[PATCH net-next 1/2] selftests: drv-net: pace ethtool_std_stats packet generation

[PATCH net-next 1/2] selftests: drv-net: pace ethtool_std_stats packet generation

[Jakub Kicinski]

mausezahn defaults to sending packets back to back at the maximum rate,
which can cause packet loss, especially if receiver is running a debug
kernel. Space the generated packets out (-d 10usec), like ethtool_rmon
already does.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
CC: andrew@lunn.ch
CC: shuah@kernel.org
CC: petrm@nvidia.com
CC: ioana.ciornei@nxp.com
CC: linux-kselftest@vger.kernel.org
---
 tools/testing/selftests/drivers/net/hw/ethtool_std_stats.sh | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/testing/selftests/drivers/net/hw/ethtool_std_stats.sh b/tools/testing/selftests/drivers/net/hw/ethtool_std_stats.sh
index c085d2a4c989..1b329b3f60c2 100755
--- a/tools/testing/selftests/drivers/net/hw/ethtool_std_stats.sh
+++ b/tools/testing/selftests/drivers/net/hw/ethtool_std_stats.sh
@@ -43,10 +43,10 @@ traffic_test()
 	done
 
 	# shellcheck disable=SC2086 # needs split options
-	run_on "$iface" "$MZ" "$iface" -q -c "$num_tx" $pkt_format
+	run_on "$iface" "$MZ" "$iface" -q -d 10usec -c "$num_tx" $pkt_format
 
 	# shellcheck disable=SC2086 # needs split options
-	run_on "$neigh" "$MZ" "$neigh" -q -c "$num_rx" $pkt_format
+	run_on "$neigh" "$MZ" "$neigh" -q -d 10usec -c "$num_rx" $pkt_format
 
 	for i in "${!counters[@]}"; do
 		read -r int grp cnt target exact_check xfail_message \
-- 
2.55.0

[PATCH net-next 2/2] selftests: drv-net: let ethtool stats settle before reading

[Jakub Kicinski]

Some devices refresh the statistics exposed via ethtool only
periodically, every stats-block-usecs (as reported by ethtool -c).
ethtool_std_stats and ethtool_rmon sample the counters immediately
after generating traffic, so on such devices they can read stale
values and fail with a delta short of the packets just sent.

Add a hw_stats_settle() helper which sleeps for 1.25x the configured
stats-block-usecs (defaulting to 20ms when the device reports no, or
a zero, period). Use it for ethtool std stats and RMON.
The 1.25x/20msec heuristic matches what the Python tests do.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
CC: andrew@lunn.ch
CC: shuah@kernel.org
CC: dsahern@kernel.org
CC: idosch@nvidia.com
CC: ioana.ciornei@nxp.com
CC: petrm@nvidia.com
CC: linux-kselftest@vger.kernel.org
---
 .../selftests/drivers/net/hw/ethtool_rmon.sh      |  2 ++
 .../selftests/drivers/net/hw/ethtool_std_stats.sh |  2 ++
 tools/testing/selftests/net/forwarding/lib.sh     | 15 +++++++++++++++
 3 files changed, 19 insertions(+)

diff --git a/tools/testing/selftests/drivers/net/hw/ethtool_rmon.sh b/tools/testing/selftests/drivers/net/hw/ethtool_rmon.sh
index 2ec19edddfaa..a074834cbe59 100755
--- a/tools/testing/selftests/drivers/net/hw/ethtool_rmon.sh
+++ b/tools/testing/selftests/drivers/net/hw/ethtool_rmon.sh
@@ -65,6 +65,8 @@ bucket_test()
 	run_on "$iface" \
 		"$MZ" "$iface" -q -c "$num_tx" -p "$len" -a own -b bcast -d 10us
 
+	hw_stats_settle "$iface"
+
 	after=$(run_on "$iface" ethtool --json -S "$iface" --groups rmon | \
 		jq -r ".[0].rmon[\"${set}-pktsNtoM\"][$bucket].val")
 
diff --git a/tools/testing/selftests/drivers/net/hw/ethtool_std_stats.sh b/tools/testing/selftests/drivers/net/hw/ethtool_std_stats.sh
index 1b329b3f60c2..09f8128c51f3 100755
--- a/tools/testing/selftests/drivers/net/hw/ethtool_std_stats.sh
+++ b/tools/testing/selftests/drivers/net/hw/ethtool_std_stats.sh
@@ -48,6 +48,8 @@ traffic_test()
 	# shellcheck disable=SC2086 # needs split options
 	run_on "$neigh" "$MZ" "$neigh" -q -d 10usec -c "$num_rx" $pkt_format
 
+	hw_stats_settle "$int"
+
 	for i in "${!counters[@]}"; do
 		read -r int grp cnt target exact_check xfail_message \
 			<<< "${counters[$i]}"
diff --git a/tools/testing/selftests/net/forwarding/lib.sh b/tools/testing/selftests/net/forwarding/lib.sh
index ac8358bcb22c..05acd4011456 100644
--- a/tools/testing/selftests/net/forwarding/lib.sh
+++ b/tools/testing/selftests/net/forwarding/lib.sh
@@ -406,6 +406,21 @@ get_ifname_by_ip()
 	__run_on "$target" ip -j addr show to "$ip_addr" | jq -r '.[].ifname'
 }
 
+# Wait for the device to refresh its HW statistics. Devices latch the stats
+# reported via ethtool only every stats-block-usecs, so sample after that.
+hw_stats_settle()
+{
+	local iface=$1; shift
+	local usecs
+
+	# Match only a non-zero integer; 0 or "n/a" use default (20msec)
+	usecs=$(run_on "$iface" ethtool -c "$iface" 2>/dev/null | \
+		sed -n 's/^stats-block-usecs:[[:space:]]*\([1-9][0-9]*\)$/\1/p')
+	usecs=${usecs:-20000}
+
+	sleep "$(echo "$usecs * 1.25 / 1000 / 1000" | bc -l)"
+}
+
 # Whether the test is conforming to the requirements and usage described in
 # drivers/net/README.rst.
 : "${DRIVER_TEST_CONFORMANT:=no}"
-- 
2.55.0

Re: [PATCH net-next 2/2] selftests: drv-net: let ethtool stats settle before reading

[Andrew Lunn]

On Sat, Aug 08, 2026 at 09:36:53AM -0700, Jakub Kicinski wrote:
> Some devices refresh the statistics exposed via ethtool only
> periodically, every stats-block-usecs (as reported by ethtool -c).

Interesting. I did not know that existed. Which also suggests not many
drivers actually set it.

I will keep an eye out for drivers which should set this, but don't.

  Andrew

Re: [PATCH net-next 1/2] selftests: drv-net: pace ethtool_std_stats packet generation

[Jakub Kicinski]

On Mon, 10 Aug 2026 11:48:32 +0200 Petr Machata wrote:
> Jakub Kicinski <kuba@kernel.org> writes:
> > mausezahn defaults to sending packets back to back at the maximum rate,
> > which can cause packet loss, especially if receiver is running a debug
> > kernel. Space the generated packets out (-d 10usec), like ethtool_rmon
> > already does.
> >
> > Signed-off-by: Jakub Kicinski <kuba@kernel.org>  
> 
> Reviewed-by: Petr Machata <petrm@nvidia.com>

FTR Petr send me a review tag on this off-list ;)

Re: [PATCH net-next 1/2] selftests: drv-net: pace ethtool_std_stats packet generation

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Sat,  8 Aug 2026 09:36:52 -0700 you wrote:
> mausezahn defaults to sending packets back to back at the maximum rate,
> which can cause packet loss, especially if receiver is running a debug
> kernel. Space the generated packets out (-d 10usec), like ethtool_rmon
> already does.
> 
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
> 
> [...]

Here is the summary with links:
  - [net-next,1/2] selftests: drv-net: pace ethtool_std_stats packet generation
    https://git.kernel.org/netdev/net-next/c/f4cf60568747
  - [net-next,2/2] selftests: drv-net: let ethtool stats settle before reading
    https://git.kernel.org/netdev/net-next/c/855631afe0f5

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html

Re: [PATCH net-next 1/2] selftests: drv-net: pace ethtool_std_stats packet generation

[Petr Machata]

Jakub Kicinski <kuba@kernel.org> writes:

> On Mon, 10 Aug 2026 11:48:32 +0200 Petr Machata wrote:
>> Jakub Kicinski <kuba@kernel.org> writes:
>> > mausezahn defaults to sending packets back to back at the maximum rate,
>> > which can cause packet loss, especially if receiver is running a debug
>> > kernel. Space the generated packets out (-d 10usec), like ethtool_rmon
>> > already does.
>> >
>> > Signed-off-by: Jakub Kicinski <kuba@kernel.org>  
>> 
>> Reviewed-by: Petr Machata <petrm@nvidia.com>
>
> FTR Petr send me a review tag on this off-list ;)

Oops, that was not the intention.
My mailer setup is currently in a bit of a flux :-|

Re: [PATCH net-next 2/2] selftests: drv-net: let ethtool stats settle before reading

[David Ahern]

On 8/8/26 10:36 AM, Jakub Kicinski wrote:
> Some devices refresh the statistics exposed via ethtool only
> periodically, every stats-block-usecs (as reported by ethtool -c).
> ethtool_std_stats and ethtool_rmon sample the counters immediately
> after generating traffic, so on such devices they can read stale
> values and fail with a delta short of the packets just sent.
> 
> Add a hw_stats_settle() helper which sleeps for 1.25x the configured
> stats-block-usecs (defaulting to 20ms when the device reports no, or
> a zero, period). Use it for ethtool std stats and RMON.
> The 1.25x/20msec heuristic matches what the Python tests do.
> 
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
> ---
> CC: andrew@lunn.ch
> CC: shuah@kernel.org
> CC: dsahern@kernel.org
> CC: idosch@nvidia.com
> CC: ioana.ciornei@nxp.com
> CC: petrm@nvidia.com
> CC: linux-kselftest@vger.kernel.org
> ---
>  .../selftests/drivers/net/hw/ethtool_rmon.sh      |  2 ++
>  .../selftests/drivers/net/hw/ethtool_std_stats.sh |  2 ++
>  tools/testing/selftests/net/forwarding/lib.sh     | 15 +++++++++++++++
>  3 files changed, 19 insertions(+)
> 


Reviewed-by: David Ahern <dsahern@kernel.org>