Merge branch 'mlxsw-show-per-band-ecn-marked-counter-on-qdisc'

Ido Schimmel says:

====================
mlxsw: Show per-band ECN-marked counter on qdisc

The RED qdisc can expose number of packets that it has marked through
the prob_marked counter (shown in iproute2 as "marked"). This counter
currently just shows number of packets marked in the SW datapath, which
in a switch deployment likely means zero.

Spectrum-3 does support per-TC counters, and in this patchset, mlxsw
supports this RED statistic properly.
====================

Link: https://lore.kernel.org/r/20211013103748.492531-1-idosch@idosch.orgSigned-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/mellanox/mlxsw/reg.h

@@ -4951,7 +4951,7 @@ enum mlxsw_reg_ppcnt_grp {
 	MLXSW_REG_PPCNT_DISCARD_CNT = 0x6,
 	MLXSW_REG_PPCNT_PRIO_CNT = 0x10,
 	MLXSW_REG_PPCNT_TC_CNT = 0x11,
-	MLXSW_REG_PPCNT_TC_CONG_TC = 0x13,
+	MLXSW_REG_PPCNT_TC_CONG_CNT = 0x13,
 };
 
 /* reg_ppcnt_grp
@@ -5371,7 +5371,7 @@ MLXSW_ITEM64(reg, ppcnt, tx_pause_duration,
 MLXSW_ITEM64(reg, ppcnt, tx_pause_transition,
 	     MLXSW_REG_PPCNT_COUNTERS_OFFSET + 0x70, 0, 64);
 
-/* Ethernet Per Traffic Group Counters */
+/* Ethernet Per Traffic Class Counters */
 
 /* reg_ppcnt_tc_transmit_queue
  * Contains the transmit queue depth in cells of traffic class
@@ -5398,6 +5398,12 @@ MLXSW_ITEM64(reg, ppcnt, tc_no_buffer_discard_uc,
 MLXSW_ITEM64(reg, ppcnt, wred_discard,
 	     MLXSW_REG_PPCNT_COUNTERS_OFFSET + 0x00, 0, 64);
 
+/* reg_ppcnt_ecn_marked_tc
+ * Access: RO
+ */
+MLXSW_ITEM64(reg, ppcnt, ecn_marked_tc,
+	     MLXSW_REG_PPCNT_COUNTERS_OFFSET + 0x08, 0, 64);
+
 static inline void mlxsw_reg_ppcnt_pack(char *payload, u8 local_port,
 					enum mlxsw_reg_ppcnt_grp grp,
 					u8 prio_tc)

drivers/net/ethernet/mellanox/mlxsw/spectrum.c

@@ -824,12 +824,16 @@ mlxsw_sp_port_get_hw_xstats(struct net_device *dev,
 
 	for (i = 0; i < TC_MAX_QUEUE; i++) {
 		err = mlxsw_sp_port_get_stats_raw(dev,
-						  MLXSW_REG_PPCNT_TC_CONG_TC,
+						  MLXSW_REG_PPCNT_TC_CONG_CNT,
 						  i, ppcnt_pl);
-		if (!err)
-			xstats->wred_drop[i] =
-				mlxsw_reg_ppcnt_wred_discard_get(ppcnt_pl);
+		if (err)
+			goto tc_cnt;
+
+		xstats->wred_drop[i] =
+			mlxsw_reg_ppcnt_wred_discard_get(ppcnt_pl);
+		xstats->tc_ecn[i] = mlxsw_reg_ppcnt_ecn_marked_tc_get(ppcnt_pl);
 
+tc_cnt:
 		err = mlxsw_sp_port_get_stats_raw(dev, MLXSW_REG_PPCNT_TC_CNT,
 						  i, ppcnt_pl);
 		if (err)

drivers/net/ethernet/mellanox/mlxsw/spectrum.h

@@ -285,6 +285,7 @@ struct mlxsw_sp_port_vlan {
 /* No need an internal lock; At worse - miss a single periodic iteration */
 struct mlxsw_sp_port_xstats {
 	u64 ecn;
+	u64 tc_ecn[TC_MAX_QUEUE];
 	u64 wred_drop[TC_MAX_QUEUE];
 	u64 tail_drop[TC_MAX_QUEUE];
 	u64 backlog[TC_MAX_QUEUE];

drivers/net/ethernet/mellanox/mlxsw/spectrum_qdisc.c

@@ -519,6 +519,7 @@ mlxsw_sp_setup_tc_qdisc_red_clean_stats(struct mlxsw_sp_port *mlxsw_sp_port,
 					       mlxsw_sp_qdisc->prio_bitmap,
 					       &stats_base->tx_packets,
 					       &stats_base->tx_bytes);
+	red_base->prob_mark = xstats->tc_ecn[tclass_num];
 	red_base->prob_drop = xstats->wred_drop[tclass_num];
 	red_base->pdrop = mlxsw_sp_xstats_tail_drop(xstats, tclass_num);
 
@@ -618,19 +619,22 @@ mlxsw_sp_qdisc_get_red_xstats(struct mlxsw_sp_port *mlxsw_sp_port,
 	int tclass_num = mlxsw_sp_qdisc->tclass_num;
 	struct mlxsw_sp_port_xstats *xstats;
 	struct red_stats *res = xstats_ptr;
-	int early_drops, pdrops;
+	int early_drops, marks, pdrops;
 
 	xstats = &mlxsw_sp_port->periodic_hw_stats.xstats;
 
 	early_drops = xstats->wred_drop[tclass_num] - xstats_base->prob_drop;
+	marks = xstats->tc_ecn[tclass_num] - xstats_base->prob_mark;
 	pdrops = mlxsw_sp_xstats_tail_drop(xstats, tclass_num) -
 		 xstats_base->pdrop;
 
 	res->pdrop += pdrops;
 	res->prob_drop += early_drops;
+	res->prob_mark += marks;
 
 	xstats_base->pdrop += pdrops;
 	xstats_base->prob_drop += early_drops;
+	xstats_base->prob_mark += marks;
 	return 0;
 }
 
@@ -648,7 +652,8 @@ mlxsw_sp_qdisc_get_red_stats(struct mlxsw_sp_port *mlxsw_sp_port,
 	stats_base = &mlxsw_sp_qdisc->stats_base;
 
 	mlxsw_sp_qdisc_get_tc_stats(mlxsw_sp_port, mlxsw_sp_qdisc, stats_ptr);
-	overlimits = xstats->wred_drop[tclass_num] - stats_base->overlimits;
+	overlimits = xstats->wred_drop[tclass_num] +
+		     xstats->tc_ecn[tclass_num] - stats_base->overlimits;
 
 	stats_ptr->qstats->overlimits += overlimits;
 	stats_base->overlimits += overlimits;

tools/testing/selftests/drivers/net/mlxsw/sch_red_core.sh

@@ -331,6 +331,14 @@ get_nmarked()
 	ethtool_stats_get $swp3 ecn_marked
 }
 
+get_qdisc_nmarked()
+{
+	local vlan=$1; shift
+
+	busywait_for_counter 1100 +1 \
+		qdisc_stats_get $swp3 $(get_qdisc_handle $vlan) .marked
+}
+
 get_qdisc_npackets()
 {
 	local vlan=$1; shift
@@ -384,14 +392,15 @@ build_backlog()
 
 check_marking()
 {
+	local get_nmarked=$1; shift
 	local vlan=$1; shift
 	local cond=$1; shift
 
 	local npackets_0=$(get_qdisc_npackets $vlan)
-	local nmarked_0=$(get_nmarked $vlan)
+	local nmarked_0=$($get_nmarked $vlan)
 	sleep 5
 	local npackets_1=$(get_qdisc_npackets $vlan)
-	local nmarked_1=$(get_nmarked $vlan)
+	local nmarked_1=$($get_nmarked $vlan)
 
 	local nmarked_d=$((nmarked_1 - nmarked_0))
 	local npackets_d=$((npackets_1 - npackets_0))
@@ -404,6 +413,7 @@ check_marking()
 ecn_test_common()
 {
 	local name=$1; shift
+	local get_nmarked=$1; shift
 	local vlan=$1; shift
 	local limit=$1; shift
 	local backlog
@@ -416,7 +426,7 @@ ecn_test_common()
 	RET=0
 	backlog=$(build_backlog $vlan $((2 * limit / 3)) udp)
 	check_err $? "Could not build the requested backlog"
-	pct=$(check_marking $vlan "== 0")
+	pct=$(check_marking "$get_nmarked" $vlan "== 0")
 	check_err $? "backlog $backlog / $limit Got $pct% marked packets, expected == 0."
 	log_test "TC $((vlan - 10)): $name backlog < limit"
 
@@ -426,22 +436,23 @@ ecn_test_common()
 	RET=0
 	backlog=$(build_backlog $vlan $((3 * limit / 2)) tcp tos=0x01)
 	check_err $? "Could not build the requested backlog"
-	pct=$(check_marking $vlan ">= 95")
+	pct=$(check_marking "$get_nmarked" $vlan ">= 95")
 	check_err $? "backlog $backlog / $limit Got $pct% marked packets, expected >= 95."
 	log_test "TC $((vlan - 10)): $name backlog > limit"
 }
 
-do_ecn_test()
+__do_ecn_test()
 {
+	local get_nmarked=$1; shift
 	local vlan=$1; shift
 	local limit=$1; shift
-	local name=ECN
+	local name=${1-ECN}; shift
 
 	start_tcp_traffic $h1.$vlan $(ipaddr 1 $vlan) $(ipaddr 3 $vlan) \
 			  $h3_mac tos=0x01
 	sleep 1
 
-	ecn_test_common "$name" $vlan $limit
+	ecn_test_common "$name" "$get_nmarked" $vlan $limit
 
 	# Up there we saw that UDP gets accepted when backlog is below the
 	# limit. Now that it is above, it should all get dropped, and backlog
@@ -455,6 +466,26 @@ do_ecn_test()
 	sleep 1
 }
 
+do_ecn_test()
+{
+	local vlan=$1; shift
+	local limit=$1; shift
+
+	__do_ecn_test get_nmarked "$vlan" "$limit"
+}
+
+do_ecn_test_perband()
+{
+	local vlan=$1; shift
+	local limit=$1; shift
+
+	# Per-band ECN counters are not supported on Spectrum-1 and Spectrum-2.
+	[[ "$DEVLINK_VIDDID" == "15b3:cb84" ||
+	   "$DEVLINK_VIDDID" == "15b3:cf6c" ]] && return
+
+	__do_ecn_test get_qdisc_nmarked "$vlan" "$limit" "per-band ECN"
+}
+
 do_ecn_nodrop_test()
 {
 	local vlan=$1; shift
@@ -465,7 +496,7 @@ do_ecn_nodrop_test()
 			  $h3_mac tos=0x01
 	sleep 1
 
-	ecn_test_common "$name" $vlan $limit
+	ecn_test_common "$name" get_nmarked $vlan $limit
 
 	# Up there we saw that UDP gets accepted when backlog is below the
 	# limit. Now that it is above, in nodrop mode, make sure it goes to
@@ -495,7 +526,7 @@ do_red_test()
 	RET=0
 	backlog=$(build_backlog $vlan $((2 * limit / 3)) tcp tos=0x01)
 	check_err $? "Could not build the requested backlog"
-	pct=$(check_marking $vlan "== 0")
+	pct=$(check_marking get_nmarked $vlan "== 0")
 	check_err $? "backlog $backlog / $limit Got $pct% marked packets, expected == 0."
 	log_test "TC $((vlan - 10)): RED backlog < limit"
 
@@ -503,7 +534,7 @@ do_red_test()
 	RET=0
 	backlog=$(build_backlog $vlan $((3 * limit / 2)) tcp tos=0x01)
 	check_fail $? "Traffic went into backlog instead of being early-dropped"
-	pct=$(check_marking $vlan "== 0")
+	pct=$(check_marking get_nmarked $vlan "== 0")
 	check_err $? "backlog $backlog / $limit Got $pct% marked packets, expected == 0."
 	local diff=$((limit - backlog))
 	pct=$((100 * diff / limit))

tools/testing/selftests/drivers/net/mlxsw/sch_red_ets.sh

@@ -4,6 +4,7 @@
 ALL_TESTS="
 	ping_ipv4
 	ecn_test
+	ecn_test_perband
 	ecn_nodrop_test
 	red_test
 	mc_backlog_test
@@ -86,6 +87,16 @@ ecn_test()
 	uninstall_qdisc
 }
 
+ecn_test_perband()
+{
+	install_qdisc ecn
+
+	do_ecn_test_perband 10 $BACKLOG1
+	do_ecn_test_perband 11 $BACKLOG2
+
+	uninstall_qdisc
+}
+
 ecn_nodrop_test()
 {
 	install_qdisc ecn nodrop

tools/testing/selftests/drivers/net/mlxsw/sch_red_root.sh

@@ -4,6 +4,7 @@
 ALL_TESTS="
 	ping_ipv4
 	ecn_test
+	ecn_test_perband
 	ecn_nodrop_test
 	red_test
 	mc_backlog_test
@@ -35,6 +36,13 @@ ecn_test()
 	uninstall_qdisc
 }
 
+ecn_test_perband()
+{
+	install_qdisc ecn
+	do_ecn_test_perband 10 $BACKLOG
+	uninstall_qdisc
+}
+
 ecn_nodrop_test()
 {
 	install_qdisc ecn nodrop