1#!/bin/bash
2# SPDX-License-Identifier: GPL-2.0
3#
4# This test sends many small packets (size is less than cell size) through the
5# switch. A shaper is used in $swp2, so the traffic is limited there. Packets
6# are queued till they will be sent.
7#
8# The idea is to verify that the switch can handle at least 85% of maximum
9# supported descrpitors by hardware. Then, we verify that the driver configures
10# firmware to allow infinite size of egress descriptor pool, and does not use a
11# lower limitation. Increase the size of the relevant pools such that the pool's
12# size does not limit the traffic.
13
14# +-----------------------+
15# | H1                    |
16# |   + $h1.111           |
17# |   | 192.0.2.33/28     |
18# |   |                   |
19# |   + $h1               |
20# +---|-------------------+
21#     |
22# +---|-----------------------------+
23# |   + $swp1                       |
24# |   | iPOOL1                      |
25# |   |                             |
26# | +-|------------------------+    |
27# | | + $swp1.111              |    |
28# | |                          |    |
29# | | BR1                      |    |
30# | |                          |    |
31# | | + $swp2.111              |    |
32# | +-|------------------------+    |
33# |   |                             |
34# |   + $swp2                       |
35# |   | ePOOL6                      |
36# |   | 1mbit                       |
37# +---+-----------------------------+
38#     |
39# +---|-------------------+
40# |   + $h2            H2 |
41# |   |                   |
42# |   + $h2.111           |
43# |     192.0.2.34/28     |
44# +-----------------------+
45#
46
47ALL_TESTS="
48	ping_ipv4
49	max_descriptors
50"
51
52lib_dir=$(dirname $0)/../../../net/forwarding
53
54NUM_NETIFS=4
55source $lib_dir/lib.sh
56source $lib_dir/devlink_lib.sh
57source mlxsw_lib.sh
58
59MAX_POOL_SIZE=$(devlink_pool_size_get)
60SHAPER_RATE=1mbit
61
62# The current TBF qdisc interface does not allow us to configure the shaper to
63# flat zero. The ASIC shaper is guaranteed to work with a granularity of
64# 200Mbps. On Spectrum-2, writing a value close to zero instead of zero works
65# well, but the performance on Spectrum-1 is unpredictable. Thus, do not run the
66# test on Spectrum-1.
67mlxsw_only_on_spectrum 2+ || exit
68
69h1_create()
70{
71	simple_if_init $h1
72
73	vlan_create $h1 111 v$h1 192.0.2.33/28
74	ip link set dev $h1.111 type vlan egress-qos-map 0:1
75}
76
77h1_destroy()
78{
79	vlan_destroy $h1 111
80
81	simple_if_fini $h1
82}
83
84h2_create()
85{
86	simple_if_init $h2
87
88	vlan_create $h2 111 v$h2 192.0.2.34/28
89}
90
91h2_destroy()
92{
93	vlan_destroy $h2 111
94
95	simple_if_fini $h2
96}
97
98switch_create()
99{
100	# pools
101	# -----
102
103	devlink_pool_size_thtype_save 1
104	devlink_pool_size_thtype_save 6
105
106	devlink_port_pool_th_save $swp1 1
107	devlink_port_pool_th_save $swp2 6
108
109	devlink_tc_bind_pool_th_save $swp1 1 ingress
110	devlink_tc_bind_pool_th_save $swp2 1 egress
111
112	devlink_pool_size_thtype_set 1 dynamic $MAX_POOL_SIZE
113	devlink_pool_size_thtype_set 6 static $MAX_POOL_SIZE
114
115	# $swp1
116	# -----
117
118	ip link set dev $swp1 up
119	vlan_create $swp1 111
120	ip link set dev $swp1.111 type vlan ingress-qos-map 0:0 1:1
121
122	devlink_port_pool_th_set $swp1 1 16
123	devlink_tc_bind_pool_th_set $swp1 1 ingress 1 16
124
125	tc qdisc replace dev $swp1 root handle 1: \
126	   ets bands 8 strict 8 priomap 7 6
127	dcb buffer set dev $swp1 prio-buffer all:0 1:1
128
129	# $swp2
130	# -----
131
132	ip link set dev $swp2 up
133	vlan_create $swp2 111
134	ip link set dev $swp2.111 type vlan egress-qos-map 0:0 1:1
135
136	devlink_port_pool_th_set $swp2 6 $MAX_POOL_SIZE
137	devlink_tc_bind_pool_th_set $swp2 1 egress 6 $MAX_POOL_SIZE
138
139	tc qdisc replace dev $swp2 root handle 1: tbf rate $SHAPER_RATE \
140		burst 128K limit 500M
141	tc qdisc replace dev $swp2 parent 1:1 handle 11: \
142		ets bands 8 strict 8 priomap 7 6
143
144	# bridge
145	# ------
146
147	ip link add name br1 type bridge vlan_filtering 0
148	ip link set dev $swp1.111 master br1
149	ip link set dev br1 up
150
151	ip link set dev $swp2.111 master br1
152}
153
154switch_destroy()
155{
156	# Do this first so that we can reset the limits to values that are only
157	# valid for the original static / dynamic setting.
158	devlink_pool_size_thtype_restore 6
159	devlink_pool_size_thtype_restore 1
160
161	# bridge
162	# ------
163
164	ip link set dev $swp2.111 nomaster
165
166	ip link set dev br1 down
167	ip link set dev $swp1.111 nomaster
168	ip link del dev br1
169
170	# $swp2
171	# -----
172
173	tc qdisc del dev $swp2 parent 1:1 handle 11:
174	tc qdisc del dev $swp2 root
175
176	devlink_tc_bind_pool_th_restore $swp2 1 egress
177	devlink_port_pool_th_restore $swp2 6
178
179	vlan_destroy $swp2 111
180	ip link set dev $swp2 down
181
182	# $swp1
183	# -----
184
185	dcb buffer set dev $swp1 prio-buffer all:0
186	tc qdisc del dev $swp1 root
187
188	devlink_tc_bind_pool_th_restore $swp1 1 ingress
189	devlink_port_pool_th_restore $swp1 1
190
191	vlan_destroy $swp1 111
192	ip link set dev $swp1 down
193}
194
195setup_prepare()
196{
197	h1=${NETIFS[p1]}
198	swp1=${NETIFS[p2]}
199
200	swp2=${NETIFS[p3]}
201	h2=${NETIFS[p4]}
202
203	h2mac=$(mac_get $h2)
204
205	vrf_prepare
206
207	h1_create
208	h2_create
209	switch_create
210}
211
212cleanup()
213{
214	pre_cleanup
215
216	switch_destroy
217	h2_destroy
218	h1_destroy
219
220	vrf_cleanup
221}
222
223ping_ipv4()
224{
225	ping_test $h1 192.0.2.34 " h1->h2"
226}
227
228percentage_used()
229{
230	local num_packets=$1; shift
231	local max_packets=$1; shift
232
233	bc <<< "
234	    scale=2
235	    100 * $num_packets / $max_packets
236	"
237}
238
239max_descriptors()
240{
241	local cell_size=$(devlink_cell_size_get)
242	local exp_perc_used=85
243	local max_descriptors
244	local pktsize=30
245
246	RET=0
247
248	max_descriptors=$(mlxsw_max_descriptors_get) || exit 1
249
250	local d0=$(ethtool_stats_get $swp2 tc_no_buffer_discard_uc_tc_1)
251
252	log_info "Send many small packets, packet size = $pktsize bytes"
253	start_traffic_pktsize $pktsize $h1.111 192.0.2.33 192.0.2.34 $h2mac
254
255	# Sleep to wait for congestion.
256	sleep 5
257
258	local d1=$(ethtool_stats_get $swp2 tc_no_buffer_discard_uc_tc_1)
259	((d1 == d0))
260	check_err $? "Drops seen on egress port: $d0 -> $d1 ($((d1 - d0)))"
261
262	# Check how many packets the switch can handle, the limitation is
263	# maximum descriptors.
264	local pkts_bytes=$(ethtool_stats_get $swp2 tc_transmit_queue_tc_1)
265	local pkts_num=$((pkts_bytes / cell_size))
266	local perc_used=$(percentage_used $pkts_num $max_descriptors)
267
268	check_err $(bc <<< "$perc_used < $exp_perc_used") \
269		"Expected > $exp_perc_used% of descriptors, handle $perc_used%"
270
271	stop_traffic
272	sleep 1
273
274	log_test "Maximum descriptors usage. The percentage used is $perc_used%"
275}
276
277trap cleanup EXIT
278setup_prepare
279setup_wait
280tests_run
281
282exit $EXIT_STATUS
283