Add reference tracking to netdev and release netdev reference in nondl on shutdown

src/driver/linux_net/drivers/net/ethernet/sfc/net_driver.h

@@ -1291,13 +1291,15 @@ struct efx_arfs_rule {
 /**
  * struct efx_async_filter_insertion - Request to asynchronously insert a filter
  * @net_dev: Reference to the netdevice
+ * @net_dev_tracker: reference tracker entry for @net_dev
  * @spec: The filter to insert
  * @work: Workitem for this request
  * @rxq_index: Identifies the channel for which this request was made
  * @flow_id: Identifies the kernel-side flow for which this request was made
  */
 struct efx_async_filter_insertion {
 	struct net_device *net_dev;
+	netdevice_tracker net_dev_tracker;
 	struct efx_filter_spec spec;
 	struct work_struct work;
 	u16 rxq_index;

src/driver/linux_net/drivers/net/ethernet/sfc/rx_common.c

@@ -1845,7 +1845,7 @@ static void efx_filter_rfs_work(struct work_struct *data)
 
 	/* Release references */
 	clear_bit(slot_idx, &efx->rps_slot_map);
-	dev_put(req->net_dev);
+	netdev_put(req->net_dev, &req->net_dev_tracker);
 
 	return;
 }
@@ -1908,7 +1908,8 @@ int efx_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
 	}
 
 	/* Queue the request */
-	dev_hold(req->net_dev = net_dev);
+	req->net_dev = net_dev;
+	netdev_hold(req->net_dev, &req->net_dev_tracker, GFP_ATOMIC);
 	INIT_WORK(&req->work, efx_filter_rfs_work);
 	req->rxq_index = rxq_index;
 	req->flow_id = flow_id;
@@ -1965,4 +1966,3 @@ bool __efx_filter_rfs_expire(struct efx_channel *channel, unsigned int quota)
 }
 
 #endif /* CONFIG_RFS_ACCEL */
-

src/driver/linux_net/drivers/net/ethernet/sfc/tc_encap_actions.c

@@ -209,7 +209,8 @@ static int efx_bind_neigh(struct efx_nic *efx,
 				EFX_TC_ERR_MSG(efx, extack, "Failed to lookup route for encap");
 				goto out_free;
 			}
-			dev_hold(neigh->egdev = dst->dev);
+			netdev_hold(neigh->egdev = dst->dev, &neigh->dev_tracker,
+				    GFP_KERNEL_ACCOUNT);
 			neigh->ttl = ip6_dst_hoplimit(dst);
 			n = dst_neigh_lookup(dst, &flow6.daddr);
 			dst_release(dst);
@@ -223,7 +224,8 @@ static int efx_bind_neigh(struct efx_nic *efx,
 				EFX_TC_ERR_MSG(efx, extack, "Failed to lookup route for encap");
 				goto out_free;
 			}
-			dev_hold(neigh->egdev = rt->dst.dev);
+			netdev_hold(neigh->egdev = rt->dst.dev, &neigh->dev_tracker,
+				    GFP_KERNEL_ACCOUNT);
 			neigh->ttl = ip4_dst_hoplimit(&rt->dst);
 			n = dst_neigh_lookup(&rt->dst, &flow4.daddr);
 			ip_rt_put(rt);
@@ -233,7 +235,7 @@ static int efx_bind_neigh(struct efx_nic *efx,
 		if (!n) {
 			rc = -ENETUNREACH;
 			EFX_TC_ERR_MSG(efx, extack, "Failed to lookup neighbour for encap");
-			dev_put(neigh->egdev);
+			netdev_put(neigh->egdev, &neigh->dev_tracker);
 			goto out_free;
 		}
 		refcount_set(&neigh->ref, 1);
@@ -273,7 +275,7 @@ static void efx_free_neigh(struct efx_neigh_binder *neigh)
 	rhashtable_remove_fast(&efx->tc->neigh_ht, &neigh->linkage,
 			       efx_neigh_ht_params);
 	synchronize_rcu();
-	dev_put(neigh->egdev);
+	netdev_put(neigh->egdev, &neigh->dev_tracker);
 	put_net(neigh->net);
 	kfree(neigh);
 }

src/driver/linux_net/drivers/net/ethernet/sfc/tc_encap_actions.h

@@ -26,6 +26,7 @@
  * @ttl: Time To Live associated with the route used
  * @dying: set when egdev is going away, to skip further updates
  * @egdev: egress device from the route lookup.  Holds a reference
+ * @dev_tracker: reference tracker entry for @egdev
  * @ref: counts encap actions referencing this entry
  * @used: jiffies of last time traffic hit any encap action using this.
  *      When counter reads update this, a new neighbour event is sent to
@@ -53,6 +54,7 @@ struct efx_neigh_binder {
 	u8 ttl;
 	bool dying;
 	struct net_device *egdev;
+	netdevice_tracker dev_tracker;
 	refcount_t ref;
 	unsigned long used;
 	struct list_head users;

src/driver/linux_resource/kernel_compat.sh

@@ -178,6 +178,9 @@ EFRM_HAVE_FOLLOW_PTE_VMA symtype follow_pte include/linux/mm.h int(struct vm_are
 
 EFRM_HAVE_LINUX_TPH_H 			file	include/linux/pci-tph.h
 
+EFX_NEED_NETDEV_HOLD		nsymbol	netdev_hold		include/linux/netdevice.h
+EFX_HAVE_DEV_HOLD_TRACK		symbol	dev_hold_track		include/linux/netdevice.h
+
 # TODO move onload-related stuff from net kernel_compat
 " | grep -E -v -e '^#' -e '^$' | sed 's/[ \t][ \t]*/:/g'
 }

src/driver/linux_resource/nondl_resource.c

@@ -8,6 +8,7 @@
 #include <ci/efrm/efrm_client.h>
 #include <ci/efrm/nondl.h>
 #include <linux/rtnetlink.h>
+#include <linux/netdevice.h>
 
 /* Maximum number of VIs we will try to create for each device. */
 #define MAX_VIS 128
@@ -94,6 +95,9 @@ void efrm_nondl_unregister_driver(struct efrm_nondl_driver *driver)
 
   rtnl_lock();
 
+  if (!nondl_driver)
+    goto out;
+
   EFRM_ASSERT(nondl_driver == driver);
 
   list_for_each_entry_safe_reverse(device, device_n, &driver->devices,
@@ -104,6 +108,7 @@ void efrm_nondl_unregister_driver(struct efrm_nondl_driver *driver)
 
   nondl_driver = NULL;
 
+out:
   rtnl_unlock();
 }
 EXPORT_SYMBOL(efrm_nondl_unregister_driver);
@@ -132,7 +137,7 @@ int efrm_nondl_register_netdev(struct net_device *netdev,
 
   INIT_LIST_HEAD(&device->node);
 
-  dev_hold(netdev);
+  netdev_hold(netdev, &device->netdev_tracker, GFP_KERNEL);
   device->netdev = netdev;
   device->n_vis = n_vis;
   device->is_up = 1;
@@ -153,7 +158,7 @@ static void efrm_nondl_cleanup_netdev(struct efrm_nondl_device *device)
   BUG_ON(device->driver);
 
   list_del(&device->node);
-  dev_put(device->netdev);
+  netdev_put(device->netdev, &device->netdev_tracker);
 
   kfree(device);
 }

src/driver/linux_resource/resource_driver.c

@@ -69,6 +69,7 @@
 #include "sfcaffinity.h"
 #include <ci/driver/resource/linux_efhw_nic.h>
 #include <ci/driver/resource/driverlink.h>
+#include <linux/reboot.h>
 #include "debugfs.h"
 
 MODULE_AUTHOR("Solarflare Communications");
@@ -206,7 +207,7 @@ linux_efrm_nic_ctor(struct linux_efhw_nic *lnic, struct device *dev,
 	/* Tie the lifetime of the kernel's state to that of our own. */
 	if( dev )
 		get_device(dev);
-	dev_hold(net_dev);
+	netdev_hold(net_dev, &nic->net_dev_tracker, GFP_KERNEL);
 
 	rc = efhw_nic_ctor(nic, res_dim, dev_type, net_dev, dev);
 	if (rc < 0)
@@ -236,7 +237,7 @@ linux_efrm_nic_ctor(struct linux_efhw_nic *lnic, struct device *dev,
 fail1:
 	if( dev )
 		put_device(dev);
-	dev_put(net_dev);
+	netdev_put(net_dev, &nic->net_dev_tracker);
 	return rc;
 }
 
@@ -259,7 +260,7 @@ linux_efrm_nic_reclaim(struct linux_efhw_nic *lnic,
 
 	/* Replace the net & pci devs */
 	get_device(dev);
-	dev_hold(net_dev);
+	netdev_hold(net_dev, &nic->net_dev_tracker, GFP_KERNEL);
 	spin_lock_bh(&nic->pci_dev_lock);
 	old_dev = nic->dev;
 	nic->dev = dev;
@@ -506,7 +507,7 @@ efrm_nic_do_unplug(struct efhw_nic* nic, bool hard)
 	spin_unlock_bh(&nic->pci_dev_lock);
 
 	EFRM_ASSERT(net_dev != NULL);
-	dev_put(net_dev);
+	netdev_put(net_dev, &nic->net_dev_tracker);
 	put_device(dev);
 
 	return 0;
@@ -631,6 +632,21 @@ static void efrm_nic_del_all(void)
 		efrm_nic_del(linux_efhw_nic(nic));
 }
 
+static int sfc_resource_shutdown_notify(struct notifier_block *unused1,
+				        unsigned long unused2, void *unused3)
+{
+	/* Due to refcounting reasons in netdev, these must
+	 * be called for the shutdown to happen without delays
+	 */
+	efrm_nondl_unregister();
+	efrm_nondl_shutdown();
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block sfc_resource_shutdown_nb = {
+	.notifier_call = sfc_resource_shutdown_notify,
+};
 
 /****************************************************************************
  *
@@ -683,6 +699,8 @@ static int init_sfc_resource(void)
 	efrm_install_sysfs_entries();
 	efrm_nondl_register();
 
+	register_reboot_notifier(&sfc_resource_shutdown_nb);
+
 	return 0;
 
 failed_notifier:
@@ -705,6 +723,8 @@ static int init_sfc_resource(void)
  ****************************************************************************/
 static void cleanup_sfc_resource(void)
 {
+	unregister_reboot_notifier(&sfc_resource_shutdown_nb);
+
 	efrm_nondl_unregister();
 	efrm_remove_sysfs_entries();
 	efrm_nondl_shutdown();

src/include/ci/driver/kernel_compat.h

@@ -546,4 +546,30 @@ static inline int efrm_follow_pfn(struct vm_area_struct *vma,
 }
 #endif
 
+#ifdef EFX_NEED_NETDEV_HOLD
+#ifdef EFX_HAVE_DEV_HOLD_TRACK
+/* Commit d62607c3fe45 ("net: rename reference+tracking helpers")
+ * renamed these.
+ */
+#define netdev_hold(_n, _t, _g)	dev_hold_track(_n, _t, _g)
+#define netdev_put(_n, _t)	dev_put_track(_n, _t)
+#else
+/* This was introduced in the same commit that adds dev_hold_track */
+typedef struct {} netdevice_tracker;
+
+static inline void netdev_hold(struct net_device *dev,
+			       netdevice_tracker *tracker __always_unused,
+			       gfp_t gfp __always_unused)
+{
+	dev_hold(dev);
+}
+
+static inline void netdev_put(struct net_device *dev,
+			      netdevice_tracker *tracker __always_unused)
+{
+	dev_put(dev);
+}
+#endif
+#endif
+
 #endif /* DRIVER_LINUX_RESOURCE_KERNEL_COMPAT_H */

src/include/ci/efhw/efhw_types.h

@@ -40,6 +40,7 @@
 #ifndef __CI_EFHW_EFAB_TYPES_H__
 #define __CI_EFHW_EFAB_TYPES_H__
 
+#include <ci/driver/kernel_compat.h>
 #include <ci/efhw/efhw_config.h>
 #include <ci/efhw/hardware_sysdep.h>
 #include <ci/efhw/iopage_types.h>
@@ -569,6 +570,7 @@ struct efhw_nic {
 	int index;
 
 	struct net_device *net_dev; /*!< Network device */
+	netdevice_tracker net_dev_tracker;
 	struct device *dev;         /*!< HW device */
 	struct pci_dev *pci_dev;    /*!< PCI device */
 	spinlock_t pci_dev_lock;    /*!< Protects access to dev & net_dev */

src/include/ci/efrm/nondl.h

@@ -8,6 +8,7 @@
 #define __EFRM_NONDL_H__
 
 #include <linux/types.h>
+#include <ci/driver/kernel_compat.h>
 
 
 /* Non-driverlink network device.
@@ -43,6 +44,7 @@ struct efrm_nondl_device {
   /* Network device currently associated with this non-driverlink
    * device. */
   struct net_device *netdev;
+  netdevice_tracker netdev_tracker;
 
   /* Number of VIs we would like to create on this device. */
   unsigned int n_vis;