I'm not surprised that there are vast numbers of latent low-probability race conditions in the linux kernel or any major software project. Having seen the testing process in both hardware and software projects, the two are not even comparable. The testing process in most software projects is dominated by fully deterministic unit tests that are very simple in nature, and make large numbers of assumptions about the behaviors and interactions with other components. Low-probability race conditions between different components is exactly the outcome I would expect from this testing process.

https://software.rajivprab.com/2019/04/28/rethinking-softwar...