Hi Coleen,
It's in the ClassLoaderData initialization. I don't know what it means, you'd
have to ask someone who knows. I don't think it matters though.
It means that if G1 concurrently scans this oop and we over-write the
uninitialized oop, the uninitialized oop would be saved in the SATB-barrier.
The oopStorage contains strong roots which are done in a safeopint.
Thanks, Robbin
Coleen
You can leave out using OopHandle. I have a patch to add the missing
functionality and add it to your code. Actually, I was looking to see how
much OopHandle is used to see if it's helping anything and there is a lot of
code using it. Most of it is to hide oop* in ClassLoaderData.
This change otherwise looks great.
Thanks, Robbin
Thanks,
Coleen
Thanks for having a look, Robbin
On 12/16/19 1:32 PM, [email protected] wrote:
I have to think about this. Could there be breakpoints in old emcp
methods that we do not remove? The metadata_do function is trying to keep
old Methods from being deleted while there are still references to them.
http://cr.openjdk.java.net/~rehn/8235912/v1/webrev/src/hotspot/share/prims/jvmtiImpl.hpp.udiff.html
+ oop* _class_holder; // keeps _method memory from being deallocated
We created the class OopHandle to encapsulate strong oopStorage references,
although it's missing oop_store. Can you use that?
Coleen
On 12/16/19 4:47 AM, Robbin Ehn wrote:
Hi all, please review.
From issue, https://bugs.openjdk.java.net/browse/JDK-8235912:
JvmtiBreakpoints are walked via VMThread oops_do (the breakpoint is in a
vm operation) before they are installed in the safeopint and after they
have been installed, walked with JvmtiCurrentBreakpoints::oops_do().
By putting the class holder inside oopStorage there is no need for this.
JvmtiCurrentBreakpoints::metadata_do is not needed because redefine
classes actually removes the breakpoints before updating them (so there is
no breakpoints to update).
We can just remove metadata_do.
I also removed some unused code.
Changeset:
http://cr.openjdk.java.net/~rehn/8235912/v1/webrev/
Passes several runs of nsk jvmti/jdi and t1-7.
Thanks, Robbin
