Kernel self tests

Posted Aug 21, 2014 3:39 UTC (Thu) by glikely (subscriber, #39601)
Parent article: Kernel self tests

In hallway-track conversations after the session, more time was spent talking about test time. Rusty Russell commented to me that if a test takes longer than about a second, then it isn't going to get run for smoke testing. I pretty much agree with that sentiment.

It is absolutely important to make it easy to execute longer running tests, and I think we're going to try to do that. However, for the purpose of smoke testing we'll need to predefine a set of "effectively instant" test cases.

Kernel self tests

Posted Aug 25, 2014 17:26 UTC (Mon) by grundler (guest, #23450) [Link]

Even if tests take longer than a second to run, they are still useful for regression testing - when one needs to confirm a subsystem or API behaves in some documented way.

What's critical for getting developers to run tests, it has to be easy to:
1) DISCOVER tests
2) RUN tests

Just like "make install" is the standard way to build a kernel, we need a standard way to discover and invoke the tests..perhaps "make list_tests" and "make run_test FOO".

ChromeOS uses autotest framework to meet those goals...but autotest has substantial infrastructure (setup/cleanup) cost (~30 seconds for a basic test) that clearly could be improved upon by some "in kernel" tests. And while python isn't that difficult, learning curve for autotest infrastructure is painful enough that I don't like re-learning it the once or twice a year I have to work on some test.