tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
selftests: net: Add busy_poll_test
Add an epoll busy poll test using netdevsim.
This test is comprised of:
- busy_poller (via busy_poller.c)
- busy_poll_test.sh which loads netdevsim, sets up network namespaces,
and runs busy_poller to receive data and socat to send data.
The selftest tests two different scenarios:
- busy poll (the pre-existing version in the kernel)
- busy poll with suspend enabled (what this series adds)
The data transmit is a 1MiB temporary file generated from /dev/urandom
and the test is considered passing if the md5sum of the input file to
socat matches the md5sum of the output file from busy_poller.
netdevsim was chosen instead of veth due to netdevsim's support for
netdev-genl.
For now, this test uses the functionality that netdevsim provides. In the
future, perhaps netdevsim can be extended to emulate device IRQs to more
thoroughly test all pre-existing kernel options (like defer_hard_irqs)
and suspend.
Signed-off-by: Joe Damato <jdamato@fastly.com>
Co-developed-by: Martin Karsten <mkarsten@uwaterloo.ca>
Signed-off-by: Martin Karsten <mkarsten@uwaterloo.ca>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20241109050245.191288-6-jdamato@fastly.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
authored by
Joe Damato
and committed by
Jakub Kicinski
347fcdc4
8a6de262
+521
+1
tools/testing/selftests/net/.gitignore
+1
tools/testing/selftests/net/.gitignore
+9
tools/testing/selftests/net/Makefile
+9
tools/testing/selftests/net/Makefile
···
97
97
TEST_PROGS += vlan_hw_filter.sh
98
98
TEST_PROGS += bpf_offload.py
99
99
TEST_PROGS += ipv6_route_update_soft_lockup.sh
100
+
TEST_PROGS += busy_poll_test.sh
101
+
102
+
# YNL files, must be before "include ..lib.mk"
103
+
YNL_GEN_FILES := busy_poller
104
+
TEST_GEN_FILES += $(YNL_GEN_FILES)
100
105
101
106
TEST_FILES := settings
102
107
TEST_FILES += in_netns.sh lib.sh net_helper.sh setup_loopback.sh setup_veth.sh
···
111
106
TEST_INCLUDES := forwarding/lib.sh
112
107
113
108
include ../lib.mk
109
+
110
+
# YNL build
111
+
YNL_GENS := netdev
112
+
include ynl.mk
114
113
115
114
$(OUTPUT)/epoll_busy_poll: LDLIBS += -lcap
116
115
$(OUTPUT)/reuseport_bpf_numa: LDLIBS += -lnuma
+165
tools/testing/selftests/net/busy_poll_test.sh
+165
tools/testing/selftests/net/busy_poll_test.sh
···
1
+
#!/bin/bash
2
+
# SPDX-License-Identifier: GPL-2.0
3
+
source net_helper.sh
4
+
5
+
NSIM_SV_ID=$((256 + RANDOM % 256))
6
+
NSIM_SV_SYS=/sys/bus/netdevsim/devices/netdevsim$NSIM_SV_ID
7
+
NSIM_CL_ID=$((512 + RANDOM % 256))
8
+
NSIM_CL_SYS=/sys/bus/netdevsim/devices/netdevsim$NSIM_CL_ID
9
+
10
+
NSIM_DEV_SYS_NEW=/sys/bus/netdevsim/new_device
11
+
NSIM_DEV_SYS_DEL=/sys/bus/netdevsim/del_device
12
+
NSIM_DEV_SYS_LINK=/sys/bus/netdevsim/link_device
13
+
NSIM_DEV_SYS_UNLINK=/sys/bus/netdevsim/unlink_device
14
+
15
+
SERVER_IP=192.168.1.1
16
+
CLIENT_IP=192.168.1.2
17
+
SERVER_PORT=48675
18
+
19
+
# busy poll config
20
+
MAX_EVENTS=8
21
+
BUSY_POLL_USECS=0
22
+
BUSY_POLL_BUDGET=16
23
+
PREFER_BUSY_POLL=1
24
+
25
+
# IRQ deferral config
26
+
NAPI_DEFER_HARD_IRQS=100
27
+
GRO_FLUSH_TIMEOUT=50000
28
+
SUSPEND_TIMEOUT=20000000
29
+
30
+
setup_ns()
31
+
{
32
+
set -e
33
+
ip netns add nssv
34
+
ip netns add nscl
35
+
36
+
NSIM_SV_NAME=$(find $NSIM_SV_SYS/net -maxdepth 1 -type d ! \
37
+
-path $NSIM_SV_SYS/net -exec basename {} \;)
38
+
NSIM_CL_NAME=$(find $NSIM_CL_SYS/net -maxdepth 1 -type d ! \
39
+
-path $NSIM_CL_SYS/net -exec basename {} \;)
40
+
41
+
# ensure the server has 1 queue
42
+
ethtool -L $NSIM_SV_NAME combined 1 2>/dev/null
43
+
44
+
ip link set $NSIM_SV_NAME netns nssv
45
+
ip link set $NSIM_CL_NAME netns nscl
46
+
47
+
ip netns exec nssv ip addr add "${SERVER_IP}/24" dev $NSIM_SV_NAME
48
+
ip netns exec nscl ip addr add "${CLIENT_IP}/24" dev $NSIM_CL_NAME
49
+
50
+
ip netns exec nssv ip link set dev $NSIM_SV_NAME up
51
+
ip netns exec nscl ip link set dev $NSIM_CL_NAME up
52
+
53
+
set +e
54
+
}
55
+
56
+
cleanup_ns()
57
+
{
58
+
ip netns del nscl
59
+
ip netns del nssv
60
+
}
61
+
62
+
test_busypoll()
63
+
{
64
+
suspend_value=${1:-0}
65
+
tmp_file=$(mktemp)
66
+
out_file=$(mktemp)
67
+
68
+
# fill a test file with random data
69
+
dd if=/dev/urandom of=${tmp_file} bs=1M count=1 2> /dev/null
70
+
71
+
timeout -k 1s 30s ip netns exec nssv ./busy_poller \
72
+
-p${SERVER_PORT} \
73
+
-b${SERVER_IP} \
74
+
-m${MAX_EVENTS} \
75
+
-u${BUSY_POLL_USECS} \
76
+
-P${PREFER_BUSY_POLL} \
77
+
-g${BUSY_POLL_BUDGET} \
78
+
-i${NSIM_SV_IFIDX} \
79
+
-s${suspend_value} \
80
+
-o${out_file}&
81
+
82
+
wait_local_port_listen nssv ${SERVER_PORT} tcp
83
+
84
+
ip netns exec nscl socat -u $tmp_file TCP:${SERVER_IP}:${SERVER_PORT}
85
+
86
+
wait
87
+
88
+
tmp_file_md5sum=$(md5sum $tmp_file | cut -f1 -d' ')
89
+
out_file_md5sum=$(md5sum $out_file | cut -f1 -d' ')
90
+
91
+
if [ "$tmp_file_md5sum" = "$out_file_md5sum" ]; then
92
+
res=0
93
+
else
94
+
echo "md5sum mismatch"
95
+
echo "input file md5sum: ${tmp_file_md5sum}";
96
+
echo "output file md5sum: ${out_file_md5sum}";
97
+
res=1
98
+
fi
99
+
100
+
rm $out_file $tmp_file
101
+
102
+
return $res
103
+
}
104
+
105
+
test_busypoll_with_suspend()
106
+
{
107
+
test_busypoll ${SUSPEND_TIMEOUT}
108
+
109
+
return $?
110
+
}
111
+
112
+
###
113
+
### Code start
114
+
###
115
+
116
+
modprobe netdevsim
117
+
118
+
# linking
119
+
120
+
echo $NSIM_SV_ID > $NSIM_DEV_SYS_NEW
121
+
echo $NSIM_CL_ID > $NSIM_DEV_SYS_NEW
122
+
udevadm settle
123
+
124
+
setup_ns
125
+
126
+
NSIM_SV_FD=$((256 + RANDOM % 256))
127
+
exec {NSIM_SV_FD}</var/run/netns/nssv
128
+
NSIM_SV_IFIDX=$(ip netns exec nssv cat /sys/class/net/$NSIM_SV_NAME/ifindex)
129
+
130
+
NSIM_CL_FD=$((256 + RANDOM % 256))
131
+
exec {NSIM_CL_FD}</var/run/netns/nscl
132
+
NSIM_CL_IFIDX=$(ip netns exec nscl cat /sys/class/net/$NSIM_CL_NAME/ifindex)
133
+
134
+
echo "$NSIM_SV_FD:$NSIM_SV_IFIDX $NSIM_CL_FD:$NSIM_CL_IFIDX" > \
135
+
$NSIM_DEV_SYS_LINK
136
+
137
+
if [ $? -ne 0 ]; then
138
+
echo "linking netdevsim1 with netdevsim2 should succeed"
139
+
cleanup_ns
140
+
exit 1
141
+
fi
142
+
143
+
test_busypoll
144
+
if [ $? -ne 0 ]; then
145
+
echo "test_busypoll failed"
146
+
cleanup_ns
147
+
exit 1
148
+
fi
149
+
150
+
test_busypoll_with_suspend
151
+
if [ $? -ne 0 ]; then
152
+
echo "test_busypoll_with_suspend failed"
153
+
cleanup_ns
154
+
exit 1
155
+
fi
156
+
157
+
echo "$NSIM_SV_FD:$NSIM_SV_IFIDX" > $NSIM_DEV_SYS_UNLINK
158
+
159
+
echo $NSIM_CL_ID > $NSIM_DEV_SYS_DEL
160
+
161
+
cleanup_ns
162
+
163
+
modprobe -r netdevsim
164
+
165
+
exit 0
+346
tools/testing/selftests/net/busy_poller.c
+346
tools/testing/selftests/net/busy_poller.c
···
1
+
// SPDX-License-Identifier: GPL-2.0
2
+
#include <assert.h>
3
+
#include <errno.h>
4
+
#include <error.h>
5
+
#include <fcntl.h>
6
+
#include <inttypes.h>
7
+
#include <limits.h>
8
+
#include <stdlib.h>
9
+
#include <stdio.h>
10
+
#include <string.h>
11
+
#include <unistd.h>
12
+
#include <ynl.h>
13
+
14
+
#include <arpa/inet.h>
15
+
#include <netinet/in.h>
16
+
17
+
#include <sys/epoll.h>
18
+
#include <sys/ioctl.h>
19
+
#include <sys/socket.h>
20
+
#include <sys/types.h>
21
+
22
+
#include <linux/genetlink.h>
23
+
#include <linux/netlink.h>
24
+
25
+
#include "netdev-user.h"
26
+
27
+
/* The below ifdef blob is required because:
28
+
*
29
+
* - sys/epoll.h does not (yet) have the ioctl definitions included. So,
30
+
* systems with older glibcs will not have them available. However,
31
+
* sys/epoll.h does include the type definition for epoll_data, which is
32
+
* needed by the user program (e.g. epoll_event.data.fd)
33
+
*
34
+
* - linux/eventpoll.h does not define the epoll_data type, it is simply an
35
+
* opaque __u64. It does, however, include the ioctl definition.
36
+
*
37
+
* Including both headers is impossible (types would be redefined), so I've
38
+
* opted instead to take sys/epoll.h, and include the blob below.
39
+
*
40
+
* Someday, when glibc is globally up to date, the blob below can be removed.
41
+
*/
42
+
#if !defined(EPOLL_IOC_TYPE)
43
+
struct epoll_params {
44
+
uint32_t busy_poll_usecs;
45
+
uint16_t busy_poll_budget;
46
+
uint8_t prefer_busy_poll;
47
+
48
+
/* pad the struct to a multiple of 64bits */
49
+
uint8_t __pad;
50
+
};
51
+
52
+
#define EPOLL_IOC_TYPE 0x8A
53
+
#define EPIOCSPARAMS _IOW(EPOLL_IOC_TYPE, 0x01, struct epoll_params)
54
+
#define EPIOCGPARAMS _IOR(EPOLL_IOC_TYPE, 0x02, struct epoll_params)
55
+
#endif
56
+
57
+
static uint32_t cfg_port = 8000;
58
+
static struct in_addr cfg_bind_addr = { .s_addr = INADDR_ANY };
59
+
static char *cfg_outfile;
60
+
static int cfg_max_events = 8;
61
+
static int cfg_ifindex;
62
+
63
+
/* busy poll params */
64
+
static uint32_t cfg_busy_poll_usecs;
65
+
static uint32_t cfg_busy_poll_budget;
66
+
static uint32_t cfg_prefer_busy_poll;
67
+
68
+
/* IRQ params */
69
+
static uint32_t cfg_defer_hard_irqs;
70
+
static uint64_t cfg_gro_flush_timeout;
71
+
static uint64_t cfg_irq_suspend_timeout;
72
+
73
+
static void usage(const char *filepath)
74
+
{
75
+
error(1, 0,
76
+
"Usage: %s -p<port> -b<addr> -m<max_events> -u<busy_poll_usecs> -P<prefer_busy_poll> -g<busy_poll_budget> -o<outfile> -d<defer_hard_irqs> -r<gro_flush_timeout> -s<irq_suspend_timeout> -i<ifindex>",
77
+
filepath);
78
+
}
79
+
80
+
static void parse_opts(int argc, char **argv)
81
+
{
82
+
int ret;
83
+
int c;
84
+
85
+
if (argc <= 1)
86
+
usage(argv[0]);
87
+
88
+
while ((c = getopt(argc, argv, "p:m:b:u:P:g:o:d:r:s:i:")) != -1) {
89
+
switch (c) {
90
+
case 'u':
91
+
cfg_busy_poll_usecs = strtoul(optarg, NULL, 0);
92
+
if (cfg_busy_poll_usecs == ULONG_MAX ||
93
+
cfg_busy_poll_usecs > UINT32_MAX)
94
+
error(1, ERANGE, "busy_poll_usecs too large");
95
+
break;
96
+
case 'P':
97
+
cfg_prefer_busy_poll = strtoul(optarg, NULL, 0);
98
+
if (cfg_prefer_busy_poll == ULONG_MAX ||
99
+
cfg_prefer_busy_poll > 1)
100
+
error(1, ERANGE,
101
+
"prefer busy poll should be 0 or 1");
102
+
break;
103
+
case 'g':
104
+
cfg_busy_poll_budget = strtoul(optarg, NULL, 0);
105
+
if (cfg_busy_poll_budget == ULONG_MAX ||
106
+
cfg_busy_poll_budget > UINT16_MAX)
107
+
error(1, ERANGE,
108
+
"busy poll budget must be [0, UINT16_MAX]");
109
+
break;
110
+
case 'p':
111
+
cfg_port = strtoul(optarg, NULL, 0);
112
+
if (cfg_port > UINT16_MAX)
113
+
error(1, ERANGE, "port must be <= 65535");
114
+
break;
115
+
case 'b':
116
+
ret = inet_aton(optarg, &cfg_bind_addr);
117
+
if (ret == 0)
118
+
error(1, errno,
119
+
"bind address %s invalid", optarg);
120
+
break;
121
+
case 'o':
122
+
cfg_outfile = strdup(optarg);
123
+
if (!cfg_outfile)
124
+
error(1, 0, "outfile invalid");
125
+
break;
126
+
case 'm':
127
+
cfg_max_events = strtol(optarg, NULL, 0);
128
+
129
+
if (cfg_max_events == LONG_MIN ||
130
+
cfg_max_events == LONG_MAX ||
131
+
cfg_max_events <= 0)
132
+
error(1, ERANGE,
133
+
"max events must be > 0 and < LONG_MAX");
134
+
break;
135
+
case 'd':
136
+
cfg_defer_hard_irqs = strtoul(optarg, NULL, 0);
137
+
138
+
if (cfg_defer_hard_irqs == ULONG_MAX ||
139
+
cfg_defer_hard_irqs > INT32_MAX)
140
+
error(1, ERANGE,
141
+
"defer_hard_irqs must be <= INT32_MAX");
142
+
break;
143
+
case 'r':
144
+
cfg_gro_flush_timeout = strtoull(optarg, NULL, 0);
145
+
146
+
if (cfg_gro_flush_timeout == ULLONG_MAX)
147
+
error(1, ERANGE,
148
+
"gro_flush_timeout must be < ULLONG_MAX");
149
+
break;
150
+
case 's':
151
+
cfg_irq_suspend_timeout = strtoull(optarg, NULL, 0);
152
+
153
+
if (cfg_irq_suspend_timeout == ULLONG_MAX)
154
+
error(1, ERANGE,
155
+
"irq_suspend_timeout must be < ULLONG_MAX");
156
+
break;
157
+
case 'i':
158
+
cfg_ifindex = strtoul(optarg, NULL, 0);
159
+
if (cfg_ifindex == ULONG_MAX)
160
+
error(1, ERANGE,
161
+
"ifindex must be < ULONG_MAX");
162
+
break;
163
+
}
164
+
}
165
+
166
+
if (!cfg_ifindex)
167
+
usage(argv[0]);
168
+
169
+
if (optind != argc)
170
+
usage(argv[0]);
171
+
}
172
+
173
+
static void epoll_ctl_add(int epfd, int fd, uint32_t events)
174
+
{
175
+
struct epoll_event ev;
176
+
177
+
ev.events = events;
178
+
ev.data.fd = fd;
179
+
if (epoll_ctl(epfd, EPOLL_CTL_ADD, fd, &ev) == -1)
180
+
error(1, errno, "epoll_ctl add fd: %d", fd);
181
+
}
182
+
183
+
static void setnonblock(int sockfd)
184
+
{
185
+
int flags;
186
+
187
+
flags = fcntl(sockfd, F_GETFL, 0);
188
+
189
+
if (fcntl(sockfd, F_SETFL, flags | O_NONBLOCK) == -1)
190
+
error(1, errno, "unable to set socket to nonblocking mode");
191
+
}
192
+
193
+
static void write_chunk(int fd, char *buf, ssize_t buflen)
194
+
{
195
+
ssize_t remaining = buflen;
196
+
char *buf_offset = buf;
197
+
ssize_t writelen = 0;
198
+
ssize_t write_result;
199
+
200
+
while (writelen < buflen) {
201
+
write_result = write(fd, buf_offset, remaining);
202
+
if (write_result == -1)
203
+
error(1, errno, "unable to write data to outfile");
204
+
205
+
writelen += write_result;
206
+
remaining -= write_result;
207
+
buf_offset += write_result;
208
+
}
209
+
}
210
+
211
+
static void setup_queue(void)
212
+
{
213
+
struct netdev_napi_get_list *napi_list = NULL;
214
+
struct netdev_napi_get_req_dump *req = NULL;
215
+
struct netdev_napi_set_req *set_req = NULL;
216
+
struct ynl_sock *ys;
217
+
struct ynl_error yerr;
218
+
uint32_t napi_id;
219
+
220
+
ys = ynl_sock_create(&ynl_netdev_family, &yerr);
221
+
if (!ys)
222
+
error(1, 0, "YNL: %s", yerr.msg);
223
+
224
+
req = netdev_napi_get_req_dump_alloc();
225
+
netdev_napi_get_req_dump_set_ifindex(req, cfg_ifindex);
226
+
napi_list = netdev_napi_get_dump(ys, req);
227
+
228
+
/* assume there is 1 NAPI configured and take the first */
229
+
if (napi_list->obj._present.id)
230
+
napi_id = napi_list->obj.id;
231
+
else
232
+
error(1, 0, "napi ID not present?");
233
+
234
+
set_req = netdev_napi_set_req_alloc();
235
+
netdev_napi_set_req_set_id(set_req, napi_id);
236
+
netdev_napi_set_req_set_defer_hard_irqs(set_req, cfg_defer_hard_irqs);
237
+
netdev_napi_set_req_set_gro_flush_timeout(set_req,
238
+
cfg_gro_flush_timeout);
239
+
netdev_napi_set_req_set_irq_suspend_timeout(set_req,
240
+
cfg_irq_suspend_timeout);
241
+
242
+
if (netdev_napi_set(ys, set_req))
243
+
error(1, 0, "can't set NAPI params: %s\n", yerr.msg);
244
+
245
+
netdev_napi_get_list_free(napi_list);
246
+
netdev_napi_get_req_dump_free(req);
247
+
netdev_napi_set_req_free(set_req);
248
+
ynl_sock_destroy(ys);
249
+
}
250
+
251
+
static void run_poller(void)
252
+
{
253
+
struct epoll_event events[cfg_max_events];
254
+
struct epoll_params epoll_params = {0};
255
+
struct sockaddr_in server_addr;
256
+
int i, epfd, nfds;
257
+
ssize_t readlen;
258
+
int outfile_fd;
259
+
char buf[1024];
260
+
int sockfd;
261
+
int conn;
262
+
int val;
263
+
264
+
outfile_fd = open(cfg_outfile, O_WRONLY | O_CREAT, 0644);
265
+
if (outfile_fd == -1)
266
+
error(1, errno, "unable to open outfile: %s", cfg_outfile);
267
+
268
+
sockfd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
269
+
if (sockfd == -1)
270
+
error(1, errno, "unable to create listen socket");
271
+
272
+
server_addr.sin_family = AF_INET;
273
+
server_addr.sin_port = htons(cfg_port);
274
+
server_addr.sin_addr = cfg_bind_addr;
275
+
276
+
/* these values are range checked during parse_opts, so casting is safe
277
+
* here
278
+
*/
279
+
epoll_params.busy_poll_usecs = cfg_busy_poll_usecs;
280
+
epoll_params.busy_poll_budget = (uint16_t)cfg_busy_poll_budget;
281
+
epoll_params.prefer_busy_poll = (uint8_t)cfg_prefer_busy_poll;
282
+
epoll_params.__pad = 0;
283
+
284
+
val = 1;
285
+
if (setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof(val)))
286
+
error(1, errno, "poller setsockopt reuseaddr");
287
+
288
+
setnonblock(sockfd);
289
+
290
+
if (bind(sockfd, (struct sockaddr *)&server_addr,
291
+
sizeof(struct sockaddr_in)))
292
+
error(0, errno, "poller bind to port: %d\n", cfg_port);
293
+
294
+
if (listen(sockfd, 1))
295
+
error(1, errno, "poller listen");
296
+
297
+
epfd = epoll_create1(0);
298
+
if (ioctl(epfd, EPIOCSPARAMS, &epoll_params) == -1)
299
+
error(1, errno, "unable to set busy poll params");
300
+
301
+
epoll_ctl_add(epfd, sockfd, EPOLLIN | EPOLLOUT | EPOLLET);
302
+
303
+
for (;;) {
304
+
nfds = epoll_wait(epfd, events, cfg_max_events, -1);
305
+
for (i = 0; i < nfds; i++) {
306
+
if (events[i].data.fd == sockfd) {
307
+
conn = accept(sockfd, NULL, NULL);
308
+
if (conn == -1)
309
+
error(1, errno,
310
+
"accepting incoming connection failed");
311
+
312
+
setnonblock(conn);
313
+
epoll_ctl_add(epfd, conn,
314
+
EPOLLIN | EPOLLET | EPOLLRDHUP |
315
+
EPOLLHUP);
316
+
} else if (events[i].events & EPOLLIN) {
317
+
for (;;) {
318
+
readlen = read(events[i].data.fd, buf,
319
+
sizeof(buf));
320
+
if (readlen > 0)
321
+
write_chunk(outfile_fd, buf,
322
+
readlen);
323
+
else
324
+
break;
325
+
}
326
+
} else {
327
+
/* spurious event ? */
328
+
}
329
+
if (events[i].events & (EPOLLRDHUP | EPOLLHUP)) {
330
+
epoll_ctl(epfd, EPOLL_CTL_DEL,
331
+
events[i].data.fd, NULL);
332
+
close(events[i].data.fd);
333
+
close(outfile_fd);
334
+
return;
335
+
}
336
+
}
337
+
}
338
+
}
339
+
340
+
int main(int argc, char *argv[])
341
+
{
342
+
parse_opts(argc, argv);
343
+
setup_queue();
344
+
run_poller();
345
+
return 0;
346
+
}