RFR (S): 8204668: Cleanup management of the java.vm.info System property
Chris Plummer
chris.plummer at oracle.com
Tue Jun 12 17:52:41 UTC 2018
Hi David,
Is there a reason you moved the PropertyList_add() call to a slightly
later point?
You moved the Arguments::update_vm_info_property() call up a bit. How
did you decide exactly how early it can be called without risk of other
flags changing?
thanks,
Chris
On 6/12/18 2:56 AM, David Holmes wrote:
> Bug: https://bugs.openjdk.java.net/browse/JDK-8204668
> webrev: http://cr.openjdk.java.net/~dholmes/8204668/webrev/
>
> JDK-8203329 fixed a problem where the native system property for the
> vm.info string was not updated after argument parsing, resulting in
> JVM TI reporting an incorrect value.
>
> Looking at the overall approach for this property it can be simplified
> quite a bit. The basic issue is that it is initialized early in VM
> startup (so it can be present for crash logs) before argument parsing,
> but some details can change due to argument parsing. If we update the
> native value immediately after argument parsing, and so before the
> properties are passed through to the Java side, then we don't need to
> execute the Java code in reset_vm_info() to perform that update.
> Additionally, if we expose the SystemProperty directly (as done for
> other properties) then we can do away with the new
> PropertyList_update_value() function that has to search for the
> property to be updated.
>
> Overall this cuts out a chunk of initialization code that may aid with
> startup costs; and simplifies the code.
>
> There's some additional history in the bug report.
>
> Testing:
> - tier 1, 2, 3
> - regression test from JDK-8203329:
> -
> serviceability/jvmti/GetSystemProperty/JvmtiGetSystemPropertyTest.java
>
> Thanks,
> David
More information about the hotspot-runtime-dev
mailing list