On Thu Aug 6, 2026 at 2:49 PM -03, Sasha Levin wrote:
> On Wed, Aug 05, 2026 at 02:54:34PM -0300, Ricardo B. Marlière (SUSE) wrote:
>> Hi, backporting commit 23b933c6223f ("bpf: Verifier support for sleepable
>> tracepoint programs") without backporting this commit breaks the selftest,
>> as seen in openSUSE CI:
>
> Agreed on the diagnosis: 23b933c6223f went into 7.1.y as a Stable-dep-of
> and changed the verifier error string without updating the expectation,
> so the verifier/sleepable.c hunk is both correct and needed there.
>
> The rest of the commit is the problem:
>
>> .../bpf/prog_tests/sleepable_tracepoints.c | 142
>> +++++++++++++++++++++
>> .../bpf/progs/test_sleepable_tracepoints.c | 112 ++++++++++++++++
>> .../bpf/progs/test_sleepable_tracepoints_fail.c | 18 +++
>> tools/testing/selftests/bpf/verifier/sleepable.c | 17 ++-
>
> The three new files use SEC("tp_btf.s/..."), SEC("raw_tp.s/...") and
> SEC("tp.s/..."). 7.1.y does not carry 0cd420a6f40c7 ("libbpf: Add section
> handlers for sleepable tracepoints"), so its libbpf has no SEC_DEF entry
> for any ".s" tracepoint variant, and sec_def_matches() will not match a
> ".s" suffix against the plain "tp_btf+"/"raw_tp+"/"tp+" entries. Every
> program in the two new progs/ files then fails to load and
> test_sleepable_tracepoints__open_and_load() fails outright - so as posted
> this trades a test_verifier failure for a permanently failing test_progs
> test.
>
> The runtime pieces are missing from 7.1.y as well - 439ebd5b5708 ("bpf:
> Add sleepable support for raw tracepoint programs"), 12628ffaf98b ("bpf:
> Add bpf_prog_run_array_sleepable()") and 57918341dd19 ("bpf: Add
> sleepable support for classic tracepoint programs") - so even with the
> libbpf section handlers the functional subtests would still fail.
>
> Could you send a v2 carrying only the
> tools/testing/selftests/bpf/verifier/sleepable.c hunk, with a note in the
> commit message explaining why the new prog_tests/ and progs/ files are
> omitted?
Will do, thanks for the analysis and review.
