RFR: 8140326: G1: Consider putting regions where evacuation failed into next collection set [v4]

[Albert Mingkun Yang]

On Wed, 2 Aug 2023 10:09:06 GMT, Thomas Schatzl <tschatzl at openjdk.org> wrote:

>> tams and stats are updated eagerly in conc-start-gc-pause and later cleared if the evac-fail region is sparse enough. Can update to tams and stats be deferred to when the fate of this evac-fail region is decided? IOW, instead of checking is-conc-start-gc-pause in multiple places, do it once only inside evac-fail-processing to restore the tams and stats invariants if really needed.
>
> It can, see a prototype at https://github.com/openjdk/jdk/compare/pr/14220...tschatzl:jdk:8140326-evacuate-retained-regions-at-any-time-work?expand=1 ; would that change be what you had in mind?

Yes; the change in `note_start_of_marking` is surprising though. Why do regions in cset need tams? I thought marking doesn't touch cset.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/14220#discussion_r1281742929