Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
selftests: drv-net: test drivers sleeping in ndo_get_stats64
Most of our tests use rtnetlink to read device stats, so they
don't expose the drivers much to paths in which device stats
are read under RCU. Add tests which hammer profcs reads to
make sure drivers:
- don't sleep while reporting stats,
- can handle parallel reads,
- can handle device going down while reading.
Set ifname on the env class in NetDrvEnv, we already do that
in NetDrvEpEnv.
KTAP version 1
1..7
ok 1 stats.check_pause
ok 2 stats.check_fec
ok 3 stats.pkt_byte_sum
ok 4 stats.qstat_by_ifindex
ok 5 stats.check_down
ok 6 stats.procfs_hammer
# completed up/down cycles: 6
ok 7 stats.procfs_downup_hammer
# Totals: pass:7 fail:0 xfail:0 xpass:0 skip:0 error:0
Reviewed-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250107022932.2087744-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
+97
-3
+1
tools/testing/selftests/drivers/net/lib/py/env.py
+1
tools/testing/selftests/drivers/net/lib/py/env.py
+91
-3
tools/testing/selftests/drivers/net/stats.py
+91
-3
tools/testing/selftests/drivers/net/stats.py
···
2
2
# SPDX-License-Identifier: GPL-2.0
3
3
4
4
import errno
5
+
import subprocess
6
+
import time
5
7
from lib.py import ksft_run, ksft_exit, ksft_pr
6
-
from lib.py import ksft_ge, ksft_eq, ksft_in, ksft_true, ksft_raises, KsftSkipEx, KsftXfailEx
8
+
from lib.py import ksft_ge, ksft_eq, ksft_is, ksft_in, ksft_lt, ksft_true, ksft_raises
9
+
from lib.py import KsftSkipEx, KsftXfailEx
7
10
from lib.py import ksft_disruptive
8
11
from lib.py import EthtoolFamily, NetdevFamily, RtnlFamily, NlError
9
12
from lib.py import NetDrvEnv
10
-
from lib.py import ip, defer
13
+
from lib.py import cmd, ip, defer
11
14
12
15
ethnl = EthtoolFamily()
13
16
netfam = NetdevFamily()
···
177
174
netfam.qstats_get({"ifindex": cfg.ifindex, "scope": "queue"}, dump=True)
178
175
179
176
177
+
def __run_inf_loop(body):
178
+
body = body.strip()
179
+
if body[-1] != ';':
180
+
body += ';'
181
+
182
+
return subprocess.Popen(f"while true; do {body} done", shell=True,
183
+
stdout=subprocess.PIPE, stderr=subprocess.PIPE)
184
+
185
+
186
+
def __stats_increase_sanely(old, new) -> None:
187
+
for k in old.keys():
188
+
ksft_ge(new[k], old[k])
189
+
ksft_lt(new[k] - old[k], 1 << 31, comment="likely wrapping error")
190
+
191
+
192
+
def procfs_hammer(cfg) -> None:
193
+
"""
194
+
Reading stats via procfs only holds the RCU lock, which is not an exclusive
195
+
lock, make sure drivers can handle parallel reads of stats.
196
+
"""
197
+
one = __run_inf_loop("cat /proc/net/dev")
198
+
defer(one.kill)
199
+
two = __run_inf_loop("cat /proc/net/dev")
200
+
defer(two.kill)
201
+
202
+
time.sleep(1)
203
+
# Make sure the processes are running
204
+
ksft_is(one.poll(), None)
205
+
ksft_is(two.poll(), None)
206
+
207
+
rtstat1 = rtnl.getlink({"ifi-index": cfg.ifindex})['stats64']
208
+
time.sleep(2)
209
+
rtstat2 = rtnl.getlink({"ifi-index": cfg.ifindex})['stats64']
210
+
__stats_increase_sanely(rtstat1, rtstat2)
211
+
# defers will kill the loops
212
+
213
+
214
+
@ksft_disruptive
215
+
def procfs_downup_hammer(cfg) -> None:
216
+
"""
217
+
Reading stats via procfs only holds the RCU lock, drivers often try
218
+
to sleep when reading the stats, or don't protect against races.
219
+
"""
220
+
# Max out the queues, we'll flip between max and 1
221
+
channels = ethnl.channels_get({'header': {'dev-index': cfg.ifindex}})
222
+
if channels['combined-count'] == 0:
223
+
rx_type = 'rx'
224
+
else:
225
+
rx_type = 'combined'
226
+
cur_queue_cnt = channels[f'{rx_type}-count']
227
+
max_queue_cnt = channels[f'{rx_type}-max']
228
+
229
+
cmd(f"ethtool -L {cfg.ifname} {rx_type} {max_queue_cnt}")
230
+
defer(cmd, f"ethtool -L {cfg.ifname} {rx_type} {cur_queue_cnt}")
231
+
232
+
# Real test stats
233
+
stats = __run_inf_loop("cat /proc/net/dev")
234
+
defer(stats.kill)
235
+
236
+
ipset = f"ip link set dev {cfg.ifname}"
237
+
defer(ip, f"link set dev {cfg.ifname} up")
238
+
# The "echo -n 1" lets us count iterations below
239
+
updown = f"{ipset} down; sleep 0.05; {ipset} up; sleep 0.05; " + \
240
+
f"ethtool -L {cfg.ifname} {rx_type} 1; " + \
241
+
f"ethtool -L {cfg.ifname} {rx_type} {max_queue_cnt}; " + \
242
+
"echo -n 1"
243
+
updown = __run_inf_loop(updown)
244
+
kill_updown = defer(updown.kill)
245
+
246
+
time.sleep(1)
247
+
# Make sure the processes are running
248
+
ksft_is(stats.poll(), None)
249
+
ksft_is(updown.poll(), None)
250
+
251
+
rtstat1 = rtnl.getlink({"ifi-index": cfg.ifindex})['stats64']
252
+
# We're looking for crashes, give it extra time
253
+
time.sleep(9)
254
+
rtstat2 = rtnl.getlink({"ifi-index": cfg.ifindex})['stats64']
255
+
__stats_increase_sanely(rtstat1, rtstat2)
256
+
257
+
kill_updown.exec()
258
+
stdout, _ = updown.communicate(timeout=5)
259
+
ksft_pr("completed up/down cycles:", len(stdout.decode('utf-8')))
260
+
261
+
180
262
def main() -> None:
181
263
with NetDrvEnv(__file__, queue_count=100) as cfg:
182
264
ksft_run([check_pause, check_fec, pkt_byte_sum, qstat_by_ifindex,
183
-
check_down],
265
+
check_down, procfs_hammer, procfs_downup_hammer],
184
266
args=(cfg, ))
185
267
ksft_exit()
186
268
+5
tools/testing/selftests/net/lib/py/ksft.py
+5
tools/testing/selftests/net/lib/py/ksft.py