Re: [PATCH bpf-next v3 00/10] Local vmtest enhancement and RV64 enabled😁

["Björn Töpel" <bjorn@kernel.org>]

Lehui, Daniel!

Pu Lehui <pulehui@huaweicloud.com> writes:

> On 2024/9/6 2:52, Daniel Borkmann wrote:
>> On 9/5/24 10:13 AM, Pu Lehui wrote:
>>> Patch 1-3 fix some problem about bpf selftests. Patch 4 add local rootfs
>>> image support for vmtest. Patch 5 enable cross-platform testing for
>>> vmtest. Patch 6-10 enable vmtest on RV64.
>>>
>>> We can now perform cross platform testing for riscv64 bpf using the
>>> following command:
>>>
>>> PLATFORM=riscv64 CROSS_COMPILE=riscv64-linux-gnu- \
>>>    tools/testing/selftests/bpf/vmtest.sh \
>>>    -l <path of local rootfs image> -- \
>>>    ./test_progs -d \
>>>        \"$(cat tools/testing/selftests/bpf/DENYLIST.riscv64 \
>>>            | cut -d'#' -f1 \
>>>            | sed -e 's/^[[:space:]]*//' \
>>>                  -e 's/[[:space:]]*$//' \
>>>            | tr -s '\n' ',' \
>>>        )\"
>>>
>>> For better regression, we rely on commit [0]. And since the work of riscv
>>> ftrace to remove stop_machine atomic replacement is in progress, we also
>>> need to revert commit [1] [2].
>>>
>>> The test platform is x86_64 architecture, and the versions of relevant
>>> components are as follows:
>>>      QEMU: 8.2.0
>>>      CLANG: 17.0.6 (align to BPF CI)
>>>      ROOTFS: ubuntu noble (generated by [3])
>>>
>>> Link: 
>>> https://lore.kernel.org/all/20240831071520.1630360-1-pulehui@huaweicloud.com/ [0]
>>> Link: 
>>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=3308172276db [1]
>>> Link: 
>>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=7caa9765465f [2]
>>> Link: https://github.com/libbpf/ci/blob/main/rootfs/mkrootfs_debian.sh 
>>> [3]
>> 
>> Nice work! Next step is upstream BPF CI integration? :)
>
> CC Björn😁

Indeed, very nice work! Every year is "The year of RISC-V BPF CI
integration". :-P

> Yeah, that's what we're most looking forward to and we've been trying to 
> move forward with that. There are currently several options, but they 
> are not very suitable yet.
>
> 1. Cross-platform testing with subset of tests (test_verifier + 
> test_progs), it will cost a bit more time.
>
> x86_64 host:
> Summary: 536/3594 PASSED, 68 SKIPPED, 0 FAILED
> real    30m 18.88s
> user    6m 52.97s
> sys     21m 3.03s
>
> 2. Cross-platform testing will parallel mode, it will meet flaky 
> problems while the time consume looks good.
>
> x86_64 host:
> real    7m 45.42s
> user    6m 13.59s
> sys     15m 41.12s
>
> 3. Real board testing, which relies on Hypervisor Extension to enable 
> kvm on qemu. We are still trying to find a suitable board.

There's a board coming out soonish with H -- let's hope it doesn't suck!
;-)

I have a CI running, that's runs all the BPF tests (and the rest of
kselftest) on various trees/branches, and it takes *hours* on QEMU TCG
[1] (the GH CI doesn't report fail/ok in the UI, so you need to download
the logs [2] -- filename *kselftest-bpf*).

Obviously this is a no-go for pre-commit/patchwork CI, but for, say,
a longer release test (post-commit), spending a couple of hours on a
test would probably be OK.

If there's a post-commit CI running somewhere, maybe we could plug in
RISC-V BPF QEMU TCG tests there? ...and then when we get proper H
machines, we can add RISC-V to Meta's PW CI as well.

Daniel, is there a "release test CI" running, or is that mostly manual
work?

Björn


[1] https://github.com/linux-riscv/linux-riscv/actions/runs/10725237865/job/29742632222#step:5:30
[2] https://github.com/linux-riscv/linux-riscv/actions/runs/10725237865/artifacts/1899421897