On Tue, Oct 18, 2022 at 2:58 AM wangyufen <wangyufen@xxxxxxxxxx> wrote:
>
>
> 在 2022/10/18 10:57, wangyufen 写道:
> >
> > 在 2022/10/13 9:51, Martin KaFai Lau 写道:
> >> On 10/11/22 2:57 AM, Wang Yufen wrote:
> >>> The program file used in the udpgro_frglist testcase is
> >>> "../bpf/nat6to4.o",
> >>> but the actual nat6to4.o file is in "bpf/" not "../bpf".
> >>> The following error occurs:
> >>> Error opening object ../bpf/nat6to4.o: No such file or directory
> >>
> >> hmm... so it sounds like the test never works...
> >>
> >> The test seems like mostly exercising the tc-bpf? It makes sense to
> >> move it to the selftests/bpf. or staying in net is also fine for now
> >> and only need to fix up the path here.
> >>
> >> However, if moving to selftests/bpf, I don't think it is a good idea
> >> to only move the bpf prog but not moving the actual test program (the
> >> script here) such that the bpf CI can continuously testing it.
> >> Otherwise, it will just drift and rot slowly like patch 2.
> >>
> >> Also, if you prefer to move it to selftests/bpf, the bpf prog cannot
> >> be moved in the current form. eg. There is some convention on the
> >> SEC name in the selftests/bpf/progs. Also, the testing script needs
> >> to be adapted to the selftests/bpf/test_progs infra.
> >
> > hmm... if moving to selftests/bpf, the actual test programs also needs
> > to move to selftests/bpf, e.g. udpgso_bench_*, in_netns.sh,
> > udpgso*.sh, which may not be a good idea.
> >
> > So, only fix up the path here.
> >
> > Also fix up the bpf/nat6to4.o compile error as following:
> >
> > make -C tools/testing/selftests/net got the following err:
> > bpf/nat6to4.c:43:10: fatal error: 'bpf/bpf_helpers.h' file not found
> > ^~~~~~~~~~~~~~~~~~~
> >
> >
> After revert commit 7b92aa9e61350("selftests net: fix kselftest net
> fatal error"),
>
> make -C tools/testing/selftests got the following err:
>
> In file included from bpf/nat6to4.c:43:
> ../../../lib/bpf/bpf_helpers.h:11:10: fatal error: 'bpf_helper_defs.h'
> file not found
> #include "bpf_helper_defs.h"
> ^~~~~~~~~~~~~~~~~~~
>
> "bpf_helper_defs.h" is generated by libbpf；
>
>
> So, there are two possible approaches: the first moving nat6to4.c and
> the actual test programs to selftests/bpf;
>
> second add make dependency on libbpf for the nat6to4.c.
>
> Which one is better?
Neither.
Martin already explained that the whole thing needs to move to selftests/bpf.
