netdev - Re: [PATCH kselftest] fix single bpf test

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <93d96c99-4712-4054-a36f-3c65c80ab3f8@linuxfoundation.org>
Date: Fri, 22 Nov 2024 08:14:58 -0700
From: Shuah Khan <skhan@...uxfoundation.org>
To: Jiayuan Chen <mrpre@....com>, linux-kselftest@...r.kernel.org,
 Mark Brown <broonie@...nel.org>
Cc: song@...nel.org, bpf@...r.kernel.org, linux-kernel@...r.kernel.org,
 netdev@...r.kernel.org, martin.lau@...ux.dev, andrii@...nel.org,
 ast@...nel.org, kpsingh@...nel.org, jolsa@...nel.org,
 Shuah Khan <skhan@...uxfoundation.org>
Subject: Re: [PATCH kselftest] fix single bpf test

On 11/18/24 07:06, Jiayuan Chen wrote:
> Currently, when testing a certain target in selftests, executing the
> command 'make TARGETS=XX -C tools/testing/selftests' succeeds for non-BPF,
> but a similar command fails for BPF:
> '''
> make TARGETS=bpf -C tools/testing/selftests
> 
> make: Entering directory '/linux-kselftest/tools/testing/selftests'
> make: *** [Makefile:197: all] Error 1
> make: Leaving directory '/linux-kselftest/tools/testing/selftests'
> '''
> 
> The reason is that the previous commit:
> commit 7a6eb7c34a78 ("selftests: Skip BPF seftests by default")
> led to the default filtering of bpf in TARGETS which make TARGETS empty.
> That commit also mentioned that building BPF tests requires external
> commands to run. This caused target like 'bpf' or 'sched_ext' defined
> in SKIP_TARGETS to need an additional specification of SKIP_TARGETS as
> empty to avoid skipping it, for example:
> '''
> make TARGETS=bpf SKIP_TARGETS="" -C tools/testing/selftests
> '''
> 
> If special steps are required to execute certain test, it is extremely
> unfair. We need a fairer way to treat different test targets.
> 

Note: Adding Mark, author for commit 7a6eb7c34a78 to the thread

The reason we did this was bpf test depends on newer versions
of LLVM tool chain.

A better solution would be to check for compile time dependencies in
bpf Makefile and check run-time dependencies from bpf test or a wrapper
script invoked from run_tests to the skip the test if test can't run.

I would like to see us go that route over addressing this problem
with SKIP_TARGETS solution.

The commit 7a6eb7c34a78 went in 4 years ago? DO we have a better
story for the LLVM tool chain to get rid of skipping bpf and sched_ext?

Running make -C tools/testing/selftests/bpf/ gave me the following error.
Does this mean we still can't include bpf in default run?

make -C tools/testing/selftests/bpf/
make: Entering directory '/linux/linux_6.12/tools/testing/selftests/bpf'

Auto-detecting system features:
...                                    llvm: [ OFF ]


   GEN     /linux/linux_6.12/tools/testing/selftests/bpf/tools/build/bpftool/vmlinux.h
libbpf: failed to find '.BTF' ELF section in /linux/linux_6.12/vmlinux
Error: failed to load BTF from /linux/linux_6.12/vmlinux: No data available
make[1]: *** [Makefile:209: /linux/linux_6.12/tools/testing/selftests/bpf/tools/build/bpftool/vmlinux.h] Error 195
make[1]: *** Deleting file '/linux/linux_6.12/tools/testing/selftests/bpf/tools/build/bpftool/vmlinux.h'
make: *** [Makefile:369: /linux/linux_6.12/tools/testing/selftests/bpf/tools/sbin/bpftool] Error 2
make: Leaving directory '/linux/linux_6.12/tools/testing/selftests/bpf'

> This commit provider a way: If a user has specified a single TARGETS,
> it indicates an expectation to run the specified target, and thus the
> object should not be skipped.
> 
> Another way is to change TARGETS to DEFAULT_TARGETS in the Makefile and
> then check if the user specified TARGETS and decide whether filter or not,
> though this approach requires too many modifications.
> Signed-off-by: Jiayuan Chen <mrpre@....com>
> ---
>   tools/testing/selftests/Makefile | 7 +++++--
>   1 file changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/testing/selftests/Makefile b/tools/testing/selftests/Makefile
> index 363d031a16f7..d76c1781ec09 100644
> --- a/tools/testing/selftests/Makefile
> +++ b/tools/testing/selftests/Makefile
> @@ -116,7 +116,7 @@ TARGETS += vDSO
>   TARGETS += mm
>   TARGETS += x86
>   TARGETS += zram
> -#Please keep the TARGETS list alphabetically sorted
> +# Please keep the TARGETS list alphabetically sorted
>   # Run "make quicktest=1 run_tests" or
>   # "make quicktest=1 kselftest" from top level Makefile
>   
> @@ -132,12 +132,15 @@ endif
>   
>   # User can optionally provide a TARGETS skiplist. By default we skip
>   # targets using BPF since it has cutting edge build time dependencies
> -# which require more effort to install.
> +# If user provide custom TARGETS, we just ignore SKIP_TARGETS so that
> +# user can easy to test single target which defined in SKIP_TARGETS
>   SKIP_TARGETS ?= bpf sched_ext
>   ifneq ($(SKIP_TARGETS),)
> +ifneq ($(words $(TARGETS)), 1)
>   	TMP := $(filter-out $(SKIP_TARGETS), $(TARGETS))
>   	override TARGETS := $(TMP)
>   endif
> +endif
>   
>   # User can set FORCE_TARGETS to 1 to require all targets to be successfully
>   # built; make will fail if any of the targets cannot be built. If
> 
> base-commit: 67b6d342fb6d5abfbeb71e0f23141b9b96cf7bb1

thanks,
-- Shuah