selftests/bpf: Switch test_vmlinux to use hrtimer_range_start_ns.
Subsystems (5)
PATCH v2 1/1 selftests/bpf: Switch test_vmlinux to use hrtimer_range_start_ns.
The test_vmlinux test uses hrtimer_nanosleep as hook to test tracing programs. But in a kernel built by clang, which performs more aggresive inlining, that function gets inlined into its caller SyS_nanosleep. Therefore, even though fentry and kprobe do hook on the function, they aren't triggered by the call to nanosleep in the test.
A possible fix is switching to use a function that is less likely to be inlined, such as hrtimer_range_start_ns. The EXPORT_SYMBOL functions shouldn't be inlined based on the description of [1], therefore safe to use for this test. Also the arguments of this function include the duration of sleep, therefore suitable for test verification.
[1] af3b56289be1 time: don't inline EXPORT_SYMBOL functions
Tested:
In a clang build kernel, before this change, the test fails:
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 switching 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
---
Changelog since v1:
- More accurate commit messages
tools/testing/selftests/bpf/progs/test_vmlinux.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-)
tools/testing/selftests/bpf/progs/test_vmlinux.c +8 -8
@@ -63,20 +63,20 @@ int BPF_PROG(handle__tp_btf, struct pt_regs *regs, long id)return 0 ; } -SEC("kprobe/hrtimer_nanosleep") -int BPF_KPROBE (handle__kprobe , -ktime_t rqtp,enum hrtimer_mode mode,clockid_t clockid) +SEC("kprobe/hrtimer_start_range_ns") +int BPF_KPROBE (handle__kprobe ,struct hrtimer *timer,ktime_t tim,u64 delta_ns, +const enum hrtimer_mode mode) { -if (rqtp ==MY_TV_NSEC ) +if (tim ==MY_TV_NSEC ) kprobe_called = true;return 0 ; } -SEC("fentry/hrtimer_nanosleep") -int BPF_PROG (handle__fentry , -ktime_t rqtp,enum hrtimer_mode mode,clockid_t clockid) +SEC("fentry/hrtimer_start_range_ns") +int BPF_PROG (handle__fentry ,struct hrtimer *timer,ktime_t tim,u64 delta_ns, +const enum hrtimer_mode mode) { -if (rqtp ==MY_TV_NSEC ) +if (tim ==MY_TV_NSEC ) fentry_called = true;return 0 ; }
Signature · 1 line
2.27.0.212.ge8ba1cc988-goog
Quoted 35 lines · Hao Luo
The
test_vmlinuxtest useshrtimer_nanosleepas hook to test tracing programs. But in a kernel built by clang, which performs more aggresive inlining, that function gets inlined into its callerSyS_nanosleep. Therefore, even though fentry and kprobe do hook on the function, they aren't triggered by the call to nanosleep in the test.A possible fix is switching to use a function that is less likely to be inlined, such as
hrtimer_range_start_ns. TheEXPORT_SYMBOLfunctions shouldn't be inlined based on the description of [1], therefore safe to use for this test. Also the arguments of this function include the duration of sleep, therefore suitable for test verification.[1]
af3b56289be1time: don't inlineEXPORT_SYMBOLfunctionsTested:
In a clang build kernel, before this change, the test fails: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 calledAfter switching 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
Quoted 2 lines
Signed-off-by Hao Luo <[email protected]>Acked-by Andrii Nakryiko <[email protected]>
Thanks!
Alexei Starovoitov
AppliedQuoted 39 lines · Yonghong Song
Quoted 35 lines · Hao Luo
The
test_vmlinuxtest useshrtimer_nanosleepas hook to test tracing programs. But in a kernel built by clang, which performs more aggresive inlining, that function gets inlined into its callerSyS_nanosleep. Therefore, even though fentry and kprobe do hook on the function, they aren't triggered by the call to nanosleep in the test.A possible fix is switching to use a function that is less likely to be inlined, such as
hrtimer_range_start_ns. TheEXPORT_SYMBOLfunctions shouldn't be inlined based on the description of [1], therefore safe to use for this test. Also the arguments of this function include the duration of sleep, therefore suitable for test verification.[1]
af3b56289be1time: don't inlineEXPORT_SYMBOLfunctionsTested:
In a clang build kernel, before this change, the test fails: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 calledAfter switching 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
Quoted 5 lines
Quoted 2 lines
Signed-off-by Hao Luo <[email protected]>Acked-by Andrii Nakryiko <[email protected]>Thanks!
Acked-by Yonghong Song <[email protected]>
Applied. Thanks