RFR: 8327250: assert(!method->is_old()) failed: Should not be installing old methods
Erik Österlund
eosterlund at openjdk.org
Wed Apr 10 13:29:11 UTC 2024
On Wed, 10 Apr 2024 12:51:22 GMT, Coleen Phillimore <coleenp at openjdk.org> wrote:
> When the [JDK-8322630](https://bugs.openjdk.org/browse/JDK-8322630) ICStubs and safepoints were removed, a check to ignore redefined methods was removed (is_old() check), but that check was in the wrong place. The LinkResolver finds a Method to call through resolution and selection and saves that Method in the returned CallInfo. The CallInfo object is temporary until the method is stored somewhere "safe", ie somewhere redefinition knows how to replace the method with a new method or deoptimize. At any safepoint, the Method could be redefined, so the CallInfo return should check for old methods and only return the equivalent new method.
>
> This change removes the one check in the interpreter and moves it to the CallInfo accessors.
>
> Tested with the failing test case with options: "-Xcomp -ea -esa -XX:CompileThreshold=100 -XX:+UnlockExperimentalVMOptions -XX:TieredStopAtLevel=2", all vmTestBase/nsk/jvmti tests with these options, and ran tiers 1-8.
This looks great. Thanks for fixing this!
Looks like there is a missing include for JvmtiExport. I don't need to see the update.
-------------
Marked as reviewed by eosterlund (Reviewer).
PR Review: https://git.openjdk.org/jdk/pull/18719#pullrequestreview-1991689029
PR Comment: https://git.openjdk.org/jdk/pull/18719#issuecomment-2047544405
More information about the hotspot-runtime-dev
mailing list