On 9/3/26 10:10 AM, Bochao Cao wrote:
Thanks Daniel, Alexis, and Emil.
Moving the test to tools/testing/selftests/drivers/net/hw/ appears to
require more than relocating the shell script. The xdp_features userspace
tool, BPF program, skeleton build rules, and test environment would also
need to be adapted to the net driver selftest framework.
Would it be acceptable to take the process lifecycle fix as a v3 in its
current location, followed by a separate migration series?
I would also be happy to work on the migration. If you would prefer to
see the migration before this fix is applied, I can prepare v3 as a
small series, keeping the framework migration and the process lifecycle
changes separate where practical.
No objections from my side to take the process lifecycle fix, I would however
suggest that this goes into net-next tree in this case: the BPF CI does not
depend on this script, and if we would merge it into bpf-next then you'd have
to wait for ~8 weeks before the trees sync during the merge window.. that way
via net-next you can continue with the migration and as next step cleanly/
conflict-free remove the test_xdp_features.sh script from BPF selftests and
adding it for net/driver HW selftests.
Emil Tsalapatis <[email protected]> 于2026年9月2日周三 15:44写道：
Emil Tsalapatis <[email protected]> 于2026年9月2日周三 15:44写道：
On Fri Aug 28, 2026 at 5:32 AM EDT, Bochao Cao wrote:
Resending as plain text because my previous reply was rejected by the
vger mailing lists. Sorry for the duplicate.
Thanks Daniel and Alexis.
Dropping the procps dependency is not the primary motivation for
this change. I agree that Debian can remove the dependency independently.
The issue addressed by this patch is process isolation in the standalone
test. The current readiness check may observe an unrelated
concurrent xdp_features process, while cleanup may terminate every
xdp_features
process on the host. In addition, a DUT which exits before listening
can leave the test waiting indefinitely.
Although this script is not currently run by the BPF CI, it remains
useful for testing real hardware,
so these process lifecycle issues can still affect users running
the test manually.
Would it be acceptable to fix these issues in the script's current location?
If the preferred direction is to move it under
tools/testing/selftests/drivers/net/hw/,
should the move be submitted first, with this fix rebased on top?
I can send a v3 that drops the Debian Closes tag and focuses
the commit message solely on the process isolation, timeout,
and cleanup fixes once the preferred location is clear.
Thanks,
Bochao
While I'd defer to Daniel and Alexis on this, imo we could merge the fix
as-is and move the file as a followup since the change is a net gain on
its own.
Wherever we end up putting the file feel free to add:
Reviewed-by: Emil Tsalapatis <[email protected]>
Bochao Cao <[email protected]> 于2026年8月28日周五 14:10写道：
Thanks Daniel and Alexis.
Dropping the procps dependency is not the primary motivation for
this change. I agree that Debian can remove the dependency independently.
The issue addressed by this patch is process isolation in the standalone
test. The current readiness check may observe an unrelated
concurrent xdp_features process, while cleanup may terminate every
xdp_features
process on the host. In addition, a DUT which exits before listening can
leave the test waiting indefinitely.
Although this script is not currently run by the BPF CI, it remains useful
for testing real hardware,
so these process lifecycle issues can still affect users running the test
manually.
Would it be acceptable to fix these issues in the script's current location?
If the preferred direction is to move it under
tools/testing/selftests/drivers/net/hw/,
should the move be submitted first, with this fix rebased on top?
I can send a v3 that drops the Debian Closes tag and focuses
the commit message solely on the process isolation, timeout,
and cleanup fixes once the preferred location is clear.
Thanks,
Bochao
Bochao Cao <[email protected]> 于2026年8月25日周二 21:01写道：
Thanks Daniel and Alexis.
Dropping the procps dependency is not the primary motivation for
this change. I agree that Debian can remove the dependency independently.
The issue addressed by this patch is process isolation in the standalone
test. The current readiness check may observe an unrelated
concurrent xdp_features process, while cleanup may terminate every
xdp_features
process on the host. In addition, a DUT which exits before listening can
leave the test waiting indefinitely.
Although this script is not currently run by the BPF CI, it remains useful
for testing real hardware,
so these process lifecycle issues can still affect users running the test
manually.
Would it be acceptable to fix these issues in the script's current location?
If the preferred direction is to move it under
tools/testing/selftests/drivers/net/hw/,
should the move be submitted first, with this fix rebased on top?
I can send a v3 that drops the Debian Closes tag and focuses
the commit message solely on the process isolation, timeout,
and cleanup fixes once the preferred location is clear.
Thanks,
Bochao
Alexis Lothoré <[email protected]> 于2026年8月24日周一 22:49写道：
Hi Daniel, thanks for the notification
On Mon Aug 24, 2026 at 1:31 PM CEST, Daniel Borkmann wrote:
[ Trimming the excessive Cc list, and adding Alexis ]
On 8/12/26 10:28 AM, Bochao Cao via B4 Relay wrote:
From: Bochao Cao <[email protected]>
test_xdp_features.sh waits for any xdp_features listener to appear and
uses pidof during cleanup. A concurrent test can therefore make another
test proceed before its own DUT is ready, and cleanup kills every
xdp_features process on the host. The readiness loop also has no timeout,
so a DUT that exits before listening leaves the test hung indefinitely.
Track one active DUT at a time, wait for ss to report that exact PID with
a bounded retry loop, and reap it after each test. Consult the shell job
table before signaling the DUT so a stale PID cannot target an unrelated
process. On failure, terminate the shell job with SIGKILL and reap it so
blocked I/O cannot hang cleanup. Install an EXIT trap and signal handlers
so failure paths also remove network setup.
Fixes: 4dba3e7852b7 ("selftests/bpf: introduce XDP compliance test tool")
Closes: https://bugs.debian.org/1136522
Signed-off-by: Bochao Cao <[email protected]>
---
Tests:
- bash -n tools/testing/selftests/bpf/test_xdp_features.sh
- make -C tools/testing/selftests/bpf xdp_features
- sudo tools/testing/selftests/bpf/test_xdp_features.sh
- verified cleanup terminates a blocked DUT without affecting an unrelated
process
---
Changes in v2:
- Clarify that avoiding name-wide process matching, rather than dropping a
dependency, is the motivation.
- Track and reap one active DUT at a time instead of retaining historical PIDs.
- Address PID reuse by signaling only the current Bash job during cleanup.
- Use SIGKILL on failure cleanup so blocked DUT I/O cannot hang wait
indefinitely.
- Link to v1:
https://patch.msgid.link/20260805-xdp-dut-process-lifecycle-gmail-v1-1-45984df8d...@gmail.com
---
tools/testing/selftests/bpf/test_xdp_features.sh | 82
++++++++++++++++++------
1 file changed, 62 insertions(+), 20 deletions(-)
Sorry for the late reply. With regards to https://bugs.debian.org/1136522,
src:linux deb
does not have to depend on this at all, so the src:linux can just get rid of
procps in
any case if this is indeed the last dependency. I'm not seeing the test being
run in our
BPF CI. I've Cc'ed Alexis as he's in the process of migrating and/or removing
tests from
tools/testing/selftests/bpf/ depending on how they fit into test_progs
framework. I'll
let him comment if there is already work in progress. It feels like this script
could be
reworked into tools/testing/selftests/drivers/net/hw/ tests and removed
altogether from
the tools/testing/selftests/bpf/ dir.
There has been an attempt to fully convert and get rid of
test_xdp_features.sh, but discussions around the corresponding series
highlighted the need for the script to remain available for testing on
real hardware. Features covered by the test_xdp_features.sh that were
not covered yet by test_progs have been added to test_progs (in
xdp_cpumap_attach), see [1]. So there's currently no active effort on
this one on my side.
Alexis
[1]
https://lore.kernel.org/bpf/[email protected]/
--
Alexis Lothoré, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
