1#!/bin/bash
2# SPDX-License-Identifier: GPL-2.0
3#
4# This test injects a 10-MB burst of traffic with VLAN tag and 802.1p priority
5# of 1. This stream is consistently prioritized as priority 1, is put to PG
6# buffer 1, and scheduled at TC 1.
7#
8# - the stream first ingresses through $swp1, where it is forwarded to $swp3
9#
10# - then it ingresses through $swp4. Here it is put to a lossless buffer and put
11#   to a small pool ("PFC pool"). The traffic is forwarded to $swp2, which is
12#   shaped, and thus the PFC pool eventually fills, therefore the headroom
13#   fills, and $swp3 is paused.
14#
15# - since $swp3 now can't send traffic, the traffic ingressing $swp1 is kept at
16#   a pool ("overflow pool"). The overflow pool needs to be large enough to
17#   contain the whole burst.
18#
19# - eventually the PFC pool gets some traffic out, headroom therefore gets some
20#   traffic to the pool, and $swp3 is unpaused again. This way the traffic is
21#   gradually forwarded from the overflow pool, through the PFC pool, out of
22#   $swp2, and eventually to $h2.
23#
24# - if PFC works, all lossless flow packets that ingress through $swp1 should
25#   also be seen ingressing $h2. If it doesn't, there will be drops due to
26#   discrepancy between the speeds of $swp1 and $h2.
27#
28# - it should all play out relatively quickly, so that SLL and HLL will not
29#   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# |   |                1Gbps |                    | 1Gbps                     |
44# |   |        PFC:enabled=1 |                    | PFC:enabled=1             |
45# | +-|----------------------|-+                +-|------------------------+  |
46# | | + $swp1.111  $swp3.111 + |                | + $swp4.111              |  |
47# | |                          |                |                          |  |
48# | | BR1                      |                | BR2                      |  |
49# | |                          |                |                          |  |
50# | |                          |                |         + $swp2.111      |  |
51# | +--------------------------+                +---------|----------------+  |
52# |                                                       |                   |
53# | iPOOL0: 500KB dynamic                                 |                   |
54# | iPOOL1: 10MB static                                   |                   |
55# | iPOOL2: 1MB static                                    + $swp2             |
56# | ePOOL4: 500KB dynamic                                 | iPOOL0            |
57# | ePOOL5: 10MB static                                   | ePOOL6            |
58# | ePOOL6: "infinite" static                             | 200Mbps shaper    |
59# +-------------------------------------------------------|-------------------+
60#                                                         |
61#                                                     +---|-------------------+
62#                                                     |   + $h2            H2 |
63#                                                     |   |                   |
64#                                                     |   + $h2.111           |
65#                                                     |     192.0.2.34/28     |
66#                                                     +-----------------------+
67#
68# iPOOL0+ePOOL4 is a helper pool for control traffic etc.
69# iPOOL1+ePOOL5 are overflow pools.
70# iPOOL2+ePOOL6 are PFC pools.
71
72ALL_TESTS="
73	ping_ipv4
74	test_qos_pfc
75"
76
77lib_dir=$(dirname $0)/../../../net/forwarding
78
79NUM_NETIFS=6
80source $lib_dir/lib.sh
81source $lib_dir/devlink_lib.sh
82
83_1KB=1000
84_100KB=$((100 * _1KB))
85_500KB=$((500 * _1KB))
86_1MB=$((1000 * _1KB))
87_10MB=$((10 * _1MB))
88
89h1_create()
90{
91	simple_if_init $h1
92	mtu_set $h1 10000
93
94	vlan_create $h1 111 v$h1 192.0.2.33/28
95}
96
97h1_destroy()
98{
99	vlan_destroy $h1 111
100
101	mtu_restore $h1
102	simple_if_fini $h1
103}
104
105h2_create()
106{
107	simple_if_init $h2
108	mtu_set $h2 10000
109
110	vlan_create $h2 111 v$h2 192.0.2.34/28
111}
112
113h2_destroy()
114{
115	vlan_destroy $h2 111
116
117	mtu_restore $h2
118	simple_if_fini $h2
119}
120
121switch_create()
122{
123	# pools
124	# -----
125
126	devlink_pool_size_thtype_save 0
127	devlink_pool_size_thtype_save 4
128	devlink_pool_size_thtype_save 1
129	devlink_pool_size_thtype_save 5
130	devlink_pool_size_thtype_save 2
131	devlink_pool_size_thtype_save 6
132
133	devlink_port_pool_th_save $swp1 1
134	devlink_port_pool_th_save $swp2 6
135	devlink_port_pool_th_save $swp3 5
136	devlink_port_pool_th_save $swp4 2
137
138	devlink_tc_bind_pool_th_save $swp1 1 ingress
139	devlink_tc_bind_pool_th_save $swp2 1 egress
140	devlink_tc_bind_pool_th_save $swp3 1 egress
141	devlink_tc_bind_pool_th_save $swp4 1 ingress
142
143	# Control traffic pools. Just reduce the size. Keep them dynamic so that
144	# we don't need to change all the uninteresting quotas.
145	devlink_pool_size_thtype_set 0 dynamic $_500KB
146	devlink_pool_size_thtype_set 4 dynamic $_500KB
147
148	# Overflow pools.
149	devlink_pool_size_thtype_set 1 static $_10MB
150	devlink_pool_size_thtype_set 5 static $_10MB
151
152	# PFC pools. As per the writ, the size of egress PFC pool should be
153	# infinice, but actually it just needs to be large enough to not matter
154	# in practice, so reuse the 10MB limit.
155	devlink_pool_size_thtype_set 2 static $_1MB
156	devlink_pool_size_thtype_set 6 static $_10MB
157
158	# $swp1
159	# -----
160
161	ip link set dev $swp1 up
162	mtu_set $swp1 10000
163	vlan_create $swp1 111
164	ip link set dev $swp1.111 type vlan ingress-qos-map 0:0 1:1
165
166	devlink_port_pool_th_set $swp1 1 $_10MB
167	devlink_tc_bind_pool_th_set $swp1 1 ingress 1 $_10MB
168
169	# Configure qdisc so that we can configure PG and therefore pool
170	# assignment.
171	tc qdisc replace dev $swp1 root handle 1: \
172	   ets bands 8 strict 8 priomap 7 6
173	dcb buffer set dev $swp1 prio-buffer all:0 1:1
174
175	# $swp2
176	# -----
177
178	ip link set dev $swp2 up
179	mtu_set $swp2 10000
180	vlan_create $swp2 111
181	ip link set dev $swp2.111 type vlan egress-qos-map 0:0 1:1
182
183	devlink_port_pool_th_set $swp2 6 $_10MB
184	devlink_tc_bind_pool_th_set $swp2 1 egress 6 $_10MB
185
186	# prio 0->TC0 (band 7), 1->TC1 (band 6). TC1 is shaped.
187	tc qdisc replace dev $swp2 root handle 1: \
188	   ets bands 8 strict 8 priomap 7 6
189	tc qdisc replace dev $swp2 parent 1:7 handle 17: \
190	   tbf rate 200Mbit burst 131072 limit 1M
191
192	# $swp3
193	# -----
194
195	ip link set dev $swp3 up
196	mtu_set $swp3 10000
197	vlan_create $swp3 111
198	ip link set dev $swp3.111 type vlan egress-qos-map 0:0 1:1
199
200	devlink_port_pool_th_set $swp3 5 $_10MB
201	devlink_tc_bind_pool_th_set $swp3 1 egress 5 $_10MB
202
203	# prio 0->TC0 (band 7), 1->TC1 (band 6)
204	tc qdisc replace dev $swp3 root handle 1: \
205	   ets bands 8 strict 8 priomap 7 6
206
207	# Need to enable PFC so that PAUSE takes effect. Therefore need to put
208	# the lossless prio into a buffer of its own. Don't bother with buffer
209	# sizes though, there is not going to be any pressure in the "backward"
210	# direction.
211	dcb buffer set dev $swp3 prio-buffer all:0 1:1
212	dcb pfc set dev $swp3 prio-pfc all:off 1:on
213
214	# $swp4
215	# -----
216
217	ip link set dev $swp4 up
218	mtu_set $swp4 10000
219	vlan_create $swp4 111
220	ip link set dev $swp4.111 type vlan ingress-qos-map 0:0 1:1
221
222	devlink_port_pool_th_set $swp4 2 $_1MB
223	devlink_tc_bind_pool_th_set $swp4 1 ingress 2 $_1MB
224
225	# Configure qdisc so that we can hand-tune headroom.
226	tc qdisc replace dev $swp4 root handle 1: \
227	   ets bands 8 strict 8 priomap 7 6
228	dcb buffer set dev $swp4 prio-buffer all:0 1:1
229	dcb pfc set dev $swp4 prio-pfc all:off 1:on
230	# PG0 will get autoconfigured to Xoff, give PG1 arbitrarily 100K, which
231	# is (-2*MTU) about 80K of delay provision.
232	dcb buffer set dev $swp4 buffer-size all:0 1:$_100KB
233
234	# bridges
235	# -------
236
237	ip link add name br1 type bridge vlan_filtering 0
238	ip link set dev $swp1.111 master br1
239	ip link set dev $swp3.111 master br1
240	ip link set dev br1 up
241
242	ip link add name br2 type bridge vlan_filtering 0
243	ip link set dev $swp2.111 master br2
244	ip link set dev $swp4.111 master br2
245	ip link set dev br2 up
246}
247
248switch_destroy()
249{
250	# Do this first so that we can reset the limits to values that are only
251	# valid for the original static / dynamic setting.
252	devlink_pool_size_thtype_restore 6
253	devlink_pool_size_thtype_restore 5
254	devlink_pool_size_thtype_restore 4
255	devlink_pool_size_thtype_restore 2
256	devlink_pool_size_thtype_restore 1
257	devlink_pool_size_thtype_restore 0
258
259	# bridges
260	# -------
261
262	ip link set dev br2 down
263	ip link set dev $swp4.111 nomaster
264	ip link set dev $swp2.111 nomaster
265	ip link del dev br2
266
267	ip link set dev br1 down
268	ip link set dev $swp3.111 nomaster
269	ip link set dev $swp1.111 nomaster
270	ip link del dev br1
271
272	# $swp4
273	# -----
274
275	dcb buffer set dev $swp4 buffer-size all:0
276	dcb pfc set dev $swp4 prio-pfc all:off
277	dcb buffer set dev $swp4 prio-buffer all:0
278	tc qdisc del dev $swp4 root
279
280	devlink_tc_bind_pool_th_restore $swp4 1 ingress
281	devlink_port_pool_th_restore $swp4 2
282
283	vlan_destroy $swp4 111
284	mtu_restore $swp4
285	ip link set dev $swp4 down
286
287	# $swp3
288	# -----
289
290	dcb pfc set dev $swp3 prio-pfc all:off
291	dcb buffer set dev $swp3 prio-buffer all:0
292	tc qdisc del dev $swp3 root
293
294	devlink_tc_bind_pool_th_restore $swp3 1 egress
295	devlink_port_pool_th_restore $swp3 5
296
297	vlan_destroy $swp3 111
298	mtu_restore $swp3
299	ip link set dev $swp3 down
300
301	# $swp2
302	# -----
303
304	tc qdisc del dev $swp2 parent 1:7
305	tc qdisc del dev $swp2 root
306
307	devlink_tc_bind_pool_th_restore $swp2 1 egress
308	devlink_port_pool_th_restore $swp2 6
309
310	vlan_destroy $swp2 111
311	mtu_restore $swp2
312	ip link set dev $swp2 down
313
314	# $swp1
315	# -----
316
317	dcb buffer set dev $swp1 prio-buffer all:0
318	tc qdisc del dev $swp1 root
319
320	devlink_tc_bind_pool_th_restore $swp1 1 ingress
321	devlink_port_pool_th_restore $swp1 1
322
323	vlan_destroy $swp1 111
324	mtu_restore $swp1
325	ip link set dev $swp1 down
326}
327
328setup_prepare()
329{
330	h1=${NETIFS[p1]}
331	swp1=${NETIFS[p2]}
332
333	swp2=${NETIFS[p3]}
334	h2=${NETIFS[p4]}
335
336	swp3=${NETIFS[p5]}
337	swp4=${NETIFS[p6]}
338
339	h2mac=$(mac_get $h2)
340
341	vrf_prepare
342
343	h1_create
344	h2_create
345	switch_create
346}
347
348cleanup()
349{
350	pre_cleanup
351
352	switch_destroy
353	h2_destroy
354	h1_destroy
355
356	vrf_cleanup
357}
358
359ping_ipv4()
360{
361	ping_test $h1 192.0.2.34
362}
363
364test_qos_pfc()
365{
366	RET=0
367
368	# 10M pool, each packet is 8K of payload + headers
369	local pkts=$((_10MB / 8050))
370	local size=$((pkts * 8050))
371	local in0=$(ethtool_stats_get $swp1 rx_octets_prio_1)
372	local out0=$(ethtool_stats_get $swp2 tx_octets_prio_1)
373
374	$MZ $h1 -p 8000 -Q 1:111 -A 192.0.2.33 -B 192.0.2.34 \
375		-a own -b $h2mac -c $pkts -t udp -q
376	sleep 2
377
378	local in1=$(ethtool_stats_get $swp1 rx_octets_prio_1)
379	local out1=$(ethtool_stats_get $swp2 tx_octets_prio_1)
380
381	local din=$((in1 - in0))
382	local dout=$((out1 - out0))
383
384	local pct_in=$((din * 100 / size))
385
386	((pct_in > 95 && pct_in < 105))
387	check_err $? "Relative ingress out of expected bounds, $pct_in% should be 100%"
388
389	((dout == din))
390	check_err $? "$((din - dout)) bytes out of $din ingressed got lost"
391
392	log_test "PFC"
393}
394
395bail_on_lldpad "configure DCB" "configure Qdiscs"
396
397trap cleanup EXIT
398setup_prepare
399setup_wait
400tests_run
401
402exit $EXIT_STATUS
403