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

   1 #!/bin/bash
   2 # SPDX-License-Identifier: GPL-2.0
   3 #
   4 # A test for switch behavior under MC overload. An issue in Spectrum chips
   5 # causes throughput of UC traffic to drop severely when a switch is under heavy
   6 # MC load. This issue can be overcome by putting the switch to MC-aware mode.
   7 # This test verifies that UC performance stays intact even as the switch is
   8 # under MC flood, and therefore that the MC-aware mode is enabled and correctly
   9 # configured.
  10 #
  11 # Because mlxsw throttles CPU port, the traffic can't actually reach userspace
  12 # at full speed. That makes it impossible to use iperf3 to simply measure the
  13 # throughput, because many packets (that reach $h3) don't get to the kernel at
  14 # all even in UDP mode (the situation is even worse in TCP mode, where one can't
  15 # hope to see more than a couple Mbps).
  16 #
  17 # So instead we send traffic with mausezahn and use RX ethtool counters at $h3.
  18 # Multicast traffic is untagged, unicast traffic is tagged with PCP 1. Therefore
  19 # each gets a different priority and we can use per-prio ethtool counters to
  20 # measure the throughput. In order to avoid prioritizing unicast traffic, prio
  21 # qdisc is installed on $swp3 and maps all priorities to the same band #7 (and
  22 # thus TC 0).
  23 #
  24 # Mausezahn can't actually saturate the links unless it's using large frames.
  25 # Thus we set MTU to 10K on all involved interfaces. Then both unicast and
  26 # multicast traffic uses 8K frames.
  27 #
  28 # +---------------------------+            +----------------------------------+
  29 # | H1                        |            |                               H2 |
  30 # |                           |            |  unicast --> + $h2.111           |
  31 # |                 multicast |            |  traffic     | 192.0.2.129/28    |
  32 # |                 traffic   |            |              | e-qos-map 0:1     |
  33 # |           $h1 + <-----    |            |              |                   |
  34 # | 192.0.2.65/28 |           |            |              + $h2               |
  35 # +---------------|-----------+            +--------------|-------------------+
  36 #                 |                                       |
  37 # +---------------|---------------------------------------|-------------------+
  38 # |         $swp1 +                                       + $swp2             |
  39 # |        >1Gbps |                                       | >1Gbps            |
  40 # | +-------------|------+                     +----------|----------------+  |
  41 # | |     $swp1.1 +      |                     |          + $swp2.111      |  |
  42 # | |                BR1 |             SW      | BR111                     |  |
  43 # | |     $swp3.1 +      |                     |          + $swp3.111      |  |
  44 # | +-------------|------+                     +----------|----------------+  |
  45 # |               \_______________________________________/                   |
  46 # |                                    |                                      |
  47 # |                                    + $swp3                                |
  48 # |                                    | 1Gbps bottleneck                     |
  49 # |                                    | prio qdisc: {0..7} -> 7              |
  50 # +------------------------------------|--------------------------------------+
  51 #                                      |
  52 #                                   +--|-----------------+
  53 #                                   |  + $h3          H3 |
  54 #                                   |  | 192.0.2.66/28   |
  55 #                                   |  |                 |
  56 #                                   |  + $h3.111         |
  57 #                                   |    192.0.2.130/28  |
  58 #                                   +--------------------+
  59
  60 ALL_TESTS="
  61         ping_ipv4
  62         test_mc_aware
  63         test_uc_aware
  64 "
  65
  66 lib_dir=$(dirname $0)/../../../net/forwarding
  67
  68 NUM_NETIFS=6
  69 source $lib_dir/lib.sh
  70 source $lib_dir/devlink_lib.sh
  71 source qos_lib.sh
  72
  73 h1_create()
  74 {
  75         simple_if_init $h1 192.0.2.65/28
  76         mtu_set $h1 10000
  77 }
  78
  79 h1_destroy()
  80 {
  81         mtu_restore $h1
  82         simple_if_fini $h1 192.0.2.65/28
  83 }
  84
  85 h2_create()
  86 {
  87         simple_if_init $h2
  88         mtu_set $h2 10000
  89
  90         vlan_create $h2 111 v$h2 192.0.2.129/28
  91         ip link set dev $h2.111 type vlan egress-qos-map 0:1
  92 }
  93
  94 h2_destroy()
  95 {
  96         vlan_destroy $h2 111
  97
  98         mtu_restore $h2
  99         simple_if_fini $h2
 100 }
 101
 102 h3_create()
 103 {
 104         simple_if_init $h3 192.0.2.66/28
 105         mtu_set $h3 10000
 106
 107         vlan_create $h3 111 v$h3 192.0.2.130/28
 108 }
 109
 110 h3_destroy()
 111 {
 112         vlan_destroy $h3 111
 113
 114         mtu_restore $h3
 115         simple_if_fini $h3 192.0.2.66/28
 116 }
 117
 118 switch_create()
 119 {
 120         ip link set dev $swp1 up
 121         mtu_set $swp1 10000
 122
 123         ip link set dev $swp2 up
 124         mtu_set $swp2 10000
 125
 126         ip link set dev $swp3 up
 127         mtu_set $swp3 10000
 128
 129         vlan_create $swp2 111
 130         vlan_create $swp3 111
 131
 132         ethtool -s $swp3 speed 1000 autoneg off
 133         tc qdisc replace dev $swp3 root handle 3: \
 134            prio bands 8 priomap 7 7 7 7 7 7 7 7
 135
 136         ip link add name br1 type bridge vlan_filtering 0
 137         ip link set dev br1 up
 138         ip link set dev $swp1 master br1
 139         ip link set dev $swp3 master br1
 140
 141         ip link add name br111 type bridge vlan_filtering 0
 142         ip link set dev br111 up
 143         ip link set dev $swp2.111 master br111
 144         ip link set dev $swp3.111 master br111
 145
 146         # Make sure that ingress quotas are smaller than egress so that there is
 147         # room for both streams of traffic to be admitted to shared buffer.
 148         devlink_port_pool_th_set $swp1 0 5
 149         devlink_tc_bind_pool_th_set $swp1 0 ingress 0 5
 150
 151         devlink_port_pool_th_set $swp2 0 5
 152         devlink_tc_bind_pool_th_set $swp2 1 ingress 0 5
 153
 154         devlink_port_pool_th_set $swp3 4 12
 155 }
 156
 157 switch_destroy()
 158 {
 159         devlink_port_pool_th_restore $swp3 4
 160
 161         devlink_tc_bind_pool_th_restore $swp2 1 ingress
 162         devlink_port_pool_th_restore $swp2 0
 163
 164         devlink_tc_bind_pool_th_restore $swp1 0 ingress
 165         devlink_port_pool_th_restore $swp1 0
 166
 167         ip link del dev br111
 168         ip link del dev br1
 169
 170         tc qdisc del dev $swp3 root handle 3:
 171         ethtool -s $swp3 autoneg on
 172
 173         vlan_destroy $swp3 111
 174         vlan_destroy $swp2 111
 175
 176         mtu_restore $swp3
 177         ip link set dev $swp3 down
 178
 179         mtu_restore $swp2
 180         ip link set dev $swp2 down
 181
 182         mtu_restore $swp1
 183         ip link set dev $swp1 down
 184 }
 185
 186 setup_prepare()
 187 {
 188         h1=${NETIFS[p1]}
 189         swp1=${NETIFS[p2]}
 190
 191         swp2=${NETIFS[p3]}
 192         h2=${NETIFS[p4]}
 193
 194         swp3=${NETIFS[p5]}
 195         h3=${NETIFS[p6]}
 196
 197         h3mac=$(mac_get $h3)
 198
 199         vrf_prepare
 200
 201         h1_create
 202         h2_create
 203         h3_create
 204         switch_create
 205 }
 206
 207 cleanup()
 208 {
 209         pre_cleanup
 210
 211         switch_destroy
 212         h3_destroy
 213         h2_destroy
 214         h1_destroy
 215
 216         vrf_cleanup
 217 }
 218
 219 ping_ipv4()
 220 {
 221         ping_test $h2 192.0.2.130
 222 }
 223
 224 test_mc_aware()
 225 {
 226         RET=0
 227
 228         local -a uc_rate
 229         start_traffic $h2.111 192.0.2.129 192.0.2.130 $h3mac
 230         uc_rate=($(measure_rate $swp2 $h3 rx_octets_prio_1 "UC-only"))
 231         check_err $? "Could not get high enough UC-only ingress rate"
 232         stop_traffic
 233         local ucth1=${uc_rate[1]}
 234
 235         start_traffic $h1 own bc bc
 236
 237         local d0=$(date +%s)
 238         local t0=$(ethtool_stats_get $h3 rx_octets_prio_0)
 239         local u0=$(ethtool_stats_get $swp1 rx_octets_prio_0)
 240
 241         local -a uc_rate_2
 242         start_traffic $h2.111 192.0.2.129 192.0.2.130 $h3mac
 243         uc_rate_2=($(measure_rate $swp2 $h3 rx_octets_prio_1 "UC+MC"))
 244         check_err $? "Could not get high enough UC+MC ingress rate"
 245         stop_traffic
 246         local ucth2=${uc_rate_2[1]}
 247
 248         local d1=$(date +%s)
 249         local t1=$(ethtool_stats_get $h3 rx_octets_prio_0)
 250         local u1=$(ethtool_stats_get $swp1 rx_octets_prio_0)
 251
 252         local deg=$(bc <<< "
 253                         scale=2
 254                         ret = 100 * ($ucth1 - $ucth2) / $ucth1
 255                         if (ret > 0) { ret } else { 0 }
 256                     ")
 257         check_err $(bc <<< "$deg > 25")
 258
 259         local interval=$((d1 - d0))
 260         local mc_ir=$(rate $u0 $u1 $interval)
 261         local mc_er=$(rate $t0 $t1 $interval)
 262
 263         stop_traffic
 264
 265         log_test "UC performance under MC overload"
 266
 267         echo "UC-only throughput  $(humanize $ucth1)"
 268         echo "UC+MC throughput    $(humanize $ucth2)"
 269         echo "Degradation         $deg %"
 270         echo
 271         echo "Full report:"
 272         echo "  UC only:"
 273         echo "    ingress UC throughput $(humanize ${uc_rate[0]})"
 274         echo "    egress UC throughput  $(humanize ${uc_rate[1]})"
 275         echo "  UC+MC:"
 276         echo "    ingress UC throughput $(humanize ${uc_rate_2[0]})"
 277         echo "    egress UC throughput  $(humanize ${uc_rate_2[1]})"
 278         echo "    ingress MC throughput $(humanize $mc_ir)"
 279         echo "    egress MC throughput  $(humanize $mc_er)"
 280         echo
 281 }
 282
 283 test_uc_aware()
 284 {
 285         RET=0
 286
 287         start_traffic $h2.111 192.0.2.129 192.0.2.130 $h3mac
 288
 289         local d0=$(date +%s)
 290         local t0=$(ethtool_stats_get $h3 rx_octets_prio_1)
 291         local u0=$(ethtool_stats_get $swp2 rx_octets_prio_1)
 292         sleep 1
 293
 294         local attempts=50
 295         local passes=0
 296         local i
 297
 298         for ((i = 0; i < attempts; ++i)); do
 299                 if $ARPING -c 1 -I $h1 -b 192.0.2.66 -q -w 0.1; then
 300                         ((passes++))
 301                 fi
 302
 303                 sleep 0.1
 304         done
 305
 306         local d1=$(date +%s)
 307         local t1=$(ethtool_stats_get $h3 rx_octets_prio_1)
 308         local u1=$(ethtool_stats_get $swp2 rx_octets_prio_1)
 309
 310         local interval=$((d1 - d0))
 311         local uc_ir=$(rate $u0 $u1 $interval)
 312         local uc_er=$(rate $t0 $t1 $interval)
 313
 314         ((attempts == passes))
 315         check_err $?
 316
 317         stop_traffic
 318
 319         log_test "MC performance under UC overload"
 320         echo "    ingress UC throughput $(humanize ${uc_ir})"
 321         echo "    egress UC throughput  $(humanize ${uc_er})"
 322         echo "    sent $attempts BC ARPs, got $passes responses"
 323 }
 324
 325 trap cleanup EXIT
 326
 327 setup_prepare
 328 setup_wait
 329
 330 tests_run
 331
 332 exit $EXIT_STATUS