aboutsummaryrefslogtreecommitdiffstatshomepage
diff options
context:
space:
mode:
authorJakub Kicinski <kuba@kernel.org>2025-04-29 14:44:39 -0700
committerJakub Kicinski <kuba@kernel.org>2025-04-29 14:44:40 -0700
commit1e0bff3bb59ca9bf84243df2a4ad3cfa6ca6f303 (patch)
treeea760162e09865b0285fb57fb2fd93a9d4e5738e
parentnet: ethernet: mtk_eth_soc: fix SER panic with 4GB+ RAM (diff)
parentselftests: net: tc_taprio: new test (diff)
downloadwireguard-linux-1e0bff3bb59ca9bf84243df2a4ad3cfa6ca6f303.tar.xz
wireguard-linux-1e0bff3bb59ca9bf84243df2a4ad3cfa6ca6f303.zip
Merge branch 'fix-felix-dsa-taprio-gates-after-clock-jump'
Vladimir Oltean says: ==================== Fix Felix DSA taprio gates after clock jump Richie Pearn presented a reproducible situation where traffic would get blocked on the NXP LS1028A switch if a certain taprio schedule was applied, and stepping the PTP clock would take place. The latter event is an expected initial occurrence, but also at runtime, for example when transitioning from one grandmaster to another. The issue is completely described in patch 1/4, which also contains the fix, but it has left me with some doubts regarding the need for vsc9959_tas_clock_adjust() in general. In order to prove to myself that vsc9959_tas_clock_adjust() is needed in general, I have written a selftest for the tc-taprio data path in patch 4/4. On the LS1028A, we can clearly see the following failures without that function: INFO: Forcing a backward clock jump TEST: ping [FAIL] INFO: Setting up taprio after PTP TEST: In band with gate [FAIL] Reception of 100 packets failed TEST: Out of band with gate [FAIL] Reception of 100 packets failed As for testing my fix from patch 1/4, that was quite a bit more complex to do automatically. In fact, I couldn't find any other schedule that would fail to be updated by vsc9959_tas_clock_adjust() as cleanly as the schedule from Richie, so I've added that specific schedule as the test_clock_jump_backward() test. The test ordering is also (unfortunately) very strategic. Running the selftest to the end dirties the GCL RAM, and when running test_clock_jump_backward() once again, the GCL entries won't be all zeroes as they were the first time around. They will contain bits and pieces of old schedules, making it very challenging to make it fail. Thus, test_clock_jump_backward() is the first in the test suite, and without patch 1/4, it is only supposed to fail the _first_ time when running after a clean boot. ==================== Link: https://patch.msgid.link/20250426144859.3128352-1-vladimir.oltean@nxp.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
-rw-r--r--drivers/net/dsa/ocelot/felix_vsc9959.c5
l---------tools/testing/selftests/drivers/net/dsa/tc_taprio.sh1
-rwxr-xr-xtools/testing/selftests/drivers/net/ocelot/psfp.sh8
-rwxr-xr-xtools/testing/selftests/net/forwarding/tc_taprio.sh421
-rw-r--r--tools/testing/selftests/net/forwarding/tsn_lib.sh26
5 files changed, 454 insertions, 7 deletions
diff --git a/drivers/net/dsa/ocelot/felix_vsc9959.c b/drivers/net/dsa/ocelot/felix_vsc9959.c
index 940f1b71226d..7b35d24c38d7 100644
--- a/drivers/net/dsa/ocelot/felix_vsc9959.c
+++ b/drivers/net/dsa/ocelot/felix_vsc9959.c
@@ -1543,7 +1543,7 @@ static void vsc9959_tas_clock_adjust(struct ocelot *ocelot)
struct tc_taprio_qopt_offload *taprio;
struct ocelot_port *ocelot_port;
struct timespec64 base_ts;
- int port;
+ int i, port;
u32 val;
mutex_lock(&ocelot->fwd_domain_lock);
@@ -1575,6 +1575,9 @@ static void vsc9959_tas_clock_adjust(struct ocelot *ocelot)
QSYS_PARAM_CFG_REG_3_BASE_TIME_SEC_MSB_M,
QSYS_PARAM_CFG_REG_3);
+ for (i = 0; i < taprio->num_entries; i++)
+ vsc9959_tas_gcl_set(ocelot, i, &taprio->entries[i]);
+
ocelot_rmw(ocelot, QSYS_TAS_PARAM_CFG_CTRL_CONFIG_CHANGE,
QSYS_TAS_PARAM_CFG_CTRL_CONFIG_CHANGE,
QSYS_TAS_PARAM_CFG_CTRL);
diff --git a/tools/testing/selftests/drivers/net/dsa/tc_taprio.sh b/tools/testing/selftests/drivers/net/dsa/tc_taprio.sh
new file mode 120000
index 000000000000..d16a65e7595d
--- /dev/null
+++ b/tools/testing/selftests/drivers/net/dsa/tc_taprio.sh
@@ -0,0 +1 @@
+run_net_forwarding_test.sh \ No newline at end of file
diff --git a/tools/testing/selftests/drivers/net/ocelot/psfp.sh b/tools/testing/selftests/drivers/net/ocelot/psfp.sh
index bed748dde4b0..8972f42dfe03 100755
--- a/tools/testing/selftests/drivers/net/ocelot/psfp.sh
+++ b/tools/testing/selftests/drivers/net/ocelot/psfp.sh
@@ -266,18 +266,14 @@ run_test()
"${base_time}" \
"${CYCLE_TIME_NS}" \
"${SHIFT_TIME_NS}" \
+ "${GATE_DURATION_NS}" \
"${NUM_PKTS}" \
"${STREAM_VID}" \
"${STREAM_PRIO}" \
"" \
"${isochron_dat}"
- # Count all received packets by looking at the non-zero RX timestamps
- received=$(isochron report \
- --input-file "${isochron_dat}" \
- --printf-format "%u\n" --printf-args "R" | \
- grep -w -v '0' | wc -l)
-
+ received=$(isochron_report_num_received "${isochron_dat}")
if [ "${received}" = "${expected}" ]; then
RET=0
else
diff --git a/tools/testing/selftests/net/forwarding/tc_taprio.sh b/tools/testing/selftests/net/forwarding/tc_taprio.sh
new file mode 100755
index 000000000000..8992aeabfe0b
--- /dev/null
+++ b/tools/testing/selftests/net/forwarding/tc_taprio.sh
@@ -0,0 +1,421 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+
+ALL_TESTS=" \
+ test_clock_jump_backward \
+ test_taprio_after_ptp \
+ test_max_sdu \
+ test_clock_jump_backward_forward \
+"
+NUM_NETIFS=4
+source tc_common.sh
+source lib.sh
+source tsn_lib.sh
+
+require_command python3
+
+# The test assumes the usual topology from the README, where h1 is connected to
+# swp1, h2 to swp2, and swp1 and swp2 are together in a bridge.
+# Additional assumption: h1 and h2 use the same PHC, and so do swp1 and swp2.
+# By synchronizing h1 to swp1 via PTP, h2 is also implicitly synchronized to
+# swp1 (and both to CLOCK_REALTIME).
+h1=${NETIFS[p1]}
+swp1=${NETIFS[p2]}
+swp2=${NETIFS[p3]}
+h2=${NETIFS[p4]}
+
+UDS_ADDRESS_H1="/var/run/ptp4l_h1"
+UDS_ADDRESS_SWP1="/var/run/ptp4l_swp1"
+
+H1_IPV4="192.0.2.1"
+H2_IPV4="192.0.2.2"
+H1_IPV6="2001:db8:1::1"
+H2_IPV6="2001:db8:1::2"
+
+# Tunables
+NUM_PKTS=100
+STREAM_VID=10
+STREAM_PRIO_1=6
+STREAM_PRIO_2=5
+STREAM_PRIO_3=4
+# PTP uses TC 0
+ALL_GATES=$((1 << 0 | 1 << STREAM_PRIO_1 | 1 << STREAM_PRIO_2))
+# Use a conservative cycle of 10 ms to allow the test to still pass when the
+# kernel has some extra overhead like lockdep etc
+CYCLE_TIME_NS=10000000
+# Create two Gate Control List entries, one OPEN and one CLOSE, of equal
+# durations
+GATE_DURATION_NS=$((CYCLE_TIME_NS / 2))
+# Give 2/3 of the cycle time to user space and 1/3 to the kernel
+FUDGE_FACTOR=$((CYCLE_TIME_NS / 3))
+# Shift the isochron base time by half the gate time, so that packets are
+# always received by swp1 close to the middle of the time slot, to minimize
+# inaccuracies due to network sync
+SHIFT_TIME_NS=$((GATE_DURATION_NS / 2))
+
+path_delay=
+
+h1_create()
+{
+ simple_if_init $h1 $H1_IPV4/24 $H1_IPV6/64
+}
+
+h1_destroy()
+{
+ simple_if_fini $h1 $H1_IPV4/24 $H1_IPV6/64
+}
+
+h2_create()
+{
+ simple_if_init $h2 $H2_IPV4/24 $H2_IPV6/64
+}
+
+h2_destroy()
+{
+ simple_if_fini $h2 $H2_IPV4/24 $H2_IPV6/64
+}
+
+switch_create()
+{
+ local h2_mac_addr=$(mac_get $h2)
+
+ ip link set $swp1 up
+ ip link set $swp2 up
+
+ ip link add br0 type bridge vlan_filtering 1
+ ip link set $swp1 master br0
+ ip link set $swp2 master br0
+ ip link set br0 up
+
+ bridge vlan add dev $swp2 vid $STREAM_VID
+ bridge vlan add dev $swp1 vid $STREAM_VID
+ bridge fdb add dev $swp2 \
+ $h2_mac_addr vlan $STREAM_VID static master
+}
+
+switch_destroy()
+{
+ ip link del br0
+}
+
+ptp_setup()
+{
+ # Set up swp1 as a master PHC for h1, synchronized to the local
+ # CLOCK_REALTIME.
+ phc2sys_start $UDS_ADDRESS_SWP1
+ ptp4l_start $h1 true $UDS_ADDRESS_H1
+ ptp4l_start $swp1 false $UDS_ADDRESS_SWP1
+}
+
+ptp_cleanup()
+{
+ ptp4l_stop $swp1
+ ptp4l_stop $h1
+ phc2sys_stop
+}
+
+txtime_setup()
+{
+ local if_name=$1
+
+ tc qdisc add dev $if_name clsact
+ # Classify PTP on TC 7 and isochron on TC 6
+ tc filter add dev $if_name egress protocol 0x88f7 \
+ flower action skbedit priority 7
+ tc filter add dev $if_name egress protocol 802.1Q \
+ flower vlan_ethtype 0xdead action skbedit priority 6
+ tc qdisc add dev $if_name handle 100: parent root mqprio num_tc 8 \
+ queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
+ map 0 1 2 3 4 5 6 7 \
+ hw 1
+ # Set up TC 5, 6, 7 for SO_TXTIME. tc-mqprio queues count from 1.
+ tc qdisc replace dev $if_name parent 100:$((STREAM_PRIO_1 + 1)) etf \
+ clockid CLOCK_TAI offload delta $FUDGE_FACTOR
+ tc qdisc replace dev $if_name parent 100:$((STREAM_PRIO_2 + 1)) etf \
+ clockid CLOCK_TAI offload delta $FUDGE_FACTOR
+ tc qdisc replace dev $if_name parent 100:$((STREAM_PRIO_3 + 1)) etf \
+ clockid CLOCK_TAI offload delta $FUDGE_FACTOR
+}
+
+txtime_cleanup()
+{
+ local if_name=$1
+
+ tc qdisc del dev $if_name clsact
+ tc qdisc del dev $if_name root
+}
+
+taprio_replace()
+{
+ local if_name="$1"; shift
+ local extra_args="$1"; shift
+
+ # STREAM_PRIO_1 always has an open gate.
+ # STREAM_PRIO_2 has a gate open for GATE_DURATION_NS (half the cycle time)
+ # STREAM_PRIO_3 always has a closed gate.
+ tc qdisc replace dev $if_name root stab overhead 24 taprio num_tc 8 \
+ queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
+ map 0 1 2 3 4 5 6 7 \
+ sched-entry S $(printf "%x" $ALL_GATES) $GATE_DURATION_NS \
+ sched-entry S $(printf "%x" $((ALL_GATES & ~(1 << STREAM_PRIO_2)))) $GATE_DURATION_NS \
+ base-time 0 flags 0x2 $extra_args
+ taprio_wait_for_admin $if_name
+}
+
+taprio_cleanup()
+{
+ local if_name=$1
+
+ tc qdisc del dev $if_name root
+}
+
+probe_path_delay()
+{
+ local isochron_dat="$(mktemp)"
+ local received
+
+ log_info "Probing path delay"
+
+ isochron_do "$h1" "$h2" "$UDS_ADDRESS_H1" "" 0 \
+ "$CYCLE_TIME_NS" "" "" "$NUM_PKTS" \
+ "$STREAM_VID" "$STREAM_PRIO_1" "" "$isochron_dat"
+
+ received=$(isochron_report_num_received "$isochron_dat")
+ if [ "$received" != "$NUM_PKTS" ]; then
+ echo "Cannot establish basic data path between $h1 and $h2"
+ exit $ksft_fail
+ fi
+
+ printf "pdelay = {}\n" > isochron_data.py
+ isochron report --input-file "$isochron_dat" \
+ --printf-format "pdelay[%u] = %d - %d\n" \
+ --printf-args "qRT" \
+ >> isochron_data.py
+ cat <<-'EOF' > isochron_postprocess.py
+ #!/usr/bin/env python3
+
+ from isochron_data import pdelay
+ import numpy as np
+
+ w = np.array(list(pdelay.values()))
+ print("{}".format(np.max(w)))
+ EOF
+ path_delay=$(python3 ./isochron_postprocess.py)
+
+ log_info "Path delay from $h1 to $h2 estimated at $path_delay ns"
+
+ if [ "$path_delay" -gt "$GATE_DURATION_NS" ]; then
+ echo "Path delay larger than gate duration, aborting"
+ exit $ksft_fail
+ fi
+
+ rm -f ./isochron_data.py 2> /dev/null
+ rm -f ./isochron_postprocess.py 2> /dev/null
+ rm -f "$isochron_dat" 2> /dev/null
+}
+
+setup_prepare()
+{
+ vrf_prepare
+
+ h1_create
+ h2_create
+ switch_create
+
+ txtime_setup $h1
+
+ # Temporarily set up PTP just to probe the end-to-end path delay.
+ ptp_setup
+ probe_path_delay
+ ptp_cleanup
+}
+
+cleanup()
+{
+ pre_cleanup
+
+ isochron_recv_stop
+ txtime_cleanup $h1
+
+ switch_destroy
+ h2_destroy
+ h1_destroy
+
+ vrf_cleanup
+}
+
+run_test()
+{
+ local base_time=$1; shift
+ local stream_prio=$1; shift
+ local expected_delay=$1; shift
+ local should_fail=$1; shift
+ local test_name=$1; shift
+ local isochron_dat="$(mktemp)"
+ local received
+ local median_delay
+
+ RET=0
+
+ # Set the shift time equal to the cycle time, which effectively
+ # cancels the default advance time. Packets won't be sent early in
+ # software, which ensures that they won't prematurely enter through
+ # the open gate in __test_out_of_band(). Also, the gate is open for
+ # long enough that this won't cause a problem in __test_in_band().
+ isochron_do "$h1" "$h2" "$UDS_ADDRESS_H1" "" "$base_time" \
+ "$CYCLE_TIME_NS" "$SHIFT_TIME_NS" "$GATE_DURATION_NS" \
+ "$NUM_PKTS" "$STREAM_VID" "$stream_prio" "" "$isochron_dat"
+
+ received=$(isochron_report_num_received "$isochron_dat")
+ [ "$received" = "$NUM_PKTS" ]
+ check_err_fail $should_fail $? "Reception of $NUM_PKTS packets"
+
+ if [ $should_fail = 0 ] && [ "$received" = "$NUM_PKTS" ]; then
+ printf "pdelay = {}\n" > isochron_data.py
+ isochron report --input-file "$isochron_dat" \
+ --printf-format "pdelay[%u] = %d - %d\n" \
+ --printf-args "qRT" \
+ >> isochron_data.py
+ cat <<-'EOF' > isochron_postprocess.py
+ #!/usr/bin/env python3
+
+ from isochron_data import pdelay
+ import numpy as np
+
+ w = np.array(list(pdelay.values()))
+ print("{}".format(int(np.median(w))))
+ EOF
+ median_delay=$(python3 ./isochron_postprocess.py)
+
+ # If the condition below is true, packets were delayed by a closed gate
+ [ "$median_delay" -gt $((path_delay + expected_delay)) ]
+ check_fail $? "Median delay $median_delay is greater than expected delay $expected_delay plus path delay $path_delay"
+
+ # If the condition below is true, packets were sent expecting them to
+ # hit a closed gate in the switch, but were not delayed
+ [ "$expected_delay" -gt 0 ] && [ "$median_delay" -lt "$expected_delay" ]
+ check_fail $? "Median delay $median_delay is less than expected delay $expected_delay"
+ fi
+
+ log_test "$test_name"
+
+ rm -f ./isochron_data.py 2> /dev/null
+ rm -f ./isochron_postprocess.py 2> /dev/null
+ rm -f "$isochron_dat" 2> /dev/null
+}
+
+__test_always_open()
+{
+ run_test 0.000000000 $STREAM_PRIO_1 0 0 "Gate always open"
+}
+
+__test_always_closed()
+{
+ run_test 0.000000000 $STREAM_PRIO_3 0 1 "Gate always closed"
+}
+
+__test_in_band()
+{
+ # Send packets in-band with the OPEN gate entry
+ run_test 0.000000000 $STREAM_PRIO_2 0 0 "In band with gate"
+}
+
+__test_out_of_band()
+{
+ # Send packets in-band with the CLOSE gate entry
+ run_test 0.005000000 $STREAM_PRIO_2 \
+ $((GATE_DURATION_NS - SHIFT_TIME_NS)) 0 \
+ "Out of band with gate"
+}
+
+run_subtests()
+{
+ __test_always_open
+ __test_always_closed
+ __test_in_band
+ __test_out_of_band
+}
+
+test_taprio_after_ptp()
+{
+ log_info "Setting up taprio after PTP"
+ ptp_setup
+ taprio_replace $swp2
+ run_subtests
+ taprio_cleanup $swp2
+ ptp_cleanup
+}
+
+__test_under_max_sdu()
+{
+ # Limit max-sdu for STREAM_PRIO_1
+ taprio_replace "$swp2" "max-sdu 0 0 0 0 0 0 100 0"
+ run_test 0.000000000 $STREAM_PRIO_1 0 0 "Under maximum SDU"
+}
+
+__test_over_max_sdu()
+{
+ # Limit max-sdu for STREAM_PRIO_1
+ taprio_replace "$swp2" "max-sdu 0 0 0 0 0 0 20 0"
+ run_test 0.000000000 $STREAM_PRIO_1 0 1 "Over maximum SDU"
+}
+
+test_max_sdu()
+{
+ ptp_setup
+ __test_under_max_sdu
+ __test_over_max_sdu
+ taprio_cleanup $swp2
+ ptp_cleanup
+}
+
+# Perform a clock jump in the past without synchronization running, so that the
+# time base remains where it was set by phc_ctl.
+test_clock_jump_backward()
+{
+ # This is a more complex schedule specifically crafted in a way that
+ # has been problematic on NXP LS1028A. Not much to test with it other
+ # than the fact that it passes traffic.
+ tc qdisc replace dev $swp2 root stab overhead 24 taprio num_tc 8 \
+ queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 map 0 1 2 3 4 5 6 7 \
+ base-time 0 sched-entry S 20 300000 sched-entry S 10 200000 \
+ sched-entry S 20 300000 sched-entry S 48 200000 \
+ sched-entry S 20 300000 sched-entry S 83 200000 \
+ sched-entry S 40 300000 sched-entry S 00 200000 flags 2
+
+ log_info "Forcing a backward clock jump"
+ phc_ctl $swp1 set 0
+
+ ping_test $h1 192.0.2.2
+ taprio_cleanup $swp2
+}
+
+# Test that taprio tolerates clock jumps.
+# Since ptp4l and phc2sys are running, it is expected for the time to
+# eventually recover (through yet another clock jump). Isochron waits
+# until that is the case.
+test_clock_jump_backward_forward()
+{
+ log_info "Forcing a backward and a forward clock jump"
+ taprio_replace $swp2
+ phc_ctl $swp1 set 0
+ ptp_setup
+ ping_test $h1 192.0.2.2
+ run_subtests
+ ptp_cleanup
+ taprio_cleanup $swp2
+}
+
+tc_offload_check
+if [[ $? -ne 0 ]]; then
+ log_test_skip "Could not test offloaded functionality"
+ exit $EXIT_STATUS
+fi
+
+trap cleanup EXIT
+
+setup_prepare
+setup_wait
+tests_run
+
+exit $EXIT_STATUS
diff --git a/tools/testing/selftests/net/forwarding/tsn_lib.sh b/tools/testing/selftests/net/forwarding/tsn_lib.sh
index b91bcd8008a9..08c044ff6689 100644
--- a/tools/testing/selftests/net/forwarding/tsn_lib.sh
+++ b/tools/testing/selftests/net/forwarding/tsn_lib.sh
@@ -2,6 +2,8 @@
# SPDX-License-Identifier: GPL-2.0
# Copyright 2021-2022 NXP
+tc_testing_scripts_dir=$(dirname $0)/../../tc-testing/scripts
+
REQUIRE_ISOCHRON=${REQUIRE_ISOCHRON:=yes}
REQUIRE_LINUXPTP=${REQUIRE_LINUXPTP:=yes}
@@ -18,6 +20,7 @@ fi
if [[ "$REQUIRE_LINUXPTP" = "yes" ]]; then
require_command phc2sys
require_command ptp4l
+ require_command phc_ctl
fi
phc2sys_start()
@@ -182,6 +185,7 @@ isochron_do()
local base_time=$1; shift
local cycle_time=$1; shift
local shift_time=$1; shift
+ local window_size=$1; shift
local num_pkts=$1; shift
local vid=$1; shift
local priority=$1; shift
@@ -212,6 +216,10 @@ isochron_do()
extra_args="${extra_args} --shift-time=${shift_time}"
fi
+ if ! [ -z "${window_size}" ]; then
+ extra_args="${extra_args} --window-size=${window_size}"
+ fi
+
if [ "${use_l2}" = "true" ]; then
extra_args="${extra_args} --l2 --etype=0xdead ${vid}"
receiver_extra_args="--l2 --etype=0xdead"
@@ -247,3 +255,21 @@ isochron_do()
cpufreq_restore ${ISOCHRON_CPU}
}
+
+isochron_report_num_received()
+{
+ local isochron_dat=$1; shift
+
+ # Count all received packets by looking at the non-zero RX timestamps
+ isochron report \
+ --input-file "${isochron_dat}" \
+ --printf-format "%u\n" --printf-args "R" | \
+ grep -w -v '0' | wc -l
+}
+
+taprio_wait_for_admin()
+{
+ local if_name="$1"; shift
+
+ "$tc_testing_scripts_dir/taprio_wait_for_admin.sh" "$(which tc)" "$if_name"
+}