On Mon, Jun 10, 2019 at 12:03 PM Hechao Li <hechaol@xxxxxx> wrote: > > Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> wrote on Mon [2019-Jun-10 11:05:28 -0700]: > > On Mon, Jun 10, 2019 at 9:57 AM Hechao Li <hechaol@xxxxxx> wrote: > > > > > > I got an error when compiling selftests/bpf: > > > > > > libbpf.c:411:10: error: implicit declaration of function 'reallocarray'; > > > did you mean 'realloc'? [-Werror=implicit-function-declaration] > > > progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0])); > > > > > > It was caused by feature-reallocarray=1 in FEATURE-DUMP.libbpf and it > > > was fixed by manually removing feature/ folder. This diff adds feature/ > > > to EXTRA_CLEAN to avoid this problem. > > > > > > Signed-off-by: Hechao Li <hechaol@xxxxxx> > > > --- > > > > There is no need to include v1 into patch prefix for a first version > > of a patch. Only v2 and further versions are added. > > > > > > > tools/testing/selftests/bpf/Makefile | 3 ++- > > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > > > diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile > > > index 2b426ae1cdc9..44fb61f4d502 100644 > > > --- a/tools/testing/selftests/bpf/Makefile > > > +++ b/tools/testing/selftests/bpf/Makefile > > > @@ -279,4 +279,5 @@ $(OUTPUT)/verifier/tests.h: $(VERIFIER_TESTS_DIR) $(VERIFIER_TEST_FILES) > > > ) > $(VERIFIER_TESTS_H)) > > > > > > EXTRA_CLEAN := $(TEST_CUSTOM_PROGS) $(ALU32_BUILD_DIR) \ > > > - $(VERIFIER_TESTS_H) $(PROG_TESTS_H) $(MAP_TESTS_H) > > > + $(VERIFIER_TESTS_H) $(PROG_TESTS_H) $(MAP_TESTS_H) \ > > > + feature > > > > It doesn't seem any of linux's Makefile do that. From brief reading of > > build/Makefile.feature, it seems like it is supposed to handle > > transparently the case where environment changes and thus a set of > > supported features changes. I also verified that FEATURE-DUMP.libbpf > > is re-generated every single time I run make in > > tools/testing/selftests/bpf, even if nothing changed at all. So I > > don't think this patch is necessary. > > > > I'm not sure what was the cause of your original problem, though. > > > > > -- > > > 2.17.1 > > > > > # Background: > > My default GCC version is 4.8.5, which caused the following error when I > run make under selftests/bpf: > libbpf.c:39:10: fatal error: libelf.h: No such file or directory > > To fix it, I have to run: > > make CC=<Path to GCC 7.x> > > The I got reallocarray not found error. By deleting feature/ folder > under selftests/bpf, it was fixed. > > # Root Cause: > > Now I found the root cause. When I run "make", which uses GCC 4.8.5, it > generates feature/test-reallocarray.d which indicates reallocarray is > enabled. However, when I switched to GCC 7.0, this file was not > re-generated and thus even FEATURE-DUMP.libbpf was re-generated, > feature-reallocarray is still 1 in it. > > This can be reproduced by the following steps: > $ cd tools/testing/selftests/bpf > $ make clean && make CC=<Path to GCC 4.8.5> > (Fail due to libelf.h not found) > $ make clean && make CC=<Path to GCC 7.x> > (Should succeed but actually fail with reallocarray not defined) > > If adding feature to EXTRA_CLEAN is not the way to go, do you have any > suggestion to fix such problem? I spent some time debugging this and I > hope to fix it so that other people with similar situation won't have to > waste time on this issue. I think adding feature to EXTRA_CLEAN is just fine. I was hesitant without understanding how this happened. Maybe condense this explanation as part of commit message, I think it's useful, thanks for explaining! Acked-by: Andrii Nakryiko <andriin@xxxxxx> > > Thanks, > Hechao
