RFR: 8332955: ubsan: runningCounters.cpp:48:61: runtime error: member call on null pointer of type 'struct VirtualSpaceList'

[Thomas Stuefe]

On Thu, 30 May 2024 06:31:05 GMT, David Holmes <dholmes at openjdk.org> wrote:

> By strange coincidence we just encountered an assertion failure during VM initalization (https://bugs.openjdk.org/browse/JDK-8332935) that leads to this secondary failure doing the NMT reporting, so I am glad to see it fixed.
> 
> But I understand @stefank 's position about the code being well-defined except during this initialization phase, and also wonder where is the "best" place to address this? Doing it in the "leaf" methods is more future-proof in a sense but it also means we open up the possibility of even more code interacting with Metaspace stats before metaspace is initialized, which seems to me to be something we would not want and would perhaps like to detect. So I'm slightly in favour of having the NMT code check if metaspace is initialized before trying to report any metaspace stats if that is feasible.

It is feasible, and likely trivial. We can do both. I opened https://bugs.openjdk.org/browse/JDK-8333290 to track this. 

I still would like this PR go through for now. I think hardening this leaf function makes sense.

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

PR Comment: https://git.openjdk.org/jdk/pull/19435#issuecomment-2139942936