418 lines · bash
1#!/bin/bash2# SPDX-License-Identifier: GPL-2.03#4# This test injects a 10-MB burst of traffic with VLAN tag and 802.1p priority5# of 1. This stream is consistently prioritized as priority 1, is put to PG6# buffer 1, and scheduled at TC 1.7#8# - the stream first ingresses through $swp1, where it is forwarded to $swp39#10# - then it ingresses through $swp4. Here it is put to a lossless buffer and put11# to a small pool ("PFC pool"). The traffic is forwarded to $swp2, which is12# shaped, and thus the PFC pool eventually fills, therefore the headroom13# fills, and $swp3 is paused.14#15# - since $swp3 now can't send traffic, the traffic ingressing $swp1 is kept at16# a pool ("overflow pool"). The overflow pool needs to be large enough to17# contain the whole burst.18#19# - eventually the PFC pool gets some traffic out, headroom therefore gets some20# traffic to the pool, and $swp3 is unpaused again. This way the traffic is21# gradually forwarded from the overflow pool, through the PFC pool, out of22# $swp2, and eventually to $h2.23#24# - if PFC works, all lossless flow packets that ingress through $swp1 should25# also be seen ingressing $h2. If it doesn't, there will be drops due to26# discrepancy between the speeds of $swp1 and $h2.27#28# - it should all play out relatively quickly, so that SLL and HLL will not29# cause drops.30#31# +-----------------------+32# | H1 |33# | + $h1.111 |34# | | 192.0.2.33/28 |35# | | |36# | + $h1 |37# +---|-------------------+ +--------------------+38# | | |39# +---|----------------------|--------------------|---------------------------+40# | + $swp1 $swp3 + + $swp4 |41# | | iPOOL1 iPOOL0 | | iPOOL2 |42# | | ePOOL4 ePOOL5 | | ePOOL4 |43# | | PFC:enabled=1 | | PFC:enabled=1 |44# | +-|----------------------|-+ +-|------------------------+ |45# | | + $swp1.111 $swp3.111 + | | + $swp4.111 | |46# | | | | | |47# | | BR1 | | BR2 | |48# | | | | | |49# | | | | + $swp2.111 | |50# | +--------------------------+ +---------|----------------+ |51# | | |52# | iPOOL0: 500KB dynamic | |53# | iPOOL1: 10MB static | |54# | iPOOL2: 1MB static + $swp2 |55# | ePOOL4: 500KB dynamic | iPOOL0 |56# | ePOOL5: 10MB static | ePOOL6 |57# | ePOOL6: "infinite" static | 200Mbps shaper |58# +-------------------------------------------------------|-------------------+59# |60# +---|-------------------+61# | + $h2 H2 |62# | | |63# | + $h2.111 |64# | 192.0.2.34/28 |65# +-----------------------+66#67# iPOOL0+ePOOL4 is a helper pool for control traffic etc.68# iPOOL1+ePOOL5 are overflow pools.69# iPOOL2+ePOOL6 are PFC pools.70 71ALL_TESTS="72 ping_ipv473 test_qos_pfc74"75 76lib_dir=$(dirname $0)/../../../net/forwarding77 78NUM_NETIFS=679source $lib_dir/lib.sh80source $lib_dir/devlink_lib.sh81 82_1KB=100083_100KB=$((100 * _1KB))84_500KB=$((500 * _1KB))85_1MB=$((1000 * _1KB))86_10MB=$((10 * _1MB))87 88h1_create()89{90 simple_if_init $h191 mtu_set $h1 1000092 93 vlan_create $h1 111 v$h1 192.0.2.33/2894}95 96h1_destroy()97{98 vlan_destroy $h1 11199 100 mtu_restore $h1101 simple_if_fini $h1102}103 104h2_create()105{106 simple_if_init $h2107 mtu_set $h2 10000108 109 vlan_create $h2 111 v$h2 192.0.2.34/28110}111 112h2_destroy()113{114 vlan_destroy $h2 111115 116 mtu_restore $h2117 simple_if_fini $h2118}119 120switch_create()121{122 local lanes_swp4123 local pg1_size124 125 # pools126 # -----127 128 devlink_pool_size_thtype_save 0129 devlink_pool_size_thtype_save 4130 devlink_pool_size_thtype_save 1131 devlink_pool_size_thtype_save 5132 devlink_pool_size_thtype_save 2133 devlink_pool_size_thtype_save 6134 135 devlink_port_pool_th_save $swp1 1136 devlink_port_pool_th_save $swp2 6137 devlink_port_pool_th_save $swp3 5138 devlink_port_pool_th_save $swp4 2139 140 devlink_tc_bind_pool_th_save $swp1 1 ingress141 devlink_tc_bind_pool_th_save $swp2 1 egress142 devlink_tc_bind_pool_th_save $swp3 1 egress143 devlink_tc_bind_pool_th_save $swp4 1 ingress144 145 # Control traffic pools. Just reduce the size. Keep them dynamic so that146 # we don't need to change all the uninteresting quotas.147 devlink_pool_size_thtype_set 0 dynamic $_500KB148 devlink_pool_size_thtype_set 4 dynamic $_500KB149 150 # Overflow pools.151 devlink_pool_size_thtype_set 1 static $_10MB152 devlink_pool_size_thtype_set 5 static $_10MB153 154 # PFC pools. As per the writ, the size of egress PFC pool should be155 # infinice, but actually it just needs to be large enough to not matter156 # in practice, so reuse the 10MB limit.157 devlink_pool_size_thtype_set 2 static $_1MB158 devlink_pool_size_thtype_set 6 static $_10MB159 160 # $swp1161 # -----162 163 ip link set dev $swp1 up164 mtu_set $swp1 10000165 vlan_create $swp1 111166 ip link set dev $swp1.111 type vlan ingress-qos-map 0:0 1:1167 168 devlink_port_pool_th_set $swp1 1 $_10MB169 devlink_tc_bind_pool_th_set $swp1 1 ingress 1 $_10MB170 171 # Configure qdisc so that we can configure PG and therefore pool172 # assignment.173 tc qdisc replace dev $swp1 root handle 1: \174 ets bands 8 strict 8 priomap 7 6175 dcb buffer set dev $swp1 prio-buffer all:0 1:1176 177 # $swp2178 # -----179 180 ip link set dev $swp2 up181 mtu_set $swp2 10000182 vlan_create $swp2 111183 ip link set dev $swp2.111 type vlan egress-qos-map 0:0 1:1184 185 devlink_port_pool_th_set $swp2 6 $_10MB186 devlink_tc_bind_pool_th_set $swp2 1 egress 6 $_10MB187 188 # prio 0->TC0 (band 7), 1->TC1 (band 6). TC1 is shaped.189 tc qdisc replace dev $swp2 root handle 1: \190 ets bands 8 strict 8 priomap 7 6191 tc qdisc replace dev $swp2 parent 1:7 handle 17: \192 tbf rate 200Mbit burst 131072 limit 1M193 194 # $swp3195 # -----196 197 ip link set dev $swp3 up198 mtu_set $swp3 10000199 vlan_create $swp3 111200 ip link set dev $swp3.111 type vlan egress-qos-map 0:0 1:1201 202 devlink_port_pool_th_set $swp3 5 $_10MB203 devlink_tc_bind_pool_th_set $swp3 1 egress 5 $_10MB204 205 # prio 0->TC0 (band 7), 1->TC1 (band 6)206 tc qdisc replace dev $swp3 root handle 1: \207 ets bands 8 strict 8 priomap 7 6208 209 # Need to enable PFC so that PAUSE takes effect. Therefore need to put210 # the lossless prio into a buffer of its own. Don't bother with buffer211 # sizes though, there is not going to be any pressure in the "backward"212 # direction.213 dcb buffer set dev $swp3 prio-buffer all:0 1:1214 dcb pfc set dev $swp3 prio-pfc all:off 1:on215 216 # $swp4217 # -----218 219 ip link set dev $swp4 up220 mtu_set $swp4 10000221 vlan_create $swp4 111222 ip link set dev $swp4.111 type vlan ingress-qos-map 0:0 1:1223 224 devlink_port_pool_th_set $swp4 2 $_1MB225 devlink_tc_bind_pool_th_set $swp4 1 ingress 2 $_1MB226 227 # Configure qdisc so that we can hand-tune headroom.228 tc qdisc replace dev $swp4 root handle 1: \229 ets bands 8 strict 8 priomap 7 6230 dcb buffer set dev $swp4 prio-buffer all:0 1:1231 dcb pfc set dev $swp4 prio-pfc all:off 1:on232 # PG0 will get autoconfigured to Xoff, give PG1 arbitrarily 100K, which233 # is (-2*MTU) about 80K of delay provision.234 pg1_size=$_100KB235 236 setup_wait_dev_with_timeout $swp4237 238 lanes_swp4=$(ethtool $swp4 | grep 'Lanes:')239 lanes_swp4=${lanes_swp4#*"Lanes: "}240 241 # 8-lane ports use two buffers among which the configured buffer242 # is split, so double the size to get twice (20K + 80K).243 if [[ $lanes_swp4 -eq 8 ]]; then244 pg1_size=$((pg1_size * 2))245 fi246 247 dcb buffer set dev $swp4 buffer-size all:0 1:$pg1_size248 249 # bridges250 # -------251 252 ip link add name br1 type bridge vlan_filtering 0253 ip link set dev $swp1.111 master br1254 ip link set dev $swp3.111 master br1255 ip link set dev br1 up256 257 ip link add name br2 type bridge vlan_filtering 0258 ip link set dev $swp2.111 master br2259 ip link set dev $swp4.111 master br2260 ip link set dev br2 up261}262 263switch_destroy()264{265 # Do this first so that we can reset the limits to values that are only266 # valid for the original static / dynamic setting.267 devlink_pool_size_thtype_restore 6268 devlink_pool_size_thtype_restore 5269 devlink_pool_size_thtype_restore 4270 devlink_pool_size_thtype_restore 2271 devlink_pool_size_thtype_restore 1272 devlink_pool_size_thtype_restore 0273 274 # bridges275 # -------276 277 ip link set dev br2 down278 ip link set dev $swp4.111 nomaster279 ip link set dev $swp2.111 nomaster280 ip link del dev br2281 282 ip link set dev br1 down283 ip link set dev $swp3.111 nomaster284 ip link set dev $swp1.111 nomaster285 ip link del dev br1286 287 # $swp4288 # -----289 290 dcb buffer set dev $swp4 buffer-size all:0291 dcb pfc set dev $swp4 prio-pfc all:off292 dcb buffer set dev $swp4 prio-buffer all:0293 tc qdisc del dev $swp4 root294 295 devlink_tc_bind_pool_th_restore $swp4 1 ingress296 devlink_port_pool_th_restore $swp4 2297 298 vlan_destroy $swp4 111299 mtu_restore $swp4300 ip link set dev $swp4 down301 302 # $swp3303 # -----304 305 dcb pfc set dev $swp3 prio-pfc all:off306 dcb buffer set dev $swp3 prio-buffer all:0307 tc qdisc del dev $swp3 root308 309 devlink_tc_bind_pool_th_restore $swp3 1 egress310 devlink_port_pool_th_restore $swp3 5311 312 vlan_destroy $swp3 111313 mtu_restore $swp3314 ip link set dev $swp3 down315 316 # $swp2317 # -----318 319 tc qdisc del dev $swp2 parent 1:7320 tc qdisc del dev $swp2 root321 322 devlink_tc_bind_pool_th_restore $swp2 1 egress323 devlink_port_pool_th_restore $swp2 6324 325 vlan_destroy $swp2 111326 mtu_restore $swp2327 ip link set dev $swp2 down328 329 # $swp1330 # -----331 332 dcb buffer set dev $swp1 prio-buffer all:0333 tc qdisc del dev $swp1 root334 335 devlink_tc_bind_pool_th_restore $swp1 1 ingress336 devlink_port_pool_th_restore $swp1 1337 338 vlan_destroy $swp1 111339 mtu_restore $swp1340 ip link set dev $swp1 down341}342 343setup_prepare()344{345 h1=${NETIFS[p1]}346 swp1=${NETIFS[p2]}347 348 swp2=${NETIFS[p3]}349 h2=${NETIFS[p4]}350 351 swp3=${NETIFS[p5]}352 swp4=${NETIFS[p6]}353 354 h2mac=$(mac_get $h2)355 356 vrf_prepare357 358 h1_create359 h2_create360 switch_create361}362 363cleanup()364{365 pre_cleanup366 367 switch_destroy368 h2_destroy369 h1_destroy370 371 vrf_cleanup372}373 374ping_ipv4()375{376 ping_test $h1 192.0.2.34377}378 379test_qos_pfc()380{381 RET=0382 383 # 10M pool, each packet is 8K of payload + headers384 local pkts=$((_10MB / 8050))385 local size=$((pkts * 8050))386 local in0=$(ethtool_stats_get $swp1 rx_octets_prio_1)387 local out0=$(ethtool_stats_get $swp2 tx_octets_prio_1)388 389 $MZ $h1 -p 8000 -Q 1:111 -A 192.0.2.33 -B 192.0.2.34 \390 -a own -b $h2mac -c $pkts -t udp -q391 sleep 2392 393 local in1=$(ethtool_stats_get $swp1 rx_octets_prio_1)394 local out1=$(ethtool_stats_get $swp2 tx_octets_prio_1)395 396 local din=$((in1 - in0))397 local dout=$((out1 - out0))398 399 local pct_in=$((din * 100 / size))400 401 ((pct_in > 95 && pct_in < 105))402 check_err $? "Relative ingress out of expected bounds, $pct_in% should be 100%"403 404 ((dout == din))405 check_err $? "$((din - dout)) bytes out of $din ingressed got lost"406 407 log_test "PFC"408}409 410bail_on_lldpad "configure DCB" "configure Qdiscs"411 412trap cleanup EXIT413setup_prepare414setup_wait415tests_run416 417exit $EXIT_STATUS418