Merge branch 'mlxsw-Further-MC-awareness-configuration'

Ido Schimmel says:

====================
mlxsw: Further MC-awareness configuration

Petr says:

Due to an issue in Spectrum chips, when unicast traffic shares the same
queue as BUM traffic, and there is congestion, the BUM traffic is
admitted to the queue anyway, thus pushing out all UC traffic. In order
to give unicast traffic precedence over BUM traffic, multicast-aware
mode is now configured on all ports. Under MC-aware mode, egress TCs
8..15 are used for BUM traffic, which has its own dedicated pool.

This patch set improves the way that the MC pool and the higher-order
TCs are integrated into the system.

In patch #1, shaper at the higher TCs is configured to the same value
that it has by default. It's better to have the corresponding artifact
in the code explicitly.

The 8 following patches gradually extend the devlink handling in mlxsw
to support the extra TCs and the new MC pool.

Patch #2 changes the way that pools are indexed in mlxsw. Instead of
using (FW index, direction) tuple to identify the pool and the
associated cache, mlxsw now uses devlink index. This change is necessary
because the new pool 15 is not contiguously adjacent to the
currently-used pools 0..3, and because it's only relevant on egress.
Using devlink index relaxes the requirement for symmetry and adjacency
imposed by using FW indexing.

In patch #3, the assumption that number of ingress TCs matches that of
egress TCs is relaxed to allow exposition of egress TCs 8..15.

In patches #4, #5 and #6, support for infinite quotas is introduced.
Infinite quotas are reported as taking all the memory in the system, but
actually use a mechanism where the infinity is configured explicitly.

In patches #7 and #8, support for configuring static pool sizes in
introduced. Statically-sized pools have been supported for a while now,
but during initialization, all pools have dynamic size. The patches
allow there to be a mix of by-default static and dynamic pools.

In patches #9 and #10, pool 15 resp. per-priority MC quotas are
explicitly configured to be in sync with the current recommendation for
handling BUM traffic in Spectrum chips.

In the following 3 patches, an mlxsw-specific selftest is added to test
the MC-awareness configuration.

First in patches #11 and #12, lib.sh is extended with functions to
collect ethtool stats, and to manage port MTU.

Then in patch #13 the selftest itself is added.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

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

@@ -8336,8 +8336,15 @@ MLXSW_ITEM32(reg, sbpr, dir, 0x00, 24, 2);
  */
 MLXSW_ITEM32(reg, sbpr, pool, 0x00, 0, 4);
 
+/* reg_sbpr_infi_size
+ * Size is infinite.
+ * Access: RW
+ */
+MLXSW_ITEM32(reg, sbpr, infi_size, 0x04, 31, 1);
+
 /* reg_sbpr_size
  * Pool size in buffer cells.
+ * Reserved when infi_size = 1.
  * Access: RW
  */
 MLXSW_ITEM32(reg, sbpr, size, 0x04, 0, 24);
@@ -8355,13 +8362,15 @@ MLXSW_ITEM32(reg, sbpr, mode, 0x08, 0, 4);
 
 static inline void mlxsw_reg_sbpr_pack(char *payload, u8 pool,
 				       enum mlxsw_reg_sbxx_dir dir,
-				       enum mlxsw_reg_sbpr_mode mode, u32 size)
+				       enum mlxsw_reg_sbpr_mode mode, u32 size,
+				       bool infi_size)
 {
 	MLXSW_REG_ZERO(sbpr, payload);
 	mlxsw_reg_sbpr_pool_set(payload, pool);
 	mlxsw_reg_sbpr_dir_set(payload, dir);
 	mlxsw_reg_sbpr_mode_set(payload, mode);
 	mlxsw_reg_sbpr_size_set(payload, size);
+	mlxsw_reg_sbpr_infi_size_set(payload, infi_size);
 }
 
 /* SBCM - Shared Buffer Class Management Register
@@ -8409,6 +8418,12 @@ MLXSW_ITEM32(reg, sbcm, min_buff, 0x18, 0, 24);
 #define MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN 1
 #define MLXSW_REG_SBXX_DYN_MAX_BUFF_MAX 14
 
+/* reg_sbcm_infi_max
+ * Max buffer is infinite.
+ * Access: RW
+ */
+MLXSW_ITEM32(reg, sbcm, infi_max, 0x1C, 31, 1);
+
 /* reg_sbcm_max_buff
  * When the pool associated to the port-pg/tclass is configured to
  * static, Maximum buffer size for the limiter configured in cells.
@@ -8418,6 +8433,7 @@ MLXSW_ITEM32(reg, sbcm, min_buff, 0x18, 0, 24);
  * 0: 0
  * i: (1/128)*2^(i-1), for i=1..14
  * 0xFF: Infinity
+ * Reserved when infi_max = 1.
  * Access: RW
  */
 MLXSW_ITEM32(reg, sbcm, max_buff, 0x1C, 0, 24);
@@ -8430,7 +8446,8 @@ MLXSW_ITEM32(reg, sbcm, pool, 0x24, 0, 4);
 
 static inline void mlxsw_reg_sbcm_pack(char *payload, u8 local_port, u8 pg_buff,
 				       enum mlxsw_reg_sbxx_dir dir,
-				       u32 min_buff, u32 max_buff, u8 pool)
+				       u32 min_buff, u32 max_buff,
+				       bool infi_max, u8 pool)
 {
 	MLXSW_REG_ZERO(sbcm, payload);
 	mlxsw_reg_sbcm_local_port_set(payload, local_port);
@@ -8438,6 +8455,7 @@ static inline void mlxsw_reg_sbcm_pack(char *payload, u8 local_port, u8 pg_buff,
 	mlxsw_reg_sbcm_dir_set(payload, dir);
 	mlxsw_reg_sbcm_min_buff_set(payload, min_buff);
 	mlxsw_reg_sbcm_max_buff_set(payload, max_buff);
+	mlxsw_reg_sbcm_infi_max_set(payload, infi_max);
 	mlxsw_reg_sbcm_pool_set(payload, pool);
 }
 

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

@@ -2804,6 +2804,13 @@ static int mlxsw_sp_port_ets_init(struct mlxsw_sp_port *mlxsw_sp_port)
 						    MLXSW_REG_QEEC_MAS_DIS);
 		if (err)
 			return err;
+
+		err = mlxsw_sp_port_ets_maxrate_set(mlxsw_sp_port,
+						    MLXSW_REG_QEEC_HIERARCY_TC,
+						    i + 8, i,
+						    MLXSW_REG_QEEC_MAS_DIS);
+		if (err)
+			return err;
 	}
 
 	/* Map all priorities to traffic class 0. */

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

+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+#
+# A test for switch behavior under MC overload. An issue in Spectrum chips
+# causes throughput of UC traffic to drop severely when a switch is under heavy
+# MC load. This issue can be overcome by putting the switch to MC-aware mode.
+# This test verifies that UC performance stays intact even as the switch is
+# under MC flood, and therefore that the MC-aware mode is enabled and correctly
+# configured.
+#
+# Because mlxsw throttles CPU port, the traffic can't actually reach userspace
+# at full speed. That makes it impossible to use iperf3 to simply measure the
+# throughput, because many packets (that reach $h3) don't get to the kernel at
+# all even in UDP mode (the situation is even worse in TCP mode, where one can't
+# hope to see more than a couple Mbps).
+#
+# So instead we send traffic with mausezahn and use RX ethtool counters at $h3.
+# Multicast traffic is untagged, unicast traffic is tagged with PCP 1. Therefore
+# each gets a different priority and we can use per-prio ethtool counters to
+# measure the throughput. In order to avoid prioritizing unicast traffic, prio
+# qdisc is installed on $swp3 and maps all priorities to the same band #7 (and
+# thus TC 0).
+#
+# Mausezahn can't actually saturate the links unless it's using large frames.
+# Thus we set MTU to 10K on all involved interfaces. Then both unicast and
+# multicast traffic uses 8K frames.
+#
+# +-----------------------+                +----------------------------------+
+# | H1                    |                |                               H2 |
+# |                       |                |  unicast --> + $h2.111           |
+# |                       |                |  traffic     | 192.0.2.129/28    |
+# |          multicast    |                |              | e-qos-map 0:1     |
+# |          traffic      |                |              |                   |
+# | $h1 + <-----          |                |              + $h2               |
+# +-----|-----------------+                +--------------|-------------------+
+#       |                                                 |
+# +-----|-------------------------------------------------|-------------------+
+# |     + $swp1                                           + $swp2             |
+# |     | >1Gbps                                          | >1Gbps            |
+# | +---|----------------+                     +----------|----------------+  |
+# | |   + $swp1.1        |                     |          + $swp2.111      |  |
+# | |                BR1 |             SW      | BR111                     |  |
+# | |   + $swp3.1        |                     |          + $swp3.111      |  |
+# | +---|----------------+                     +----------|----------------+  |
+# |     \_________________________________________________/                   |
+# |                                    |                                      |
+# |                                    + $swp3                                |
+# |                                    | 1Gbps bottleneck                     |
+# |                                    | prio qdisc: {0..7} -> 7              |
+# +------------------------------------|--------------------------------------+
+#                                      |
+#                                   +--|-----------------+
+#                                   |  + $h3          H3 |
+#                                   |  |                 |
+#                                   |  + $h3.111         |
+#                                   |    192.0.2.130/28  |
+#                                   +--------------------+
+
+ALL_TESTS="
+	ping_ipv4
+	test_mc_aware
+"
+
+lib_dir=$(dirname $0)/../../../net/forwarding
+
+NUM_NETIFS=6
+source $lib_dir/lib.sh
+
+h1_create()
+{
+	simple_if_init $h1
+	mtu_set $h1 10000
+}
+
+h1_destroy()
+{
+	mtu_restore $h1
+	simple_if_fini $h1
+}
+
+h2_create()
+{
+	simple_if_init $h2
+	mtu_set $h2 10000
+
+	vlan_create $h2 111 v$h2 192.0.2.129/28
+	ip link set dev $h2.111 type vlan egress-qos-map 0:1
+}
+
+h2_destroy()
+{
+	vlan_destroy $h2 111
+
+	mtu_restore $h2
+	simple_if_fini $h2
+}
+
+h3_create()
+{
+	simple_if_init $h3
+	mtu_set $h3 10000
+
+	vlan_create $h3 111 v$h3 192.0.2.130/28
+}
+
+h3_destroy()
+{
+	vlan_destroy $h3 111
+
+	mtu_restore $h3
+	simple_if_fini $h3
+}
+
+switch_create()
+{
+	ip link set dev $swp1 up
+	mtu_set $swp1 10000
+
+	ip link set dev $swp2 up
+	mtu_set $swp2 10000
+
+	ip link set dev $swp3 up
+	mtu_set $swp3 10000
+
+	vlan_create $swp2 111
+	vlan_create $swp3 111
+
+	ethtool -s $swp3 speed 1000 autoneg off
+	tc qdisc replace dev $swp3 root handle 3: \
+	   prio bands 8 priomap 7 7 7 7 7 7 7 7
+
+	ip link add name br1 type bridge vlan_filtering 0
+	ip link set dev br1 up
+	ip link set dev $swp1 master br1
+	ip link set dev $swp3 master br1
+
+	ip link add name br111 type bridge vlan_filtering 0
+	ip link set dev br111 up
+	ip link set dev $swp2.111 master br111
+	ip link set dev $swp3.111 master br111
+}
+
+switch_destroy()
+{
+	ip link del dev br111
+	ip link del dev br1
+
+	tc qdisc del dev $swp3 root handle 3:
+	ethtool -s $swp3 autoneg on
+
+	vlan_destroy $swp3 111
+	vlan_destroy $swp2 111
+
+	mtu_restore $swp3
+	ip link set dev $swp3 down
+
+	mtu_restore $swp2
+	ip link set dev $swp2 down
+
+	mtu_restore $swp1
+	ip link set dev $swp1 down
+}
+
+setup_prepare()
+{
+	h1=${NETIFS[p1]}
+	swp1=${NETIFS[p2]}
+
+	swp2=${NETIFS[p3]}
+	h2=${NETIFS[p4]}
+
+	swp3=${NETIFS[p5]}
+	h3=${NETIFS[p6]}
+
+	h3mac=$(mac_get $h3)
+
+	vrf_prepare
+
+	h1_create
+	h2_create
+	h3_create
+	switch_create
+}
+
+cleanup()
+{
+	pre_cleanup
+
+	switch_destroy
+	h3_destroy
+	h2_destroy
+	h1_destroy
+
+	vrf_cleanup
+}
+
+ping_ipv4()
+{
+	ping_test $h2 192.0.2.130
+}
+
+humanize()
+{
+	local speed=$1; shift
+
+	for unit in bps Kbps Mbps Gbps; do
+		if (($(echo "$speed < 1024" | bc))); then
+			break
+		fi
+
+		speed=$(echo "scale=1; $speed / 1024" | bc)
+	done
+
+	echo "$speed${unit}"
+}
+
+rate()
+{
+	local t0=$1; shift
+	local t1=$1; shift
+	local interval=$1; shift
+
+	echo $((8 * (t1 - t0) / interval))
+}
+
+check_rate()
+{
+	local rate=$1; shift
+	local min=$1; shift
+	local what=$1; shift
+
+	if ((rate > min)); then
+		return 0
+	fi
+
+	echo "$what $(humanize $ir) < $(humanize $min_ingress)" > /dev/stderr
+	return 1
+}
+
+measure_uc_rate()
+{
+	local what=$1; shift
+
+	local interval=10
+	local i
+	local ret=0
+
+	# Dips in performance might cause momentary ingress rate to drop below
+	# 1Gbps. That wouldn't saturate egress and MC would thus get through,
+	# seemingly winning bandwidth on account of UC. Demand at least 2Gbps
+	# average ingress rate to somewhat mitigate this.
+	local min_ingress=2147483648
+
+	mausezahn $h2.111 -p 8000 -A 192.0.2.129 -B 192.0.2.130 -c 0 \
+		-a own -b $h3mac -t udp -q &
+	sleep 1
+
+	for i in {5..0}; do
+		local t0=$(ethtool_stats_get $h3 rx_octets_prio_1)
+		local u0=$(ethtool_stats_get $swp2 rx_octets_prio_1)
+		sleep $interval
+		local t1=$(ethtool_stats_get $h3 rx_octets_prio_1)
+		local u1=$(ethtool_stats_get $swp2 rx_octets_prio_1)
+
+		local ir=$(rate $u0 $u1 $interval)
+		local er=$(rate $t0 $t1 $interval)
+
+		if check_rate $ir $min_ingress "$what ingress rate"; then
+			break
+		fi
+
+		# Fail the test if we can't get the throughput.
+		if ((i == 0)); then
+			ret=1
+		fi
+	done
+
+	# Suppress noise from killing mausezahn.
+	{ kill %% && wait; } 2>/dev/null
+
+	echo $ir $er
+	exit $ret
+}
+
+test_mc_aware()
+{
+	RET=0
+
+	local -a uc_rate
+	uc_rate=($(measure_uc_rate "UC-only"))
+	check_err $? "Could not get high enough UC-only ingress rate"
+	local ucth1=${uc_rate[1]}
+
+	mausezahn $h1 -p 8000 -c 0 -a own -b bc -t udp -q &
+
+	local d0=$(date +%s)
+	local t0=$(ethtool_stats_get $h3 rx_octets_prio_0)
+	local u0=$(ethtool_stats_get $swp1 rx_octets_prio_0)
+
+	local -a uc_rate_2
+	uc_rate_2=($(measure_uc_rate "UC+MC"))
+	check_err $? "Could not get high enough UC+MC ingress rate"
+	local ucth2=${uc_rate_2[1]}
+
+	local d1=$(date +%s)
+	local t1=$(ethtool_stats_get $h3 rx_octets_prio_0)
+	local u1=$(ethtool_stats_get $swp1 rx_octets_prio_0)
+
+	local deg=$(bc <<< "
+			scale=2
+			ret = 100 * ($ucth1 - $ucth2) / $ucth1
+			if (ret > 0) { ret } else { 0 }
+		    ")
+	check_err $(bc <<< "$deg > 10")
+
+	local interval=$((d1 - d0))
+	local mc_ir=$(rate $u0 $u1 $interval)
+	local mc_er=$(rate $t0 $t1 $interval)
+
+	# Suppress noise from killing mausezahn.
+	{ kill %% && wait; } 2>/dev/null
+
+	log_test "UC performace under MC overload"
+
+	echo "UC-only throughput  $(humanize $ucth1)"
+	echo "UC+MC throughput    $(humanize $ucth2)"
+	echo "Degradation         $deg %"
+	echo
+	echo "Full report:"
+	echo "  UC only:"
+	echo "    ingress UC throughput $(humanize ${uc_rate[0]})"
+	echo "    egress UC throughput  $(humanize ${uc_rate[1]})"
+	echo "  UC+MC:"
+	echo "    ingress UC throughput $(humanize ${uc_rate_2[0]})"
+	echo "    egress UC throughput  $(humanize ${uc_rate_2[1]})"
+	echo "    ingress MC throughput $(humanize $mc_ir)"
+	echo "    egress MC throughput  $(humanize $mc_er)"
+}
+
+trap cleanup EXIT
+
+setup_prepare
+setup_wait
+
+tests_run
+
+exit $EXIT_STATUS

tools/testing/selftests/net/forwarding/lib.sh

@@ -494,6 +494,14 @@ tc_rule_stats_get()
 	    | jq '.[1].options.actions[].stats.packets'
 }
 
+ethtool_stats_get()
+{
+	local dev=$1; shift
+	local stat=$1; shift
+
+	ethtool -S $dev | grep "^ *$stat:" | head -n 1 | cut -d: -f2
+}
+
 mac_get()
 {
 	local if_name=$1
@@ -541,6 +549,23 @@ forwarding_restore()
 	sysctl_restore net.ipv4.conf.all.forwarding
 }
 
+declare -A MTU_ORIG
+mtu_set()
+{
+	local dev=$1; shift
+	local mtu=$1; shift
+
+	MTU_ORIG["$dev"]=$(ip -j link show dev $dev | jq -e '.[].mtu')
+	ip link set dev $dev mtu $mtu
+}
+
+mtu_restore()
+{
+	local dev=$1; shift
+
+	ip link set dev $dev mtu ${MTU_ORIG["$dev"]}
+}
+
 tc_offload_check()
 {
 	local num_netifs=${1:-$NUM_NETIFS}