Re: [PATCH 4/8] selftests/bpf: fix object files installation

From: Yauheni Kaliuta <yauheni.kaliuta@redhat.com>
To: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Cc: bpf <bpf@vger.kernel.org>, Jiri Benc <jbenc@redhat.com>,
	Jiri Olsa <jolsa@redhat.com>, Andrii Nakryiko <andriin@fb.com>,
	Daniel Borkmann <daniel@iogearbox.net>
Subject: Re: [PATCH 4/8] selftests/bpf: fix object files installation
Date: Thu, 28 May 2020 21:46:37 +0300	[thread overview]
Message-ID: <CANoWswmmjKmC8L8Fo27cUzGAF_vhWBgY52PForTfsffir7Qpyg@mail.gmail.com> (raw)
In-Reply-To: <CAEf4BzbgGnJ8d+q5KyjdXYMHFOoyi7UJpKWLS1t_oNh2dhKy+g@mail.gmail.com>

On Thu, May 28, 2020 at 9:40 PM Andrii Nakryiko
<andrii.nakryiko@gmail.com> wrote:
>
> On Tue, May 26, 2020 at 10:17 PM Yauheni Kaliuta
> <yauheni.kaliuta@redhat.com> wrote:
> >
> > Hi, Andrii!
> >
> > >>>>> On Tue, 26 May 2020 15:30:19 -0700, Andrii Nakryiko  wrote:
> >
> >  > On Thu, May 21, 2020 at 9:14 PM Yauheni Kaliuta
> >  > <yauheni.kaliuta@redhat.com> wrote:
> >  >>
> >  >> There are problems with bpf test programs object files:
> >  >>
> >  >> 1) some of them are build for flavored test runner and should be
> >  >> installed in the subdirectory;
> >  >> 2) it's possible that the same file mentioned several times (added
> >  >> for every different unflavored test runner);
> >  >> 3) some generated files are not treated properly.
> >  >>
> >  >> Fix 1) by adding subdirectory to the list. rsync -a in the install
> >  >> target will handle it.
> >  >>
> >  >> Fix 2) by filtering the list. Performance should not matter for such
> >  >> amount of files.
> >  >>
> >  >> Fix 3) by use proper (TEST_GEN_FILES) variable for the list.
> >  >>
> >  >> Fixes: 309b81f0fdc4 ("selftests/bpf: Install generated test progs")
> >  >> Fixes: e47a179997ce ("bpf, testing: Add missing object file to
> >  >> TEST_FILES")
> >  >>
> >  >> Signed-off-by: Yauheni Kaliuta <yauheni.kaliuta@redhat.com>
> >  >> ---
> >  >> tools/testing/selftests/bpf/Makefile | 9 ++++++---
> >  >> 1 file changed, 6 insertions(+), 3 deletions(-)
> >  >>
> >  >> diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile
> >  >> index 19091dbc8ca4..1ba3d72c3261 100644
> >  >> --- a/tools/testing/selftests/bpf/Makefile
> >  >> +++ b/tools/testing/selftests/bpf/Makefile
> >  >> @@ -42,8 +42,7 @@ ifneq ($(BPF_GCC),)
> >  >> TEST_GEN_PROGS += test_progs-bpf_gcc
> >  >> endif
> >  >>
> >  >> -TEST_GEN_FILES =
> >  >> -TEST_FILES = test_lwt_ip_encap.o \
> >  >> +TEST_GEN_FILES = test_lwt_ip_encap.o \
> >  >> test_tc_edt.o
> >  >>
> >  >> BTF_C_FILES = $(wildcard progs/btf_dump_test_case_*.c)
> >  >> @@ -273,7 +272,11 @@ TRUNNER_BPF_OBJS := $$(patsubst %.c,$$(TRUNNER_OUTPUT)/%.o, $$(TRUNNER_BPF_SRCS)
> >  >> TRUNNER_BPF_SKELS := $$(patsubst %.c,$$(TRUNNER_OUTPUT)/%.skel.h,      \
> >  >> $$(filter-out $(SKEL_BLACKLIST),       \
> >  >> $$(TRUNNER_BPF_SRCS)))
> >  >> -TEST_GEN_FILES += $$(TRUNNER_BPF_OBJS)
> >  >> +
> >  >> +TO_ADD := $(if $2,$$(TRUNNER_OUTPUT),$$(TRUNNER_BPF_OBJS))
> >  >> +$$(foreach i,$$(TO_ADD),\
> >  >> +       $$(eval \
> >  >> +               TEST_GEN_FILES += $$(if $$(filter $$i,$$(TEST_GEN_FILES)),,$$i)))
> >
> >  > This makes me cringe. Can we not have three levels of nested
> >  > evals, please? I also didn't get exactly what's the problem
> >  > you are trying to solve, could you give some example, please?
> >
> > It's sort of `unique` functionality.
>
> `unique` in make world is just $(sort $VAR). Isn't that a more
> light-weight and generic way to avoid duplicates in lib.mk?

Oh, my bad, totally forgot it. Sure! Thanks!

>
> >
> > With the current approach TEST_GEN_FILES has at least 2 copies of
> > an object file (for call test_progs and test_maps) which is both
> > inaccurate and increasing the length of the variable (even if
> > copying the same file should not cause problems).
> >
> >
> > (Without sub-directory handling it's even overwritten by
> > flavoured binaries in between).
> >
> > BTW, how would you like to change $(call ...) with $(value ...)?
> > It will get rid of one level of indirection but requires
> > rule-specific variables for rule generation, since some
> > evaluations are done in recipies.
>
> I don't exactly understand the implications, so don't know. But the
> less changes to this Makefile, the happier I am, so... :)

So, no way ;)

It would be relatively many changes, but more simple code without extra $$.

>
> >
> >  >>
> >  >> # Evaluate rules now with extra TRUNNER_XXX variables above already defined
> >  >> $$(eval $$(call DEFINE_TEST_RUNNER_RULES,$1,$2))
> >  >> --
> >  >> 2.26.2
> >  >>
> >
> >
> > --
> > WBR,
> > Yauheni Kaliuta
> >
>

-- 
WBR, Yauheni