[5.4.y] selftests/bpf build broken by "bpf: Zero-fill..."

[5.4.y] selftests/bpf build broken by "bpf: Zero-fill..."

[Kamal Mostafa]

Hi Sasha-

This v5.4.78 commit breaks the tools/testing/selftests/bpf build:

[linux-5.4.y] c602ad2b52dc bpf: Zero-fill re-used per-cpu map element

Like this:

	prog_tests/map_init.c:5:10: fatal error: test_map_init.skel.h: No such file or directory
	    5 | #include "test_map_init.skel.h"

Because tools/testing/selftests/bpf/Makefile in v5.4 does not have the
"skeleton header generation" stuff (circa v5.6).

Reverting c602ad2b52dc from linux-5.4.y fixes it.

 -Kamal

Re: [5.4.y] selftests/bpf build broken by "bpf: Zero-fill..."

[Kamal Mostafa]

On Fri, Dec 04, 2020 at 10:28:46AM -0800, Kamal Mostafa wrote:
> Hi Sasha-
> 
> This v5.4.78 commit breaks the tools/testing/selftests/bpf build:
> 
> [linux-5.4.y] c602ad2b52dc bpf: Zero-fill re-used per-cpu map element
> 
> Like this:
> 
> 	prog_tests/map_init.c:5:10: fatal error: test_map_init.skel.h: No such file or directory
> 	    5 | #include "test_map_init.skel.h"
> 
> Because tools/testing/selftests/bpf/Makefile in v5.4 does not have the
> "skeleton header generation" stuff (circa v5.6).
> 
> Reverting c602ad2b52dc from linux-5.4.y fixes it.

Another option would be to just drop the selftest from linux-5.4.y,
but keep the beneficial change to kernel/bpf/hashtab.c.

(We're leaning towards that approach for Ubuntu).

 -Kamal

Re: [5.4.y] selftests/bpf build broken by "bpf: Zero-fill..."

[David Verbeiren]

Hi Kamal, Sasha,

On Mon, Dec 7, 2020 at 11:42 PM Kamal Mostafa <kamal@canonical.com> wrote:
>
> On Fri, Dec 04, 2020 at 10:28:46AM -0800, Kamal Mostafa wrote:
> > Hi Sasha-
> >
> > This v5.4.78 commit breaks the tools/testing/selftests/bpf build:
> >
> > [linux-5.4.y] c602ad2b52dc bpf: Zero-fill re-used per-cpu map element
> >
> > Like this:
> >
> >       prog_tests/map_init.c:5:10: fatal error: test_map_init.skel.h: No such file or directory
> >           5 | #include "test_map_init.skel.h"
> >
> > Because tools/testing/selftests/bpf/Makefile in v5.4 does not have the
> > "skeleton header generation" stuff (circa v5.6).
> >
> > Reverting c602ad2b52dc from linux-5.4.y fixes it.
>
> Another option would be to just drop the selftest from linux-5.4.y,
> but keep the beneficial change to kernel/bpf/hashtab.c.
>
> (We're leaning towards that approach for Ubuntu).
>
>  -Kamal

An alternative could be to use the initial version of the selftest I had
proposed before learning about the skeleton approach.

You can find it here:
https://lore.kernel.org/bpf/20201029111730.6881-1-david.verbeiren@tessares.net/

I also think it would be good to keep the fix of course.

-David

Re: [5.4.y] selftests/bpf build broken by "bpf: Zero-fill..."

[Sasha Levin]

On Mon, Dec 07, 2020 at 02:42:40PM -0800, Kamal Mostafa wrote:
>On Fri, Dec 04, 2020 at 10:28:46AM -0800, Kamal Mostafa wrote:
>> Hi Sasha-
>>
>> This v5.4.78 commit breaks the tools/testing/selftests/bpf build:
>>
>> [linux-5.4.y] c602ad2b52dc bpf: Zero-fill re-used per-cpu map element
>>
>> Like this:
>>
>> 	prog_tests/map_init.c:5:10: fatal error: test_map_init.skel.h: No such file or directory
>> 	    5 | #include "test_map_init.skel.h"
>>
>> Because tools/testing/selftests/bpf/Makefile in v5.4 does not have the
>> "skeleton header generation" stuff (circa v5.6).
>>
>> Reverting c602ad2b52dc from linux-5.4.y fixes it.
>
>Another option would be to just drop the selftest from linux-5.4.y,
>but keep the beneficial change to kernel/bpf/hashtab.c.
>
>(We're leaning towards that approach for Ubuntu).

That's what I did, thanks!

-- 
Thanks,
Sasha