start building kselftests under CI

[nickdesaulniers]

Once ClangBuiltLinux/linux#1698 is resolved, we should wire up at least 1 build of the kselftests so that they don't regress //building//.

Running them will be more complicated and should be tracked in a second distinct issue, since that will involve fetching the build artifacts, repacking them in our userspace images, then running them upon boot, and then trying to understand the output.

Perhaps one build for x86_64 is sufficient for now.

cc @JustinStitt

[gctucker]

Out of interest, are you planning to use make kselftest-merge to create the kernel config?

[nickdesaulniers]

I'm unfamiliar with that make target. Do you recommend that we should be using that instead? Or in addition?

[gctucker]

It's the standard upstream way of creating a kernel config with all the fragments for all the kselftests. I think it ought to work, and if it doesn't then the CI should help identify issues. It's something to discuss with Shuah I think in terms of actual recommendation.

FIY So far, KernelCI has been relying on its own way to collate all the kselftest fragments. I think kselftest-merge wasn't really ready when we started. Now with the new API & Pipeline, the build jobs are being reworked and it would seem like a logical move to use it. We'll see how it goes. Also, some kselftests can be run with a plain defconfig build.