RFR: JDK-8262068: Improve G1 Full GC by skipping compaction for regions with high survival ratio

[Hamlin Li]

On Sat, 6 Mar 2021 10:02:35 GMT, Hamlin Li <mli at openjdk.org> wrote:

>> I manage to get a crash locally using a fastdebug build on Linux x64. I haven't had time to look at the error in detail, but it is the same crash. It happened in different test so just running a lot of testing should trigger it in your environment as well.
>
> Hi Stefan, Thanks a lot for the information, It's very helpful!

HI Stefan,

Seems we have fixed the previous crashes.
Would you mind to help to review the fixes and the whole change set when available? Thanks.
Summarize the previous 2 commits as below.

e267bc3: 
  1. fix the crash when get object size which is caused by class unloading, by using "_bitmap->get_next_marked_addr(next_addr, limit);" rather than call "obj->size();" directly.
  2. some misc improvement when initializing G1RegionMarkStatsCache.
a28a93f: 
  1. fix the bot crash by cross_threshold.
  2. fix initialization issue of G1RegionMarkStatsCache, and assert issue in previous commit e267bc3.

-------------

PR: https://git.openjdk.java.net/jdk/pull/2760