I recently read the following paper about CPALockator: Analysis of Correct Synchronization of Operating System Components.
Basically CPALockator is a CPAchecker fork/branch/something that also does thread-modular analysis.

In the evaluation section, they mention having constructed a set of 425 data race detection benchmarks from Linux 4.2.6 device drivers. There's no mention of the availability of this benchmark set, but I managed to find this, which seems like it's the same set: https://gitlab.com/sosy-lab/software/ldv-benchmarks/-/tree/master/linux-4.2.6-races.

I wasn't aware of this earlier and I'm not sure why this didn't get added to sv-benchmarks for data-race. Nevertheless, this might be a useful set of benchmarks for us to look at because:

1. There's 425 of them.
2. They're from the Linux kernel (so should be sufficiently large).
3. Since both CPALockator and Goblint are thread-modular, there's hope that we can also analyze them successfully.

The same repository also seems to have some data-race benchmarks for Linux 4.18: https://gitlab.com/sosy-lab/software/ldv-benchmarks/-/tree/master/linux-4.18-races. And a handful of commit-based data-race benchmarks in Linux (although most don't have before vs after like the README claims): https://gitlab.com/sosy-lab/software/ldv-benchmarks/-/tree/main/ldv-commits-races.

Some of the benchmarks are in Unknowns subdirectories but still contain task definitions with expected verdicts (not sure based on what). These should probably be excluded to be on the safe side.