Reading bionic commit 6e096fe0027f723f746f8179503784cb3579294a
("bpf/tests: Do not PASS tests without actually testing the result"),
and looking at the specific failing tests, it looks to be the case that
the test change has shown these failures starting now. Also included in
focal.
There is a
Tested on P8 node entei with 4.15.0-162-generic
Kernel 4.15.0-162-generic + Source 4.15.0-162-generic
[12008.662733] test_bpf: #145 ALU_MOD_K: 3 % 1 = 0 jited:1 ret 3 != 0 FAIL (1
times)
[12008.663071] test_bpf: #148 ALU64_MOD_K: 3 % 1 = 0 jited:1 ret 3 != 0 FAIL (1
times)
[12012.150242]
These are the BPF related changes between 5.4.0-89 and 5.4.0-90. Nothing
where I would immediately think of causing division to go wrong...
git log Ubuntu-5.4.0-89.100..origin/master-prep --grep="bpf" --oneline
760480fda833 selftests/bpf: Enlarge select() timeout for test_maps
5c85b4d0b051
BTW here is the test summary, it looks like we have 2 cases removed in
5.4.0-90. (in lib/test
Test results:
* Kernel: 5.4.0-89 + Source: 5.4.0-89 = Summary: 378 PASSED, 0 FAILED, [366/366
JIT'ed]
* Kernel: 5.4.0-89 + Source: 5.4.0-90 = Summary: 378 PASSED, 0 FAILED, [366/366
JIT'ed]
* Kernel:
Manually tested on a P8 node (gulpin) with 5.4.0-89-generic and
5.4.0-90-generic
Conclusion:
It looks like a real kernel regression.
The following two test cases are failing with 5.4.0-90-generic (passed with
5.4.0-89-generic)
[ 736.795192] test_bpf: #142 ALU_MOD_K: 3 % 1 = 0 jited:1 ret 3 != 0