[PATCH net] selftests: drv-net: rss_ctx: Add short delay between per-context traffic checks
A few packets may still be sent and received during the termination of the iperf processes. These late packets cause failures when they arrive on queues expected to be empty.
Add a one second delay between repeated _send_traffic_check() calls in rss_ctx tests to ensure such packets are processed before the next traffic checks are performed.
Example failure observed:
Check failed 2 != 0 traffic on inactive queues (context 1): [0, 0, 1, 1, 386385, 397196, 0, 0, 0, 0, ...]
Check failed 4 != 0 traffic on inactive queues (context 2): [0, 0, 0, 0, 2, 2, 247152, 253013, 0, 0, ...]
Check failed 2 != 0 traffic on inactive queues (context 3): [0, 0, 0, 0, 0, 0, 1, 1, 282434, 283070, ...]
Note: While the `noise` parameter could be used to tolerate these late packets, it would be inappropriate here. `noise` tolerates far more traffic than acceptable in this case, risking false positives. Inactive queues are supposed to see zero traffic.
Fixes: 847aa551fa78 ("selftests: drv-net: rss_ctx: factor out send traffic and check") Reviewed-by: Gal Pressman gal@nvidia.com Reviewed-by: Carolina Jubran cjubran@nvidia.com Signed-off-by: Nimrod Oren noren@nvidia.com --- tools/testing/selftests/drivers/net/hw/rss_ctx.py | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/tools/testing/selftests/drivers/net/hw/rss_ctx.py b/tools/testing/selftests/drivers/net/hw/rss_ctx.py index 7bb552f8b182..19be69227693 100755 --- a/tools/testing/selftests/drivers/net/hw/rss_ctx.py +++ b/tools/testing/selftests/drivers/net/hw/rss_ctx.py @@ -4,6 +4,7 @@ import datetime import random import re +import time from lib.py import ksft_run, ksft_pr, ksft_exit from lib.py import ksft_eq, ksft_ne, ksft_ge, ksft_in, ksft_lt, ksft_true, ksft_raises from lib.py import NetDrvEpEnv @@ -492,6 +493,7 @@ def test_rss_context(cfg, ctx_cnt=1, create_with_cfg=None): { 'target': (2+i*2, 3+i*2), 'noise': (0, 1), 'empty': list(range(2, 2+i*2)) + list(range(4+i*2, 2+2*ctx_cnt)) }) + time.sleep(1)
if requested_ctx_cnt != ctx_cnt: raise KsftSkipEx(f"Tested only {ctx_cnt} contexts, wanted {requested_ctx_cnt}") @@ -559,6 +561,7 @@ def test_rss_context_out_of_order(cfg, ctx_cnt=4): }
_send_traffic_check(cfg, ports[i], f"context {i}", expected) + time.sleep(1)
# Use queues 0 and 1 for normal traffic ethtool(f"-X {cfg.ifname} equal 2")
On Sun, 29 Jun 2025 14:18:12 +0300 Nimrod Oren wrote:
A few packets may still be sent and received during the termination of the iperf processes. These late packets cause failures when they arrive on queues expected to be empty.
Add a one second delay between repeated _send_traffic_check() calls in rss_ctx tests to ensure such packets are processed before the next traffic checks are performed.
Sprinklings sleeps should be last resort. Is there a way to wait for iperf to shut down cleanly, or wait for the socket to be closed fully? Like wait_port_listen() ?
On 02/07/2025 3:23, Jakub Kicinski wrote:
On Sun, 29 Jun 2025 14:18:12 +0300 Nimrod Oren wrote:
A few packets may still be sent and received during the termination of the iperf processes. These late packets cause failures when they arrive on queues expected to be empty.
Add a one second delay between repeated _send_traffic_check() calls in rss_ctx tests to ensure such packets are processed before the next traffic checks are performed.
Sprinklings sleeps should be last resort. Is there a way to wait for iperf to shut down cleanly, or wait for the socket to be closed fully? Like wait_port_listen() ?
The socket may end up in TIME_WAIT state, so waiting for it to be fully closed can take ~2 mins. We could opt for a single sleep during GenerateTraffic.stop(), but that would also affect tests that aren't impacted by late packets. WDYT?
On Wed, 9 Jul 2025 20:10:14 +0300 Nimrod Oren wrote:
On 02/07/2025 3:23, Jakub Kicinski wrote:
On Sun, 29 Jun 2025 14:18:12 +0300 Nimrod Oren wrote:
A few packets may still be sent and received during the termination of the iperf processes. These late packets cause failures when they arrive on queues expected to be empty.
Add a one second delay between repeated _send_traffic_check() calls in rss_ctx tests to ensure such packets are processed before the next traffic checks are performed.
Sprinklings sleeps should be last resort. Is there a way to wait for iperf to shut down cleanly, or wait for the socket to be closed fully? Like wait_port_listen() ?
The socket may end up in TIME_WAIT state, so waiting for it to be fully closed can take ~2 mins.
TIME_WAIT is as good as CLOSED for our purposes. Once we got a FIN the chances of more traffic should be minuscule.
linux-kselftest-mirror@lists.linaro.org
-
Jakub Kicinski -
Nimrod Oren