On Mon, Dec 14, 2020 at 12:21 PM Kamal Mostafa <kamal@xxxxxxxxxxxxx> wrote:
>
> If Makefile cannot find any of the vmlinux's in its VMLINUX_BTF_PATHS list,
> it tries to run btftool incorrectly, with VMLINUX_BTF unset:
>
> bpftool btf dump file $(VMLINUX_BTF) format c
>
> Such that the keyword 'format' is misinterpreted as the path to vmlinux.
> The resulting build error message is fairly cryptic:
>
> GEN vmlinux.h
> Error: failed to load BTF from format: No such file or directory
>
> This patch makes the failure reason clearer by yielding this instead:
>
> Makefile:...: *** cannot find a vmlinux for VMLINUX_BTF at any of
> "{paths}". Stop.
>
> Fixes: acbd06206bbb ("selftests/bpf: Add vmlinux.h selftest exercising tracing of syscalls")
> Cc: stable@xxxxxxxxxxxxxxx # 5.7+
> Signed-off-by: Kamal Mostafa <kamal@xxxxxxxxxxxxx>
> ---
> tools/testing/selftests/bpf/Makefile | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile
> index 542768f5195b..93ed34ef6e3f 100644
> --- a/tools/testing/selftests/bpf/Makefile
> +++ b/tools/testing/selftests/bpf/Makefile
> @@ -196,6 +196,9 @@ $(BUILD_DIR)/libbpf $(BUILD_DIR)/bpftool $(BUILD_DIR)/resolve_btfids $(INCLUDE_D
> $(INCLUDE_DIR)/vmlinux.h: $(VMLINUX_BTF) | $(BPFTOOL) $(INCLUDE_DIR)
> ifeq ($(VMLINUX_H),)
> $(call msg,GEN,,$@)
> +ifeq ($(VMLINUX_BTF),)
> +$(error cannot find a vmlinux for VMLINUX_BTF at any of "$(VMLINUX_BTF_PATHS)")
> +endif
> $(Q)$(BPFTOOL) btf dump file $(VMLINUX_BTF) format c > $@
> else
> $(call msg,CP,,$@)
This handles only one possible case where $(VMLINUX_BTF) is expected.
Given $(VMLINUX_BTF) is now almost mandatory to have (bpf_testmod.ko
is dependent on it, for instance), I think it's acceptable to just add
such a check right after VMLINUX_BTF definition, outside of any
specific rule. The downside is that some unrelated targets (e.g.,
non-test_progs tests) will fail if VMLINUX_BTF is not found, even if
they don't need it, but the most common use case us to build
test_progs, as well as bpf_testmod.ko and runqslower, so I think it's
an OK trade off. Worst case, one can work around such with
VMLINUX_BTF=whatever command-line overload.
> --
> 2.17.1
>
