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 <[email protected]>
Reviewed-by: Willem de Bruijn <[email protected]>
Link: https://patch.msgid.link/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Author: samikhawaja

Documentation/netlink/specs/netdev.yaml

@@ -283,6 +283,14 @@ attribute-sets:
         doc: The timeout, in nanoseconds, of how long to suspend irq
              processing, if event polling finds events
         type: uint
+      -
+        name: threaded
+        doc: Whether the NAPI is configured to operate in threaded polling
+             mode. If this is set to 1 then the NAPI context operates in
+             threaded polling mode.
+        type: uint
+        checks:
+          max: 1
   -
     name: xsk-info
     attributes: []
@@ -694,6 +702,7 @@ operations:
             - defer-hard-irqs
             - gro-flush-timeout
             - irq-suspend-timeout
+            - threaded
       dump:
         request:
           attributes:
@@ -746,6 +755,7 @@ operations:
             - defer-hard-irqs
             - gro-flush-timeout
             - irq-suspend-timeout
+            - threaded
     -
       name: bind-tx
       doc: Bind dmabuf to netdev for TX

Documentation/networking/napi.rst

@@ -444,7 +444,14 @@ dependent). The NAPI instance IDs will be assigned in the opposite
 order than the process IDs of the kernel threads.
 
 Threaded NAPI is controlled by writing 0/1 to the ``threaded`` file in
-netdev's sysfs directory.
+netdev's sysfs directory. It can also be enabled for a specific NAPI using
+netlink interface.
+
+For example, using the script:
+
+.. code-block:: bash
+
+  $ ynl --family netdev --do napi-set --json='{"id": 66, "threaded": 1}'
 
 .. rubric:: Footnotes
 

include/linux/netdevice.h

@@ -369,6 +369,7 @@ struct napi_config {
 	u64 irq_suspend_timeout;
 	u32 defer_hard_irqs;
 	cpumask_t affinity_mask;
+	bool threaded;
 	unsigned int napi_id;
 };
 

include/uapi/linux/netdev.h

@@ -134,6 +134,7 @@ enum {
 	NETDEV_A_NAPI_DEFER_HARD_IRQS,
 	NETDEV_A_NAPI_GRO_FLUSH_TIMEOUT,
 	NETDEV_A_NAPI_IRQ_SUSPEND_TIMEOUT,
+	NETDEV_A_NAPI_THREADED,
 
 	__NETDEV_A_NAPI_MAX,
 	NETDEV_A_NAPI_MAX = (__NETDEV_A_NAPI_MAX - 1)

net/core/dev.c

@@ -6961,16 +6961,38 @@ static void napi_stop_kthread(struct napi_struct *napi)
 	napi->thread = NULL;
 }
 
+int napi_set_threaded(struct napi_struct *napi, bool threaded)
+{
+	if (threaded) {
+		if (!napi->thread) {
+			int err = napi_kthread_create(napi);
+
+			if (err)
+				return err;
+		}
+	}
+
+	if (napi->config)
+		napi->config->threaded = threaded;
+
+	if (!threaded && napi->thread) {
+		napi_stop_kthread(napi);
+	} else {
+		/* Make sure kthread is created before THREADED bit is set. */
+		smp_mb__before_atomic();
+		assign_bit(NAPI_STATE_THREADED, &napi->state, threaded);
+	}
+
+	return 0;
+}
+
 int dev_set_threaded(struct net_device *dev, bool threaded)
 {
 	struct napi_struct *napi;
 	int err = 0;
 
 	netdev_assert_locked_or_invisible(dev);
 
-	if (dev->threaded == threaded)
-		return 0;
-
 	if (threaded) {
 		list_for_each_entry(napi, &dev->napi_list, dev_list) {
 			if (!napi->thread) {
@@ -7221,6 +7243,8 @@ static void napi_restore_config(struct napi_struct *n)
 		napi_hash_add(n);
 		n->config->napi_id = n->napi_id;
 	}
+
+	WARN_ON_ONCE(napi_set_threaded(n, n->config->threaded));
 }
 
 static void napi_save_config(struct napi_struct *n)

net/core/dev.h

@@ -315,6 +315,13 @@ static inline void napi_set_irq_suspend_timeout(struct napi_struct *n,
 	WRITE_ONCE(n->irq_suspend_timeout, timeout);
 }
 
+static inline bool napi_get_threaded(struct napi_struct *n)
+{
+	return test_bit(NAPI_STATE_THREADED, &n->state);
+}
+
+int napi_set_threaded(struct napi_struct *n, bool threaded);
+
 int rps_cpumask_housekeeping(struct cpumask *mask);
 
 #if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL)

net/core/netdev-genl-gen.c

@@ -92,11 +92,12 @@ static const struct nla_policy netdev_bind_rx_nl_policy[NETDEV_A_DMABUF_FD + 1]
 };
 
 /* NETDEV_CMD_NAPI_SET - do */
-static const struct nla_policy netdev_napi_set_nl_policy[NETDEV_A_NAPI_IRQ_SUSPEND_TIMEOUT + 1] = {
+static const struct nla_policy netdev_napi_set_nl_policy[NETDEV_A_NAPI_THREADED + 1] = {
 	[NETDEV_A_NAPI_ID] = { .type = NLA_U32, },
 	[NETDEV_A_NAPI_DEFER_HARD_IRQS] = NLA_POLICY_FULL_RANGE(NLA_U32, &netdev_a_napi_defer_hard_irqs_range),
 	[NETDEV_A_NAPI_GRO_FLUSH_TIMEOUT] = { .type = NLA_UINT, },
 	[NETDEV_A_NAPI_IRQ_SUSPEND_TIMEOUT] = { .type = NLA_UINT, },
+	[NETDEV_A_NAPI_THREADED] = NLA_POLICY_MAX(NLA_UINT, 1),
 };
 
 /* NETDEV_CMD_BIND_TX - do */
@@ -193,7 +194,7 @@ static const struct genl_split_ops netdev_nl_ops[] = {
 		.cmd		= NETDEV_CMD_NAPI_SET,
 		.doit		= netdev_nl_napi_set_doit,
 		.policy		= netdev_napi_set_nl_policy,
-		.maxattr	= NETDEV_A_NAPI_IRQ_SUSPEND_TIMEOUT,
+		.maxattr	= NETDEV_A_NAPI_THREADED,
 		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
 	},
 	{

net/core/netdev-genl.c

@@ -184,6 +184,10 @@ netdev_nl_napi_fill_one(struct sk_buff *rsp, struct napi_struct *napi,
 	if (napi->irq >= 0 && nla_put_u32(rsp, NETDEV_A_NAPI_IRQ, napi->irq))
 		goto nla_put_failure;
 
+	if (nla_put_uint(rsp, NETDEV_A_NAPI_THREADED,
+			 napi_get_threaded(napi)))
+		goto nla_put_failure;
+
 	if (napi->thread) {
 		pid = task_pid_nr(napi->thread);
 		if (nla_put_u32(rsp, NETDEV_A_NAPI_PID, pid))
@@ -322,8 +326,18 @@ netdev_nl_napi_set_config(struct napi_struct *napi, struct genl_info *info)
 {
 	u64 irq_suspend_timeout = 0;
 	u64 gro_flush_timeout = 0;
+	u8 threaded = 0;
 	u32 defer = 0;
 
+	if (info->attrs[NETDEV_A_NAPI_THREADED]) {
+		int ret;
+
+		threaded = nla_get_uint(info->attrs[NETDEV_A_NAPI_THREADED]);
+		ret = napi_set_threaded(napi, !!threaded);
+		if (ret)
+			return ret;
+	}
+
 	if (info->attrs[NETDEV_A_NAPI_DEFER_HARD_IRQS]) {
 		defer = nla_get_u32(info->attrs[NETDEV_A_NAPI_DEFER_HARD_IRQS]);
 		napi_set_defer_hard_irqs(napi, defer);

tools/include/uapi/linux/netdev.h

@@ -134,6 +134,7 @@ enum {
 	NETDEV_A_NAPI_DEFER_HARD_IRQS,
 	NETDEV_A_NAPI_GRO_FLUSH_TIMEOUT,
 	NETDEV_A_NAPI_IRQ_SUSPEND_TIMEOUT,
+	NETDEV_A_NAPI_THREADED,
 
 	__NETDEV_A_NAPI_MAX,
 	NETDEV_A_NAPI_MAX = (__NETDEV_A_NAPI_MAX - 1)

tools/testing/selftests/net/nl_netdev.py

@@ -35,6 +35,91 @@ def napi_list_check(nf) -> None:
                 ksft_eq(len(napis), 100,
                         comment=f"queue count after reset queue {q} mode {i}")
 
+def napi_set_threaded(nf) -> None:
+    """
+    Test that verifies various cases of napi threaded
+    set and unset at napi and device level.
+    """
+    with NetdevSimDev(queue_count=2) as nsimdev:
+        nsim = nsimdev.nsims[0]
+
+        ip(f"link set dev {nsim.ifname} up")
+
+        napis = nf.napi_get({'ifindex': nsim.ifindex}, dump=True)
+        ksft_eq(len(napis), 2)
+
+        napi0_id = napis[0]['id']
+        napi1_id = napis[1]['id']
+
+        # set napi threaded and verify
+        nf.napi_set({'id': napi0_id, 'threaded': 1})
+        napi0 = nf.napi_get({'id': napi0_id})
+        ksft_eq(napi0['threaded'], 1)
+        ksft_ne(napi0.get('pid'), None)
+
+        # check it is not set for napi1
+        napi1 = nf.napi_get({'id': napi1_id})
+        ksft_eq(napi1['threaded'], 0)
+        ksft_eq(napi1.get('pid'), None)
+
+        ip(f"link set dev {nsim.ifname} down")
+        ip(f"link set dev {nsim.ifname} up")
+
+        # verify if napi threaded is still set
+        napi0 = nf.napi_get({'id': napi0_id})
+        ksft_eq(napi0['threaded'], 1)
+        ksft_ne(napi0.get('pid'), None)
+
+        # check it is still not set for napi1
+        napi1 = nf.napi_get({'id': napi1_id})
+        ksft_eq(napi1['threaded'], 0)
+        ksft_eq(napi1.get('pid'), None)
+
+        # unset napi threaded and verify
+        nf.napi_set({'id': napi0_id, 'threaded': 0})
+        napi0 = nf.napi_get({'id': napi0_id})
+        ksft_eq(napi0['threaded'], 0)
+        ksft_eq(napi0.get('pid'), None)
+
+        # set threaded at device level
+        system(f"echo 1 > /sys/class/net/{nsim.ifname}/threaded")
+
+        # check napi threaded is set for both napis
+        napi0 = nf.napi_get({'id': napi0_id})
+        ksft_eq(napi0['threaded'], 1)
+        ksft_ne(napi0.get('pid'), None)
+        napi1 = nf.napi_get({'id': napi1_id})
+        ksft_eq(napi1['threaded'], 1)
+        ksft_ne(napi1.get('pid'), None)
+
+        # unset threaded at device level
+        system(f"echo 0 > /sys/class/net/{nsim.ifname}/threaded")
+
+        # check napi threaded is unset for both napis
+        napi0 = nf.napi_get({'id': napi0_id})
+        ksft_eq(napi0['threaded'], 0)
+        ksft_eq(napi0.get('pid'), None)
+        napi1 = nf.napi_get({'id': napi1_id})
+        ksft_eq(napi1['threaded'], 0)
+        ksft_eq(napi1.get('pid'), None)
+
+        # set napi threaded for napi0
+        nf.napi_set({'id': napi0_id, 'threaded': 1})
+        napi0 = nf.napi_get({'id': napi0_id})
+        ksft_eq(napi0['threaded'], 1)
+        ksft_ne(napi0.get('pid'), None)
+
+        # unset threaded at device level
+        system(f"echo 0 > /sys/class/net/{nsim.ifname}/threaded")
+
+        # check napi threaded is unset for both napis
+        napi0 = nf.napi_get({'id': napi0_id})
+        ksft_eq(napi0['threaded'], 0)
+        ksft_eq(napi0.get('pid'), None)
+        napi1 = nf.napi_get({'id': napi1_id})
+        ksft_eq(napi1['threaded'], 0)
+        ksft_eq(napi1.get('pid'), None)
+
 def dev_set_threaded(nf) -> None:
     """
     Test that verifies various cases of napi threaded
@@ -56,17 +141,21 @@ def dev_set_threaded(nf) -> None:
 
         # check napi threaded is set for both napis
         napi0 = nf.napi_get({'id': napi0_id})
+        ksft_eq(napi0['threaded'], 1)
         ksft_ne(napi0.get('pid'), None)
         napi1 = nf.napi_get({'id': napi1_id})
+        ksft_eq(napi1['threaded'], 1)
         ksft_ne(napi1.get('pid'), None)
 
         # unset threaded
         system(f"echo 0 > /sys/class/net/{nsim.ifname}/threaded")
 
         # check napi threaded is unset for both napis
         napi0 = nf.napi_get({'id': napi0_id})
+        ksft_eq(napi0['threaded'], 0)
         ksft_eq(napi0.get('pid'), None)
         napi1 = nf.napi_get({'id': napi1_id})
+        ksft_eq(napi1['threaded'], 0)
         ksft_eq(napi1.get('pid'), None)
 
 def nsim_rxq_reset_down(nf) -> None:
@@ -156,7 +245,7 @@ def get_pp():
 def main() -> None:
     nf = NetdevFamily()
     ksft_run([empty_check, lo_check, page_pool_check, napi_list_check,
-              dev_set_threaded, nsim_rxq_reset_down],
+              dev_set_threaded, napi_set_threaded, nsim_rxq_reset_down],
              args=(nf, ))
     ksft_exit()