RFR: 8319048: Monitor deflation unlink phase prolongs time to safepoint [v8]

[Daniel D. Daugherty]

On Tue, 21 Nov 2023 10:23:39 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

>> src/hotspot/share/runtime/synchronizer.cpp line 122:
>> 
>>> 120:           // walks searching for new prev during unlink under heavy list inserts.
>>> 121:           break;
>>> 122:         }
>> 
>> Is there a reason to make this check and bail out while executing the `do ... while`
>> and gathering the first batch of deflated monitors? Why can't you finish gathering
>> your first batch and let the head change logic below just take care of it?
>
> My thinking was two-fold: a) small batch size saves us from TTSP hiccup in gathering loop, but under heavy inserts we might end up searching for batch head through a long list -- and that is where small batch size would not save us; b) since `prev != nullptr` after the first update, we do not incur lots of overheads with checking in the gathering loop.

I still don't get it, but I don't want to hold this PR up any longer.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/16412#discussion_r1406771911