Add ndo_set_rx_mode_async callback that drivers can implement instead
of the legacy ndo_set_rx_mode. The legacy callback runs under the
netif_addr_lock spinlock with BHs disabled, preventing drivers from
sleeping. The async variant runs from a work queue with rtnl_lock and
netdev_lock_ops held, in fully sleepable context.

When __dev_set_rx_mode() sees ndo_set_rx_mode_async, it schedules
dev_rx_mode_work instead of calling the driver inline. The work
function takes two snapshots of each address list (uc/mc) under
the addr_lock, then drops the lock and calls the driver with the
work copies. After the driver returns, it reconciles the snapshots
back to the real lists under the lock.

Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
 Documentation/networking/netdevices.rst |  8 +++
 include/linux/netdevice.h               | 20 ++++++
 net/core/dev.c                          | 94 +++++++++++++++++++++++--
 3 files changed, 115 insertions(+), 7 deletions(-)

diff --git a/Documentation/networking/netdevices.rst b/Documentation/networking/netdevices.rst
index 35704d115312..dc83d78d3b27 100644
--- a/Documentation/networking/netdevices.rst
+++ b/Documentation/networking/netdevices.rst
@@ -289,6 +289,14 @@ struct net_device synchronization rules
 ndo_set_rx_mode:
 	Synchronization: netif_addr_lock spinlock.
 	Context: BHs disabled
+	Notes: Deprecated in favor of sleepable ndo_set_rx_mode_async.
+
+ndo_set_rx_mode_async:
+	Synchronization: rtnl_lock() semaphore. In addition, netdev instance
+	lock if the driver implements queue management or shaper API.
+	Context: process (from a work queue)
+	Notes: Sleepable version of ndo_set_rx_mode. Receives snapshots
+	of the unicast and multicast address lists.
 
 ndo_setup_tc:
 	``TC_SETUP_BLOCK`` and ``TC_SETUP_FT`` are running under NFT locks
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 469b7cdb3237..7ede1f56bd70 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1117,6 +1117,16 @@ struct netdev_net_notifier {
  *	This function is called device changes address list filtering.
  *	If driver handles unicast address filtering, it should set
  *	IFF_UNICAST_FLT in its priv_flags.
+ *	Cannot sleep, called with netif_addr_lock_bh held.
+ *	Deprecated in favor of sleepable ndo_set_rx_mode_async.
+ *
+ * void (*ndo_set_rx_mode_async)(struct net_device *dev,
+ *				  struct netdev_hw_addr_list *uc,
+ *				  struct netdev_hw_addr_list *mc);
+ *	Sleepable version of ndo_set_rx_mode. Called from a work queue
+ *	with rtnl_lock and netdev_lock_ops(dev) held. The uc/mc parameters
+ *	are snapshots of the address lists - iterate with
+ *	netdev_hw_addr_list_for_each(ha, uc).
  *
  * int (*ndo_set_mac_address)(struct net_device *dev, void *addr);
  *	This function  is called when the Media Access Control address
@@ -1437,6 +1447,9 @@ struct net_device_ops {
 	void			(*ndo_change_rx_flags)(struct net_device *dev,
 						       int flags);
 	void			(*ndo_set_rx_mode)(struct net_device *dev);
+	void			(*ndo_set_rx_mode_async)(struct net_device *dev,
+					struct netdev_hw_addr_list *uc,
+					struct netdev_hw_addr_list *mc);
 	int			(*ndo_set_mac_address)(struct net_device *dev,
 						       void *addr);
 	int			(*ndo_validate_addr)(struct net_device *dev);
@@ -1903,6 +1916,7 @@ enum netdev_reg_state {
  *				has been enabled due to the need to listen to
  *				additional unicast addresses in a device that
  *				does not implement ndo_set_rx_mode()
+ *	@rx_mode_work:		Work queue entry for ndo_set_rx_mode_async()
  *	@uc:			unicast mac addresses
  *	@mc:			multicast mac addresses
  *	@dev_addrs:		list of device hw addresses
@@ -2293,6 +2307,7 @@ struct net_device {
 	unsigned int		promiscuity;
 	unsigned int		allmulti;
 	bool			uc_promisc;
+	struct work_struct	rx_mode_work;
 #ifdef CONFIG_LOCKDEP
 	unsigned char		nested_level;
 #endif
@@ -4661,6 +4676,11 @@ static inline bool netif_device_present(const struct net_device *dev)
 	return test_bit(__LINK_STATE_PRESENT, &dev->state);
 }
 
+static inline bool netif_up_and_present(const struct net_device *dev)
+{
+	return (dev->flags & IFF_UP) && netif_device_present(dev);
+}
+
 void netif_device_detach(struct net_device *dev);
 
 void netif_device_attach(struct net_device *dev);
diff --git a/net/core/dev.c b/net/core/dev.c
index 200d44883fc1..77fdbe836754 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2381,6 +2381,8 @@ static void netstamp_clear(struct work_struct *work)
 static DECLARE_WORK(netstamp_work, netstamp_clear);
 #endif
 
+static struct workqueue_struct *rx_mode_wq;
+
 void net_enable_timestamp(void)
 {
 #ifdef CONFIG_JUMP_LABEL
@@ -9669,22 +9671,83 @@ int netif_set_allmulti(struct net_device *dev, int inc, bool notify)
 	return 0;
 }
 
-/*
- *	Upload unicast and multicast address lists to device and
- *	configure RX filtering. When the device doesn't support unicast
- *	filtering it is put in promiscuous mode while unicast addresses
- *	are present.
+static void dev_rx_mode_work(struct work_struct *work)
+{
+	struct net_device *dev = container_of(work, struct net_device,
+					      rx_mode_work);
+	struct netdev_hw_addr_list uc_snap, mc_snap, uc_ref, mc_ref;
+	const struct net_device_ops *ops = dev->netdev_ops;
+	int err;
+
+	__hw_addr_init(&uc_snap);
+	__hw_addr_init(&mc_snap);
+	__hw_addr_init(&uc_ref);
+	__hw_addr_init(&mc_ref);
+
+	rtnl_lock();
+	netdev_lock_ops(dev);
+
+	if (!netif_up_and_present(dev))
+		goto out;
+
+	if (ops->ndo_set_rx_mode_async) {
+		netif_addr_lock_bh(dev);
+
+		err = __hw_addr_list_snapshot(&uc_snap, &dev->uc,
+					      dev->addr_len);
+		if (!err)
+			err = __hw_addr_list_snapshot(&uc_ref, &dev->uc,
+						      dev->addr_len);
+		if (!err)
+			err = __hw_addr_list_snapshot(&mc_snap, &dev->mc,
+						      dev->addr_len);
+		if (!err)
+			err = __hw_addr_list_snapshot(&mc_ref, &dev->mc,
+						      dev->addr_len);
+		netif_addr_unlock_bh(dev);
+
+		if (err) {
+			__hw_addr_flush(&uc_snap);
+			__hw_addr_flush(&uc_ref);
+			__hw_addr_flush(&mc_snap);
+			goto out;
+		}
+
+		ops->ndo_set_rx_mode_async(dev, &uc_snap, &mc_snap);
+
+		netif_addr_lock_bh(dev);
+		__hw_addr_list_reconcile(&dev->uc, &uc_snap,
+					 &uc_ref, dev->addr_len);
+		__hw_addr_list_reconcile(&dev->mc, &mc_snap,
+					 &mc_ref, dev->addr_len);
+		netif_addr_unlock_bh(dev);
+	}
+
+out:
+	netdev_unlock_ops(dev);
+	rtnl_unlock();
+}
+
+/**
+ * __dev_set_rx_mode() - upload unicast and multicast address lists to device
+ * and configure RX filtering.
+ * @dev: device
+ *
+ * When the device doesn't support unicast filtering it is put in promiscuous
+ * mode while unicast addresses are present.
  */
 void __dev_set_rx_mode(struct net_device *dev)
 {
 	const struct net_device_ops *ops = dev->netdev_ops;
 
 	/* dev_open will call this function so the list will stay sane. */
-	if (!(dev->flags&IFF_UP))
+	if (!netif_up_and_present(dev))
 		return;
 
-	if (!netif_device_present(dev))
+	if (ops->ndo_set_rx_mode_async) {
+		queue_work(rx_mode_wq, &dev->rx_mode_work);
 		return;
+	}
 
 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
 		/* Unicast addresses changes may only happen under the rtnl,
@@ -11708,6 +11771,16 @@ void netdev_run_todo(void)
 
 	__rtnl_unlock();
 
+	/* Make sure all pending rx_mode work completes before returning.
+	 *
+	 * rx_mode_wq may be NULL during early boot:
+	 * core_initcall(netlink_proto_init) vs subsys_initcall(net_dev_init).
+	 *
+	 * Check current_work() to avoid flushing from the wq.
+	 */
+	if (rx_mode_wq && !current_work())
+		flush_workqueue(rx_mode_wq);
+
 	/* Wait for rcu callbacks to finish before next phase */
 	if (!list_empty(&list))
 		rcu_barrier();
@@ -12099,6 +12172,7 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
 #endif
 
 	mutex_init(&dev->lock);
+	INIT_WORK(&dev->rx_mode_work, dev_rx_mode_work);
 
 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
 	setup(dev);
@@ -12203,6 +12277,8 @@ void free_netdev(struct net_device *dev)
 
 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
 
+	cancel_work_sync(&dev->rx_mode_work);
+
 	/* Flush device addresses */
 	dev_addr_flush(dev);
 
@@ -13296,6 +13372,10 @@ static int __init net_dev_init(void)
 	if (register_pernet_device(&default_device_ops))
 		goto out;
 
+	rx_mode_wq = alloc_ordered_workqueue("rx_mode_wq", 0);
+	if (!rx_mode_wq)
+		goto out;
+
 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
 
-- 
2.53.0

> -----Original Message-----
> From: Intel-wired-lan <intel-wired-lan-bounces@osuosl.org> On Behalf
> Of Stanislav Fomichev
> Sent: Wednesday, March 18, 2026 4:03 PM
> To: netdev@vger.kernel.org
> Cc: davem@davemloft.net; edumazet@google.com; kuba@kernel.org;
> pabeni@redhat.com; horms@kernel.org; corbet@lwn.net;
> skhan@linuxfoundation.org; andrew+netdev@lunn.ch;
> michael.chan@broadcom.com; pavan.chebbi@broadcom.com; Nguyen, Anthony
> L <anthony.l.nguyen@intel.com>; Kitszel, Przemyslaw
> <przemyslaw.kitszel@intel.com>; saeedm@nvidia.com; tariqt@nvidia.com;
> mbloch@nvidia.com; alexanderduyck@fb.com; kernel-team@meta.com;
> johannes@sipsolutions.net; sd@queasysnail.net; jianbol@nvidia.com;
> dtatulea@nvidia.com; sdf@fomichev.me; mohsin.bashr@gmail.com; Keller,
> Jacob E <jacob.e.keller@intel.com>; willemb@google.com;
> skhawaja@google.com; bestswngs@gmail.com; linux-doc@vger.kernel.org;
> linux-kernel@vger.kernel.org; intel-wired-lan@lists.osuosl.org; linux-
> rdma@vger.kernel.org; linux-wireless@vger.kernel.org; linux-
> kselftest@vger.kernel.org; leon@kernel.org
> Subject: [Intel-wired-lan] [PATCH net-next v2 03/13] net: introduce
> ndo_set_rx_mode_async and dev_rx_mode_work
> 
> Add ndo_set_rx_mode_async callback that drivers can implement instead
> of the legacy ndo_set_rx_mode. The legacy callback runs under the
> netif_addr_lock spinlock with BHs disabled, preventing drivers from
> sleeping. The async variant runs from a work queue with rtnl_lock and
> netdev_lock_ops held, in fully sleepable context.
> 
> When __dev_set_rx_mode() sees ndo_set_rx_mode_async, it schedules
> dev_rx_mode_work instead of calling the driver inline. The work
> function takes two snapshots of each address list (uc/mc) under the
> addr_lock, then drops the lock and calls the driver with the work
> copies. After the driver returns, it reconciles the snapshots back to
> the real lists under the lock.
> 
> Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
> ---
>  Documentation/networking/netdevices.rst |  8 +++
>  include/linux/netdevice.h               | 20 ++++++
>  net/core/dev.c                          | 94 +++++++++++++++++++++++-
> -
>  3 files changed, 115 insertions(+), 7 deletions(-)
> 
> diff --git a/Documentation/networking/netdevices.rst
> b/Documentation/networking/netdevices.rst
> index 35704d115312..dc83d78d3b27 100644
> --- a/Documentation/networking/netdevices.rst
> +++ b/Documentation/networking/netdevices.rst
> @@ -289,6 +289,14 @@ struct net_device synchronization rules
>  ndo_set_rx_mode:
>  	Synchronization: netif_addr_lock spinlock.
>  	Context: BHs disabled


...

> 
> -/*
> - *	Upload unicast and multicast address lists to device and
> - *	configure RX filtering. When the device doesn't support unicast
> - *	filtering it is put in promiscuous mode while unicast addresses
> - *	are present.
> +static void dev_rx_mode_work(struct work_struct *work) {
> +	struct net_device *dev = container_of(work, struct net_device,
> +					      rx_mode_work);
> +	struct netdev_hw_addr_list uc_snap, mc_snap, uc_ref, mc_ref;
> +	const struct net_device_ops *ops = dev->netdev_ops;
> +	int err;
> +
> +	__hw_addr_init(&uc_snap);
> +	__hw_addr_init(&mc_snap);
> +	__hw_addr_init(&uc_ref);
> +	__hw_addr_init(&mc_ref);
> +
> +	rtnl_lock();
> +	netdev_lock_ops(dev);
> +
> +	if (!netif_up_and_present(dev))
> +		goto out;
> +
> +	if (ops->ndo_set_rx_mode_async) {
> +		netif_addr_lock_bh(dev);
> +
> +		err = __hw_addr_list_snapshot(&uc_snap, &dev->uc,
> +					      dev->addr_len);
> +		if (!err)
> +			err = __hw_addr_list_snapshot(&uc_ref, &dev->uc,
> +						      dev->addr_len);
> +		if (!err)
> +			err = __hw_addr_list_snapshot(&mc_snap, &dev->mc,
> +						      dev->addr_len);
> +		if (!err)
> +			err = __hw_addr_list_snapshot(&mc_ref, &dev->mc,
> +						      dev->addr_len);
> +		netif_addr_unlock_bh(dev);
> +
> +		if (err) {
> +			__hw_addr_flush(&uc_snap);
> +			__hw_addr_flush(&uc_ref);
> +			__hw_addr_flush(&mc_snap);
Shouldn't here go cleanup for symmetry? 
			__hw_addr_flush(&mc_ref);

> +			goto out;
> +		}
> +
> +		ops->ndo_set_rx_mode_async(dev, &uc_snap, &mc_snap);
> +
> +		netif_addr_lock_bh(dev);
> +		__hw_addr_list_reconcile(&dev->uc, &uc_snap,
> +					 &uc_ref, dev->addr_len);
> +		__hw_addr_list_reconcile(&dev->mc, &mc_snap,
> +					 &mc_ref, dev->addr_len);
> +		netif_addr_unlock_bh(dev);
> +	}
> +
> +out:
> +	netdev_unlock_ops(dev);
> +	rtnl_unlock();
> +}

...

> --
> 2.53.0

> -----Original Message-----
> From: Intel-wired-lan <intel-wired-lan-bounces@osuosl.org> On Behalf
> Of Stanislav Fomichev
> Sent: Wednesday, March 18, 2026 4:03 PM
> To: netdev@vger.kernel.org
> Cc: davem@davemloft.net; edumazet@google.com; kuba@kernel.org;
> pabeni@redhat.com; horms@kernel.org; corbet@lwn.net;
> skhan@linuxfoundation.org; andrew+netdev@lunn.ch;
> michael.chan@broadcom.com; pavan.chebbi@broadcom.com; Nguyen, Anthony
> L <anthony.l.nguyen@intel.com>; Kitszel, Przemyslaw
> <przemyslaw.kitszel@intel.com>; saeedm@nvidia.com; tariqt@nvidia.com;
> mbloch@nvidia.com; alexanderduyck@fb.com; kernel-team@meta.com;
> johannes@sipsolutions.net; sd@queasysnail.net; jianbol@nvidia.com;
> dtatulea@nvidia.com; sdf@fomichev.me; mohsin.bashr@gmail.com; Keller,
> Jacob E <jacob.e.keller@intel.com>; willemb@google.com;
> skhawaja@google.com; bestswngs@gmail.com; linux-doc@vger.kernel.org;
> linux-kernel@vger.kernel.org; intel-wired-lan@lists.osuosl.org; linux-
> rdma@vger.kernel.org; linux-wireless@vger.kernel.org; linux-
> kselftest@vger.kernel.org; leon@kernel.org
> Subject: [Intel-wired-lan] [PATCH net-next v2 03/13] net: introduce
> ndo_set_rx_mode_async and dev_rx_mode_work
> 
> Add ndo_set_rx_mode_async callback that drivers can implement instead
> of the legacy ndo_set_rx_mode. The legacy callback runs under the
> netif_addr_lock spinlock with BHs disabled, preventing drivers from
> sleeping. The async variant runs from a work queue with rtnl_lock and
> netdev_lock_ops held, in fully sleepable context.
> 
> When __dev_set_rx_mode() sees ndo_set_rx_mode_async, it schedules
> dev_rx_mode_work instead of calling the driver inline. The work
> function takes two snapshots of each address list (uc/mc) under the
> addr_lock, then drops the lock and calls the driver with the work
> copies. After the driver returns, it reconciles the snapshots back to
> the real lists under the lock.
> 
> Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
> ---
>  Documentation/networking/netdevices.rst |  8 +++
>  include/linux/netdevice.h               | 20 ++++++
>  net/core/dev.c                          | 94 +++++++++++++++++++++++-
> -
>  3 files changed, 115 insertions(+), 7 deletions(-)
> 
> diff --git a/Documentation/networking/netdevices.rst
> b/Documentation/networking/netdevices.rst
> index 35704d115312..dc83d78d3b27 100644
> --- a/Documentation/networking/netdevices.rst
> +++ b/Documentation/networking/netdevices.rst
> @@ -289,6 +289,14 @@ struct net_device synchronization rules
>  ndo_set_rx_mode:
>  	Synchronization: netif_addr_lock spinlock.
>  	Context: BHs disabled

...

>  	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
>  	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
> 
> --
> 2.53.0

Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>