
Locally tested before and after change, using GC logging to see
<br>
precleaning time drop from 15+ minutes to 20 seconds.
<br>
<br>
</blockquote>
<br>
</blockquote>
<p><br>
</p>
</body>
</html>