On 8/8/22 4:25 PM, Andrii Nakryiko wrote:
On Sun, Aug 7, 2022 at 10:51 AM Yonghong Song <yhs@xxxxxx> wrote:
Add two program tests with kfunc return types u8/s16.
With previous patch, xlated codes looks like below:
...
; return bpf_kfunc_call_test4((struct sock *)sk, (1 << 16) + 0xff00, (1 << 16) + 0xff);
5: (bf) r1 = r0
6: (b4) w2 = 130816
7: (b4) w3 = 65791
8: (85) call bpf_kfunc_call_test4#8931696
9: (67) r0 <<= 48
10: (c7) r0 s>>= 48
11: (bc) w6 = w0
; }
12: (bc) w0 = w6
13: (95) exit
...
; return bpf_kfunc_call_test5((struct sock *)sk, (1 << 8) + 1, (1 << 8) + 2);
5: (bf) r1 = r0
6: (b4) w2 = 257
7: (b4) w3 = 258
8: (85) call bpf_kfunc_call_test5#8931712
9: (67) r0 <<= 56
10: (77) r0 >>= 56
11: (bc) w6 = w0
; }
12: (bc) w0 = w6
13: (95) exit
For return type s16, proper sign extension for the return value is done
for kfunc bpf_kfunc_call_test4(). For return type s8, proper zero
extension for the return value is done for bpf_kfunc_call_test5().
Without the previous patch, the test kfunc_call will fail with
...
test_main:FAIL:test4-retval unexpected test4-retval: actual 196607 != expected 4294967295
...
test_main:FAIL:test5-retval unexpected test5-retval: actual 515 != expected 3
Signed-off-by: Yonghong Song <yhs@xxxxxx>
---
net/bpf/test_run.c | 12 +++++++
.../selftests/bpf/prog_tests/kfunc_call.c | 10 ++++++
.../selftests/bpf/progs/kfunc_call_test.c | 32 +++++++++++++++++++
3 files changed, 54 insertions(+)
diff --git a/net/bpf/test_run.c b/net/bpf/test_run.c
index cbc9cd5058cb..3a17ab4107f5 100644
--- a/net/bpf/test_run.c
+++ b/net/bpf/test_run.c
@@ -551,6 +551,16 @@ struct sock * noinline bpf_kfunc_call_test3(struct sock *sk)
return sk;
}
+s16 noinline bpf_kfunc_call_test4(struct sock *sk, u32 a, u32 b)
+{
+ return a + b;
+}
+
+u8 noinline bpf_kfunc_call_test5(struct sock *sk, u32 a, u32 b)
+{
+ return a + b;
+}
Is there any upside of adding this to net/bpf/test_run.c instead of
defining it in bpf_testmod?
I put these two functions in test_run.c since bpf_kfunc_call_test{1,2,3}
are defined here and they easily fit the existing kfunc_call testing code.
But yes, I just checked the bpf_testmod.c. Looks like I am able
to define kfunc's in bpf_testmod. Will respin in v2 with this change.
+
struct prog_test_member1 {
int a;
};
[...]
