RFR: 8305819: LogConfigurationTest intermittently fails on AArch64 [v2]

Tue May 9 08:43:29 UTC 2023

On Mon, 8 May 2023 04:50:21 GMT, David Holmes <dholmes at openjdk.org> wrote:

> If the iteration of the output list falls outside of the RCU reader protection then I think we need to treat it like any other lock-free code. That will at least repair the currently detected bug.
> 
> After that we should look at replacing the this "ad-hoc" synchronization (lock + RCU + barriers) with a more direct synchronization mechanism (rwLock?) - which I think is what @theRealAph is advocating for to get rid of any UB associated with RCU.

To be clear, I'm happy with any _legal_ C++ code.
In my opinion it doesn't help to have custom synchronization mechanisms local to some parts of HotSpot, mostly because people underestimate how hard it is to get such mechanisms right.

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

PR Comment: https://git.openjdk.org/jdk/pull/13421#issuecomment-1539697461