On 9/26/26 06:31, Masami Hiramatsu (Google) wrote:
On Fri, 25 Sep 2026 16:35:41 +0300
Eva Kurchatova <[email protected]> wrote:
kprobe_eventname.tc takes the last kallsyms entry whose name ends in
vfs_read and puts a probe on it. On a kernel built with
CONFIG_GCOV_PROFILE_ALL=y the counter symbol gcc emits for the function
sorts after it:
ffffffff819bd1e0 T vfs_read
ffffffff845013e0 t __gcov_.vfs_read
so the test picks __gcov_.vfs_read, which is not a function entry, and
the write fails:
kprobe_eventname.tc: line 10: echo: write error: Invalid argument
Leave those symbols out of the choice. The test then picks vfs_read and
passes; without a coverage build nothing changes, as no such symbol
exists there.
This is the same class of problem as commit 23a4b108accc
("tracing/selftests: Fix kprobe event name test for .isra. functions"),
where a compiler-generated symbol was picked instead of a real function.
Oops, good catch! This looks good to me.
Acked-by: Masami Hiramatsu (Google) <[email protected]>
Shuah, can you pick this fix via your selftest tree? I think this is a
kind of selftest fix/improvement.
Thanks,
Signed-off-by: Eva Kurchatova <[email protected]>
Applied to linux-kselftest next branch for the next merge window.
thanks,
-- Shuah
