RFR: 8078644: CDS needs to support JVMTI CFLH

serguei.spitsyn at oracle.com serguei.spitsyn at oracle.com
Wed Sep 7 23:29:29 UTC 2016

Hi Jiangli,

Just to let you know that I'm reviewing this.


On 9/6/16 17:14, Jiangli Zhou wrote:
> Hi,
> Please review the following change thats support JVMTI class_file_load_hook(CFLH) during initial loading of shared classes. The webrev also removes the temporary workaround that disables CDS when JVMTI CFLH is required (JDK-8141341 <https://bugs.openjdk.java.net/browse/JDK-8141341>).
> 	webrev:http://cr.openjdk.java.net/~jiangli/8078644/webrev.00/
>          bug: JDK-8078644 <https://bugs.openjdk.java.net/browse/JDK-8078644>
> Class file data is needed when posting CFLH. For shared classes, the class file data are now archived at dump time. That avoids re-reading the class file data from the JAR file or reconstituting the data at runtime, which would add class loading overhead for shared classes.
> To minimize the runtime memory overhead, the archived class file data are stored in a separate shared space, called ‘optional data space’ (od).  The ‘od’ space a new region in the archive. It does not increase runtime memory usage when JvmtiExport::should_post_class_file_load_hook() is false. The memory contains the archived class file data is only paged in when the VM needs to post CFLH.  The ‘od’ space can be shared by different processes.
> When loading shared class at runtime, we now call JvmtiExport::post_class_file_load_hook() with the archive class file data if needed. If JVMTI agent modifies the class, new InstanceKlass is created using the modified class data and the archived class is ignored.
> Tested with JPRT, CDS tests and QA tests.
> Thanks,
> Jiangli

More information about the hotspot-runtime-dev mailing list