Re: [PATCH 0/3] selftests/bpf: Fixes for out-of-tree builds

[Eduard Zingerman <eddyz87@gmail.com>]

On Tue, 2026-09-22 at 15:31 -0300, Ricardo B. Marlière (SUSE) wrote:
> On Mon Sep 21, 2026 at 6:04 PM -03, Eduard Zingerman wrote:
> > On Mon, 2026-09-21 at 14:40 -0600, Nico Pache (Red Hat) wrote:
> > 
> > ...
> > 
> > > Given your series was defferred, and there is a "rewrite" coming, what would
> > > be the best approach to working together (between us three) to get these
> > > fixes in that series too. 
> > 
> > The, "rewrite", as you call it, had landed today.
> > So the series in question can be re-spun.
> > Regarding your patch on top of it, it appears you can just post is separately.
> > However, I am inclined to be confrontational towards changes like this one.
> > The purpose of the selftests makefile is to run selftests as appropriate
> > for current tip of the kernel tree using a very specific kernel configuration.
> 
> Ideally, if a Kconfig symbol is needed and the kernel under test does not
> have it, tests are skipped. But with BPF they were simply not built,
> rendering the tests useless for anything other than in-tree vmlinux with
> tools/selftests/bpf/config*.
> 
> > Main users of this makefile are BPF developers and BPF CI.
> 
> But they shouldn't be the *only* ones, IMHO. Enabling distros to build and
> run the tests is also valuable to catch regressions in their kernels, don't
> you think?
> 
> > Support for distros customizing the makefile is a non-goal.
> > The "PERMISSIVE" change done enough damage already.
> 
> It's ugly, I agree, but what damage has it caused? I see that it's going to
> be reverted anyway, but I would appreciate if you could elaborate.

It's the first time I see a makefile that builds stuff not fully but
90% of it and returns success. The damage is the need to maintain it
going forward that already cost some time.

Do you have a list of kernel configs you had issues with?