On Fri, Mar 03, 2023 at 01:41:00PM +0100, Paolo Abeni wrote:
On Fri, 2023-03-03 at 12:44 +0100, Greg Kroah-Hartman wrote:
On Fri, Mar 03, 2023 at 12:39:07PM +0100, Paolo Abeni wrote:
Additionally, some self-tests check for known bugs/regressions. Running them on older kernel will cause real trouble, and checking for bug presence in the running kernel would be problematic at best, I think.
No, not at all, why wouldn't you want to test for know bugs and regressions and fail? That's a great thing to do, and so you will know to backport those bugfixes to those older kernels if you have to use them.
I'm sorry, I likely was not clear at all. What I mean is that the self- test for a bug may trigger e.g. memory corruption on the bugged kernel (or more specifically to networking, the infamous, recurring "unregister_netdevice: waiting for ...") which in turn could cause random failures later.
If that specific case runs on older (unpatched) kernel will screw the overall tests results. The same could happen in less-detectable way for old bugs non explicitly checked by any test, but still triggered by the test-suite. As a consequence I expect that the results observed running newer self-tests on older kernel are unreliable.
For the stable/LTS kernel trees, they should _never_ be unreliable, otherwise that means we have missed a needed fix and so we need to resolve that.
Which is why I always recommend running the latest selftests on all older kernels, and have for a very long time now.
thanks,
greg k-h