tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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# | + $h3.111 |
57# | 192.0.2.130/28 |
58# +--------------------+
59
60ALL_TESTS="
61 ping_ipv4
62 test_mc_aware
63 test_uc_aware
64"
65
66lib_dir=$(dirname $0)/../../../net/forwarding
67
68NUM_NETIFS=6
69source $lib_dir/lib.sh
70source $lib_dir/devlink_lib.sh
71source qos_lib.sh
72
73h1_create()
74{
75 simple_if_init $h1 192.0.2.65/28
76 mtu_set $h1 10000
77}
78
79h1_destroy()
80{
81 mtu_restore $h1
82 simple_if_fini $h1 192.0.2.65/28
83}
84
85h2_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
94h2_destroy()
95{
96 vlan_destroy $h2 111
97
98 mtu_restore $h2
99 simple_if_fini $h2
100}
101
102h3_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
110h3_destroy()
111{
112 vlan_destroy $h3 111
113
114 mtu_restore $h3
115 simple_if_fini $h3 192.0.2.66/28
116}
117
118switch_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_save $swp1 0
149 devlink_port_pool_th_set $swp1 0 5
150 devlink_tc_bind_pool_th_save $swp1 0 ingress
151 devlink_tc_bind_pool_th_set $swp1 0 ingress 0 5
152
153 devlink_port_pool_th_save $swp2 0
154 devlink_port_pool_th_set $swp2 0 5
155 devlink_tc_bind_pool_th_save $swp2 1 ingress
156 devlink_tc_bind_pool_th_set $swp2 1 ingress 0 5
157
158 devlink_port_pool_th_save $swp3 4
159 devlink_port_pool_th_set $swp3 4 12
160}
161
162switch_destroy()
163{
164 devlink_port_pool_th_restore $swp3 4
165
166 devlink_tc_bind_pool_th_restore $swp2 1 ingress
167 devlink_port_pool_th_restore $swp2 0
168
169 devlink_tc_bind_pool_th_restore $swp1 0 ingress
170 devlink_port_pool_th_restore $swp1 0
171
172 ip link del dev br111
173 ip link del dev br1
174
175 tc qdisc del dev $swp3 root handle 3:
176 ethtool -s $swp3 autoneg on
177
178 vlan_destroy $swp3 111
179 vlan_destroy $swp2 111
180
181 mtu_restore $swp3
182 ip link set dev $swp3 down
183
184 mtu_restore $swp2
185 ip link set dev $swp2 down
186
187 mtu_restore $swp1
188 ip link set dev $swp1 down
189}
190
191setup_prepare()
192{
193 h1=${NETIFS[p1]}
194 swp1=${NETIFS[p2]}
195
196 swp2=${NETIFS[p3]}
197 h2=${NETIFS[p4]}
198
199 swp3=${NETIFS[p5]}
200 h3=${NETIFS[p6]}
201
202 h3mac=$(mac_get $h3)
203
204 vrf_prepare
205
206 h1_create
207 h2_create
208 h3_create
209 switch_create
210}
211
212cleanup()
213{
214 pre_cleanup
215
216 switch_destroy
217 h3_destroy
218 h2_destroy
219 h1_destroy
220
221 vrf_cleanup
222}
223
224ping_ipv4()
225{
226 ping_test $h2 192.0.2.130
227}
228
229test_mc_aware()
230{
231 RET=0
232
233 local -a uc_rate
234 start_traffic $h2.111 192.0.2.129 192.0.2.130 $h3mac
235 uc_rate=($(measure_rate $swp2 $h3 rx_octets_prio_1 "UC-only"))
236 check_err $? "Could not get high enough UC-only ingress rate"
237 stop_traffic
238 local ucth1=${uc_rate[1]}
239
240 start_traffic $h1 192.0.2.65 bc bc
241
242 local d0=$(date +%s)
243 local t0=$(ethtool_stats_get $h3 rx_octets_prio_0)
244 local u0=$(ethtool_stats_get $swp1 rx_octets_prio_0)
245
246 local -a uc_rate_2
247 start_traffic $h2.111 192.0.2.129 192.0.2.130 $h3mac
248 uc_rate_2=($(measure_rate $swp2 $h3 rx_octets_prio_1 "UC+MC"))
249 check_err $? "Could not get high enough UC+MC ingress rate"
250 stop_traffic
251 local ucth2=${uc_rate_2[1]}
252
253 local d1=$(date +%s)
254 local t1=$(ethtool_stats_get $h3 rx_octets_prio_0)
255 local u1=$(ethtool_stats_get $swp1 rx_octets_prio_0)
256
257 local deg=$(bc <<< "
258 scale=2
259 ret = 100 * ($ucth1 - $ucth2) / $ucth1
260 if (ret > 0) { ret } else { 0 }
261 ")
262
263 # Minimum shaper of 200Mbps on MC TCs should cause about 20% of
264 # degradation on 1Gbps link.
265 check_err $(bc <<< "$deg < 15") "Minimum shaper not in effect"
266 check_err $(bc <<< "$deg > 25") "MC traffic degrades UC performance too much"
267
268 local interval=$((d1 - d0))
269 local mc_ir=$(rate $u0 $u1 $interval)
270 local mc_er=$(rate $t0 $t1 $interval)
271
272 stop_traffic
273
274 log_test "UC performance under MC overload"
275
276 echo "UC-only throughput $(humanize $ucth1)"
277 echo "UC+MC throughput $(humanize $ucth2)"
278 echo "Degradation $deg %"
279 echo
280 echo "Full report:"
281 echo " UC only:"
282 echo " ingress UC throughput $(humanize ${uc_rate[0]})"
283 echo " egress UC throughput $(humanize ${uc_rate[1]})"
284 echo " UC+MC:"
285 echo " ingress UC throughput $(humanize ${uc_rate_2[0]})"
286 echo " egress UC throughput $(humanize ${uc_rate_2[1]})"
287 echo " ingress MC throughput $(humanize $mc_ir)"
288 echo " egress MC throughput $(humanize $mc_er)"
289 echo
290}
291
292test_uc_aware()
293{
294 RET=0
295
296 start_traffic $h2.111 192.0.2.129 192.0.2.130 $h3mac
297
298 local d0=$(date +%s)
299 local t0=$(ethtool_stats_get $h3 rx_octets_prio_1)
300 local u0=$(ethtool_stats_get $swp2 rx_octets_prio_1)
301 sleep 1
302
303 local attempts=50
304 local passes=0
305 local i
306
307 for ((i = 0; i < attempts; ++i)); do
308 if $ARPING -c 1 -I $h1 -b 192.0.2.66 -q -w 1; then
309 ((passes++))
310 fi
311
312 sleep 0.1
313 done
314
315 local d1=$(date +%s)
316 local t1=$(ethtool_stats_get $h3 rx_octets_prio_1)
317 local u1=$(ethtool_stats_get $swp2 rx_octets_prio_1)
318
319 local interval=$((d1 - d0))
320 local uc_ir=$(rate $u0 $u1 $interval)
321 local uc_er=$(rate $t0 $t1 $interval)
322
323 ((attempts == passes))
324 check_err $?
325
326 stop_traffic
327
328 log_test "MC performance under UC overload"
329 echo " ingress UC throughput $(humanize ${uc_ir})"
330 echo " egress UC throughput $(humanize ${uc_er})"
331 echo " sent $attempts BC ARPs, got $passes responses"
332}
333
334trap cleanup EXIT
335
336setup_prepare
337setup_wait
338
339tests_run
340
341exit $EXIT_STATUS