RFR: Remove safepoint-cleanup piggybacking code

Roman Kennke rkennke at redhat.com
Mon Sep 17 09:04:10 UTC 2018


The code that attempts to piggy-back monitor deflation and nmethod
marking on thread scanning requires quite a lot of changes against
upstream. To my knowledge, it has not worked very reliably, and hasn't
provided a very significant benefit either. It has also bitrotten a bit
because we never enabled it by default and therefore it's never been
part of our testing.

I have asked if upstream has interest in accepting it and the reply has
been that monitor deflation is being worked on to be done concurrently
(and probably there's ways to come up with a deflation scheme such that
it doesn't become a problem to begin with), and nmethod marking should
not be done in safepoints to begin with, but instead use thread local
handshakes.

I propose to remove/revert this stuff:

http://cr.openjdk.java.net/~rkennke/remove-sp-cleanup/webrev.00/

Testing: tier3_gc_shenandoah

WDYT?

Roman



More information about the shenandoah-dev mailing list