request for review (S)
Keith McGuigan
keith.mcguigan at oracle.com
Mon Jun 4 05:46:25 PDT 2012
On 6/2/2012 1:57 AM, David Holmes wrote:
> Hi Keith,
>
> On 2/06/2012 3:58 AM, Keith McGuigan wrote:
>> Hello,
>>
>> This fixes up a hole which I'm reasonably confident is the cause of
>> http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7166498
>>
>> Webrev: http://cr.openjdk.java.net/~kamg/7166498/
>> (the comment in the code explains what's going on)
>
> So it's really bcs->bcp() that may have changed, and hence the
> aligned_bcp calculated from that. Seems ok.
>
> I hadn't realized we have safepoint issues for data other than oops.
> Makes we wonder how can we detect/prevent this kind of issue. Also makes
> me wonder how we might better flag data that isn't safe across a
> safepoint, and methods that might lead to a safepoint?
Yes, this thought had crossed my mind as well. Likely we'd have to
remove all accessors that give direct access to Java heap memory and use
some sort of wrapper (similar to what we do with oops and
-XX:+CheckUnhandledOops).
But... with perm gen elimination coming around, many of the places where
we access the heap will be eliminated since metadata won't be in the
heap anymore. So the general problem will be mostly mitigated soon anyway.
--
- Keith
More information about the hotspot-runtime-dev
mailing list