On Tue, Jun 30, 2020 at 1:47 PM Hao Luo <hao...@google.com> wrote:
>
> The test_vmlinux test uses hrtimer_nanosleep as hook to test tracing
> programs. But it seems Clang may have done an aggressive optimization,
> causing fentry and kprobe to not hook on this function properly on a
> Clang build kernel.
>
> A possible fix is switching to use a more reliable function, e.g. the
> ones exported to kernel modules such as hrtimer_range_start_ns. After
> we switch to using hrtimer_range_start_ns, the test passes again even
> on a clang build kernel.
>
> Tested:
> In a clang build kernel, the test fail even when the flags
> {fentry, kprobe}_called are set unconditionally in handle__kprobe()
> and handle__fentry(), which implies the programs do not hook on
> hrtimer_nanosleep() properly. This could be because clang's code
> transformation is too aggressive.
>
> test_vmlinux:PASS:skel_open 0 nsec
> test_vmlinux:PASS:skel_attach 0 nsec
> test_vmlinux:PASS:tp 0 nsec
> test_vmlinux:PASS:raw_tp 0 nsec
> test_vmlinux:PASS:tp_btf 0 nsec
> test_vmlinux:FAIL:kprobe not called
> test_vmlinux:FAIL:fentry not called
>
> After we switch to hrtimer_range_start_ns, the test passes.
>
> test_vmlinux:PASS:skel_open 0 nsec
> test_vmlinux:PASS:skel_attach 0 nsec
> test_vmlinux:PASS:tp 0 nsec
> test_vmlinux:PASS:raw_tp 0 nsec
> test_vmlinux:PASS:tp_btf 0 nsec
> test_vmlinux:PASS:kprobe 0 nsec
> test_vmlinux:PASS:fentry 0 nsec
>
> Signed-off-by: Hao Luo <hao...@google.com>
> ---
Took me a bit of jumping around to find how it is related to nanosleep
call :) But seems like it's unconditionally called, so should be fine.
Acked-by: Andrii Nakryiko <andr...@fb.com>
> tools/testing/selftests/bpf/progs/test_vmlinux.c | 16 ++++++++--------
> 1 file changed, 8 insertions(+), 8 deletions(-)
>
[...]
