Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Add support to set NAPI threaded for individual NAPI
A net device has a threaded sysctl that can be used to enable threaded
NAPI polling on all of the NAPI contexts under that device. Allow
enabling threaded NAPI polling at individual NAPI level using netlink.
Extend the netlink operation `napi-set` and allow setting the threaded
attribute of a NAPI. This will enable the threaded polling on a NAPI
context.
Add a test in `nl_netdev.py` that verifies various cases of threaded
NAPI being set at NAPI and at device level.
Tested
./tools/testing/selftests/net/nl_netdev.py
TAP version 13
1..7
ok 1 nl_netdev.empty_check
ok 2 nl_netdev.lo_check
ok 3 nl_netdev.page_pool_check
ok 4 nl_netdev.napi_list_check
ok 5 nl_netdev.dev_set_threaded
ok 6 nl_netdev.napi_set_threaded
ok 7 nl_netdev.nsim_rxq_reset_down
# Totals: pass:7 fail:0 xfail:0 xpass:0 skip:0 error:0
Signed-off-by: Samiullah Khawaja <skhawaja@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250710211203.3979655-1-skhawaja@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
authored by
Samiullah Khawaja
and committed by
Jakub Kicinski
2677010e
a44312d5
+162
-7
+10
Documentation/netlink/specs/netdev.yaml
+10
Documentation/netlink/specs/netdev.yaml
···
283
283
doc: The timeout, in nanoseconds, of how long to suspend irq
284
284
processing, if event polling finds events
285
285
type: uint
286
+
-
287
+
name: threaded
288
+
doc: Whether the NAPI is configured to operate in threaded polling
289
+
mode. If this is set to 1 then the NAPI context operates in
290
+
threaded polling mode.
291
+
type: uint
292
+
checks:
293
+
max: 1
286
294
-
287
295
name: xsk-info
288
296
attributes: []
···
702
694
- defer-hard-irqs
703
695
- gro-flush-timeout
704
696
- irq-suspend-timeout
697
+
- threaded
705
698
dump:
706
699
request:
707
700
attributes:
···
755
746
- defer-hard-irqs
756
747
- gro-flush-timeout
757
748
- irq-suspend-timeout
749
+
- threaded
758
750
-
759
751
name: bind-tx
760
752
doc: Bind dmabuf to netdev for TX
+8
-1
Documentation/networking/napi.rst
+8
-1
Documentation/networking/napi.rst
···
444
444
order than the process IDs of the kernel threads.
445
445
446
446
Threaded NAPI is controlled by writing 0/1 to the ``threaded`` file in
447
-
netdev's sysfs directory.
447
+
netdev's sysfs directory. It can also be enabled for a specific NAPI using
448
+
netlink interface.
449
+
450
+
For example, using the script:
451
+
452
+
.. code-block:: bash
453
+
454
+
$ ynl --family netdev --do napi-set --json='{"id": 66, "threaded": 1}'
448
455
449
456
.. rubric:: Footnotes
450
457
+1
include/linux/netdevice.h
+1
include/linux/netdevice.h
+1
include/uapi/linux/netdev.h
+1
include/uapi/linux/netdev.h
+27
-3
net/core/dev.c
+27
-3
net/core/dev.c
···
6961
6961
napi->thread = NULL;
6962
6962
}
6963
6963
6964
+
int napi_set_threaded(struct napi_struct *napi, bool threaded)
6965
+
{
6966
+
if (threaded) {
6967
+
if (!napi->thread) {
6968
+
int err = napi_kthread_create(napi);
6969
+
6970
+
if (err)
6971
+
return err;
6972
+
}
6973
+
}
6974
+
6975
+
if (napi->config)
6976
+
napi->config->threaded = threaded;
6977
+
6978
+
if (!threaded && napi->thread) {
6979
+
napi_stop_kthread(napi);
6980
+
} else {
6981
+
/* Make sure kthread is created before THREADED bit is set. */
6982
+
smp_mb__before_atomic();
6983
+
assign_bit(NAPI_STATE_THREADED, &napi->state, threaded);
6984
+
}
6985
+
6986
+
return 0;
6987
+
}
6988
+
6964
6989
int dev_set_threaded(struct net_device *dev, bool threaded)
6965
6990
{
6966
6991
struct napi_struct *napi;
6967
6992
int err = 0;
6968
6993
6969
6994
netdev_assert_locked_or_invisible(dev);
6970
-
6971
-
if (dev->threaded == threaded)
6972
-
return 0;
6973
6995
6974
6996
if (threaded) {
6975
6997
list_for_each_entry(napi, &dev->napi_list, dev_list) {
···
7243
7221
napi_hash_add(n);
7244
7222
n->config->napi_id = n->napi_id;
7245
7223
}
7224
+
7225
+
WARN_ON_ONCE(napi_set_threaded(n, n->config->threaded));
7246
7226
}
7247
7227
7248
7228
static void napi_save_config(struct napi_struct *n)
+7
net/core/dev.h
+7
net/core/dev.h
···
315
315
WRITE_ONCE(n->irq_suspend_timeout, timeout);
316
316
}
317
317
318
+
static inline bool napi_get_threaded(struct napi_struct *n)
319
+
{
320
+
return test_bit(NAPI_STATE_THREADED, &n->state);
321
+
}
322
+
323
+
int napi_set_threaded(struct napi_struct *n, bool threaded);
324
+
318
325
int rps_cpumask_housekeeping(struct cpumask *mask);
319
326
320
327
#if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL)
+3
-2
net/core/netdev-genl-gen.c
+3
-2
net/core/netdev-genl-gen.c
···
92
92
};
93
93
94
94
/* NETDEV_CMD_NAPI_SET - do */
95
-
static const struct nla_policy netdev_napi_set_nl_policy[NETDEV_A_NAPI_IRQ_SUSPEND_TIMEOUT + 1] = {
95
+
static const struct nla_policy netdev_napi_set_nl_policy[NETDEV_A_NAPI_THREADED + 1] = {
96
96
[NETDEV_A_NAPI_ID] = { .type = NLA_U32, },
97
97
[NETDEV_A_NAPI_DEFER_HARD_IRQS] = NLA_POLICY_FULL_RANGE(NLA_U32, &netdev_a_napi_defer_hard_irqs_range),
98
98
[NETDEV_A_NAPI_GRO_FLUSH_TIMEOUT] = { .type = NLA_UINT, },
99
99
[NETDEV_A_NAPI_IRQ_SUSPEND_TIMEOUT] = { .type = NLA_UINT, },
100
+
[NETDEV_A_NAPI_THREADED] = NLA_POLICY_MAX(NLA_UINT, 1),
100
101
};
101
102
102
103
/* NETDEV_CMD_BIND_TX - do */
···
194
193
.cmd = NETDEV_CMD_NAPI_SET,
195
194
.doit = netdev_nl_napi_set_doit,
196
195
.policy = netdev_napi_set_nl_policy,
197
-
.maxattr = NETDEV_A_NAPI_IRQ_SUSPEND_TIMEOUT,
196
+
.maxattr = NETDEV_A_NAPI_THREADED,
198
197
.flags = GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
199
198
},
200
199
{
+14
net/core/netdev-genl.c
+14
net/core/netdev-genl.c
···
184
184
if (napi->irq >= 0 && nla_put_u32(rsp, NETDEV_A_NAPI_IRQ, napi->irq))
185
185
goto nla_put_failure;
186
186
187
+
if (nla_put_uint(rsp, NETDEV_A_NAPI_THREADED,
188
+
napi_get_threaded(napi)))
189
+
goto nla_put_failure;
190
+
187
191
if (napi->thread) {
188
192
pid = task_pid_nr(napi->thread);
189
193
if (nla_put_u32(rsp, NETDEV_A_NAPI_PID, pid))
···
326
322
{
327
323
u64 irq_suspend_timeout = 0;
328
324
u64 gro_flush_timeout = 0;
325
+
u8 threaded = 0;
329
326
u32 defer = 0;
327
+
328
+
if (info->attrs[NETDEV_A_NAPI_THREADED]) {
329
+
int ret;
330
+
331
+
threaded = nla_get_uint(info->attrs[NETDEV_A_NAPI_THREADED]);
332
+
ret = napi_set_threaded(napi, !!threaded);
333
+
if (ret)
334
+
return ret;
335
+
}
330
336
331
337
if (info->attrs[NETDEV_A_NAPI_DEFER_HARD_IRQS]) {
332
338
defer = nla_get_u32(info->attrs[NETDEV_A_NAPI_DEFER_HARD_IRQS]);
+1
tools/include/uapi/linux/netdev.h
+1
tools/include/uapi/linux/netdev.h
+90
-1
tools/testing/selftests/net/nl_netdev.py
+90
-1
tools/testing/selftests/net/nl_netdev.py
···
35
35
ksft_eq(len(napis), 100,
36
36
comment=f"queue count after reset queue {q} mode {i}")
37
37
38
+
def napi_set_threaded(nf) -> None:
39
+
"""
40
+
Test that verifies various cases of napi threaded
41
+
set and unset at napi and device level.
42
+
"""
43
+
with NetdevSimDev(queue_count=2) as nsimdev:
44
+
nsim = nsimdev.nsims[0]
45
+
46
+
ip(f"link set dev {nsim.ifname} up")
47
+
48
+
napis = nf.napi_get({'ifindex': nsim.ifindex}, dump=True)
49
+
ksft_eq(len(napis), 2)
50
+
51
+
napi0_id = napis[0]['id']
52
+
napi1_id = napis[1]['id']
53
+
54
+
# set napi threaded and verify
55
+
nf.napi_set({'id': napi0_id, 'threaded': 1})
56
+
napi0 = nf.napi_get({'id': napi0_id})
57
+
ksft_eq(napi0['threaded'], 1)
58
+
ksft_ne(napi0.get('pid'), None)
59
+
60
+
# check it is not set for napi1
61
+
napi1 = nf.napi_get({'id': napi1_id})
62
+
ksft_eq(napi1['threaded'], 0)
63
+
ksft_eq(napi1.get('pid'), None)
64
+
65
+
ip(f"link set dev {nsim.ifname} down")
66
+
ip(f"link set dev {nsim.ifname} up")
67
+
68
+
# verify if napi threaded is still set
69
+
napi0 = nf.napi_get({'id': napi0_id})
70
+
ksft_eq(napi0['threaded'], 1)
71
+
ksft_ne(napi0.get('pid'), None)
72
+
73
+
# check it is still not set for napi1
74
+
napi1 = nf.napi_get({'id': napi1_id})
75
+
ksft_eq(napi1['threaded'], 0)
76
+
ksft_eq(napi1.get('pid'), None)
77
+
78
+
# unset napi threaded and verify
79
+
nf.napi_set({'id': napi0_id, 'threaded': 0})
80
+
napi0 = nf.napi_get({'id': napi0_id})
81
+
ksft_eq(napi0['threaded'], 0)
82
+
ksft_eq(napi0.get('pid'), None)
83
+
84
+
# set threaded at device level
85
+
system(f"echo 1 > /sys/class/net/{nsim.ifname}/threaded")
86
+
87
+
# check napi threaded is set for both napis
88
+
napi0 = nf.napi_get({'id': napi0_id})
89
+
ksft_eq(napi0['threaded'], 1)
90
+
ksft_ne(napi0.get('pid'), None)
91
+
napi1 = nf.napi_get({'id': napi1_id})
92
+
ksft_eq(napi1['threaded'], 1)
93
+
ksft_ne(napi1.get('pid'), None)
94
+
95
+
# unset threaded at device level
96
+
system(f"echo 0 > /sys/class/net/{nsim.ifname}/threaded")
97
+
98
+
# check napi threaded is unset for both napis
99
+
napi0 = nf.napi_get({'id': napi0_id})
100
+
ksft_eq(napi0['threaded'], 0)
101
+
ksft_eq(napi0.get('pid'), None)
102
+
napi1 = nf.napi_get({'id': napi1_id})
103
+
ksft_eq(napi1['threaded'], 0)
104
+
ksft_eq(napi1.get('pid'), None)
105
+
106
+
# set napi threaded for napi0
107
+
nf.napi_set({'id': napi0_id, 'threaded': 1})
108
+
napi0 = nf.napi_get({'id': napi0_id})
109
+
ksft_eq(napi0['threaded'], 1)
110
+
ksft_ne(napi0.get('pid'), None)
111
+
112
+
# unset threaded at device level
113
+
system(f"echo 0 > /sys/class/net/{nsim.ifname}/threaded")
114
+
115
+
# check napi threaded is unset for both napis
116
+
napi0 = nf.napi_get({'id': napi0_id})
117
+
ksft_eq(napi0['threaded'], 0)
118
+
ksft_eq(napi0.get('pid'), None)
119
+
napi1 = nf.napi_get({'id': napi1_id})
120
+
ksft_eq(napi1['threaded'], 0)
121
+
ksft_eq(napi1.get('pid'), None)
122
+
38
123
def dev_set_threaded(nf) -> None:
39
124
"""
40
125
Test that verifies various cases of napi threaded
···
141
56
142
57
# check napi threaded is set for both napis
143
58
napi0 = nf.napi_get({'id': napi0_id})
59
+
ksft_eq(napi0['threaded'], 1)
144
60
ksft_ne(napi0.get('pid'), None)
145
61
napi1 = nf.napi_get({'id': napi1_id})
62
+
ksft_eq(napi1['threaded'], 1)
146
63
ksft_ne(napi1.get('pid'), None)
147
64
148
65
# unset threaded
···
152
65
153
66
# check napi threaded is unset for both napis
154
67
napi0 = nf.napi_get({'id': napi0_id})
68
+
ksft_eq(napi0['threaded'], 0)
155
69
ksft_eq(napi0.get('pid'), None)
156
70
napi1 = nf.napi_get({'id': napi1_id})
71
+
ksft_eq(napi1['threaded'], 0)
157
72
ksft_eq(napi1.get('pid'), None)
158
73
159
74
def nsim_rxq_reset_down(nf) -> None:
···
245
156
def main() -> None:
246
157
nf = NetdevFamily()
247
158
ksft_run([empty_check, lo_check, page_pool_check, napi_list_check,
248
-
dev_set_threaded, nsim_rxq_reset_down],
159
+
dev_set_threaded, napi_set_threaded, nsim_rxq_reset_down],
249
160
args=(nf, ))
250
161
ksft_exit()
251
162