Re: [PATCH v1 1/2] perf build: Support building BPF skeletons with bpf-gcc

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] perf build: bpf-gcc version check bypassed due to mutually exclusive GCC arguments
--

commit b2c0a31ec137ebce8b46644a6edfe507e7c3b298
Author: Ian Rogers <irogers@google.com>

perf build: Support building BPF skeletons with bpf-gcc

This commit supports building perf BPF skeletons with bpf-gcc (>= 14.0.0)
in addition to clang. It updates the build system to prefer bpf-gcc when
building with GCC, and falls back to Clang if bpf-gcc is unavailable or
too old.

> diff --git a/tools/perf/Makefile.config b/tools/perf/Makefile.config
> --- a/tools/perf/Makefile.config
> +++ b/tools/perf/Makefile.config
[ ... ]
> +    define check_bpf_gcc
> +      ifneq ($$(call get-executable,$$(BPF_GCC)),)
> +        BPF_GCC_VERSION := $$(shell $$(BPF_GCC) -dumpfullversion -dumpversion 2>/dev/null)

[Severity: Medium]
Does passing both -dumpfullversion and -dumpversion simultaneously cause
GCC to fail with a fatal error?

If GCC rejects these mutually exclusive arguments, the error output is
suppressed by 2>/dev/null, resulting in BPF_GCC_VERSION evaluating to an
empty string.

> +        ifneq ($$(call version-lt3,$$(BPF_GCC_VERSION),14.0.0),1)
> +          have_bpf_gcc := 1
> +        endif

When version-lt3 is called with an empty string against 14.0.0, the
underlying awk expression evaluates this as false (0). The outer check
ifneq (0,1) then evaluates to true, unconditionally setting
have_bpf_gcc := 1.

Can this regression result in the version check being completely bypassed,
allowing older, unsupported versions of bpf-gcc to be incorrectly
permitted to build BPF skeletons instead of properly falling back to Clang?

> +      endif
> +    endef

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261013221643.1815686-1-irogers@google.com?part=1