> diff --git a/tools/testing/selftests/bpf/prog_tests/btf_show_void.c
> b/tools/testing/selftests/bpf/prog_tests/btf_show_void.c
> new file mode 100644
> index 0000000000000..5468331991958
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/prog_tests/btf_show_void.c
> @@ -0,0 +1,57 @@
> +// SPDX-License-Identifier: GPL-2.0
> +#include <test_progs.h>
> +#include <bpf/btf.h>
> +#include "btf_show_void.skel.h"
> +
> +/*
> + * bpf_snprintf_btf() with the type_id of a "const void" (a modifier that
> + * resolves to void, present in the vmlinux BTF) used to NULL-deref in
> + * btf_modifier_show(). A fixed kernel prints the "<unsupported kind:0>"
> + * placeholder; on an unfixed kernel this oopses the task (and panics it
> under
> + * panic_on_oops), so it doubles as a reproducer.
> + */
> +void test_btf_show_void(void)
A subsystem pattern flags this as potentially concerning: this adds a
new standalone test file pair for one narrow case (a modifier resolving
to void in bpf_snprintf_btf(), exercising the btf_modifier_show() ->
btf_df_show() fallback). The directory already has
prog_tests/snprintf_btf.c (serial_test_snprintf_btf, driving
progs/netif_receive_skb.c which includes progs/btf_ptr.h) that exercises
bpf_snprintf_btf() output formatting for a range of types.
Would this case fit better as an additional case in that existing test
rather than as a new binary/skeleton pair?
The counter-argument: snprintf_btf.c is structured very differently - it
attaches a netif_receive_skb raw tracepoint and triggers it with
system("ping -c 1 127.0.0.1"), and still uses the deprecated CHECK()
macros. The new test instead needs a type_id discovered from the vmlinux
BTF at runtime injected through rodata, and drives the program with
bpf_prog_test_run_opts() rather than a real event. Folding it into the
existing test would mean adding a second program and a second run
mechanism to that file. That is a plausible engineering reason for a
separate file, and the new file is the more modern idiom.
> +{
> + LIBBPF_OPTS(bpf_test_run_opts, topts);
> + const struct btf_type *t;
> + struct btf_show_void *skel;
> + int i, n, cv = 0, err;
> + char ctx[16] = {};
> + struct btf *btf;
> +
> + btf = btf__parse("/sys/kernel/btf/vmlinux", NULL);
> + if (!ASSERT_OK_PTR(btf, "btf__parse vmlinux"))
> + return;
Both environment prerequisites of this test are reported as hard failures
rather than skips.
On a kernel built without CONFIG_DEBUG_INFO_BTF the /sys/kernel/btf/vmlinux
file does not exist, btf__parse() fails with ENOENT, and the test reports
a failure instead of a skip. The path is also hardcoded; btf__load_vmlinux_btf()
(used by 14 other prog_tests, e.g. fill_link_info.c, tracing_multi.c,
btf_module.c) additionally searches /boot/vmlinux-$(uname -r) and the
module build dirs.
> +
> + n = btf__type_cnt(btf);
> + for (i = 1; i < n; i++) {
> + t = btf__type_by_id(btf, i);
> + if (btf_kind(t) == BTF_KIND_CONST && t->type == 0) {
> + cv = i;
> + break;
> + }
> + }
> + if (!ASSERT_GT(cv, 0, "find const void in vmlinux BTF"))
> + goto out_btf;
The test scans the vmlinux BTF for the first BTF_KIND_CONST whose
->type == 0 and hard-fails via ASSERT_GT when none exists. Whether a
standalone const void node survives in a given kernel's BTF is a property
of that build (pahole version, dedup, whether anything in that kernel
actually uses const void *), not a property of the kernel logic under test.
If it is ever absent the test reports a kernel regression that is not
there, with no indication that the input the test needed simply was not
present.
Would a test__skip() with a reason string on both arms (the pattern used
in prog_tests/snprintf_btf.c) keep an environment difference from being
read as a failure?
Mitigating context: all BPF selftests already require CONFIG_DEBUG_INFO_BTF=y,
and the exact idiom btf__parse("/sys/kernel/btf/vmlinux", NULL) followed by
ASSERT_OK_PTR is used unchanged in prog_tests/libbpf_str.c (4 sites),
prog_tests/libbpf_probes.c (2 sites) and prog_tests/btf_sysfs.c, so the
first arm matches existing neighbours. const void is also very likely
present in any vmlinux BTF, since const void * appears in many exported
prototypes. The concern is robustness/diagnosability of the failure mode,
not a functional defect.
> +
> + skel = btf_show_void__open();
> + if (!ASSERT_OK_PTR(skel, "skel_open"))
> + goto out_btf;
> + skel->rodata->const_void_id = cv;
> + if (!ASSERT_OK(btf_show_void__load(skel), "skel_load"))
> + goto out_skel;
> +
> + topts.ctx_in = ctx;
> + topts.ctx_size_in = sizeof(ctx);
> + err =
> bpf_prog_test_run_opts(bpf_program__fd(skel->progs.dump_const_void),
> + &topts);
> + if (!ASSERT_OK(err, "test_run"))
> + goto out_skel;
> +
> + ASSERT_EQ(skel->bss->ret, sizeof("<unsupported kind:0>") - 1, "ret");
> + ASSERT_STREQ(skel->bss->out, "<unsupported kind:0>", "placeholder");
> +out_skel:
> + btf_show_void__destroy(skel);
> +out_btf:
> + btf__free(btf);
> +}
---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md
CI run summary: https://github.com/kernel-patches/bpf/actions/runs/33300136294
