Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next

Pull networking updates from Davic Miller:

 1) Support busy polling generically, for all NAPI drivers.  From Eric
    Dumazet.

 2) Add byte/packet counter support to nft_ct, from Floriani Westphal.

 3) Add RSS/XPS support to mvneta driver, from Gregory Clement.

 4) Implement IPV6_HDRINCL socket option for raw sockets, from Hannes
    Frederic Sowa.

 5) Add support for T6 adapter to cxgb4 driver, from Hariprasad Shenai.

 6) Add support for VLAN device bridging to mlxsw switch driver, from
    Ido Schimmel.

 7) Add driver for Netronome NFP4000/NFP6000, from Jakub Kicinski.

 8) Provide hwmon interface to mlxsw switch driver, from Jiri Pirko.

 9) Reorganize wireless drivers into per-vendor directories just like we
    do for ethernet drivers.  From Kalle Valo.

10) Provide a way for administrators "destroy" connected sockets via the
    SOCK_DESTROY socket netlink diag operation.  From Lorenzo Colitti.

11) Add support to add/remove multicast routes via netlink, from Nikolay
    Aleksandrov.

12) Make TCP keepalive settings per-namespace, from Nikolay Borisov.

13) Add forwarding and packet duplication facilities to nf_tables, from
    Pablo Neira Ayuso.

14) Dead route support in MPLS, from Roopa Prabhu.

15) TSO support for thunderx chips, from Sunil Goutham.

16) Add driver for IBM's System i/p VNIC protocol, from Thomas Falcon.

17) Rationalize, consolidate, and more completely document the checksum
    offloading facilities in the networking stack.  From Tom Herbert.

18) Support aborting an ongoing scan in mac80211/cfg80211, from
    Vidyullatha Kanchanapally.

19) Use per-bucket spinlock for bpf hash facility, from Tom Leiming.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1375 commits)
  net: bnxt: always return values from _bnxt_get_max_rings
  net: bpf: reject invalid shifts
  phonet: properly unshare skbs in phonet_rcv()
  dwc_eth_qos: Fix dma address for multi-fragment skbs
  phy: remove an unneeded condition
  mdio: remove an unneed condition
  mdio_bus: NULL dereference on allocation error
  net: Fix typo in netdev_intersect_features
  net: freescale: mac-fec: Fix build error from phy_device API change
  net: freescale: ucc_geth: Fix build error from phy_device API change
  bonding: Prevent IPv6 link local address on enslaved devices
  IB/mlx5: Add flow steering support
  net/mlx5_core: Export flow steering API
  net/mlx5_core: Make ipv4/ipv6 location more clear
  net/mlx5_core: Enable flow steering support for the IB driver
  net/mlx5_core: Initialize namespaces only when supported by device
  net/mlx5_core: Set priority attributes
  net/mlx5_core: Connect flow tables
  net/mlx5_core: Introduce modify flow table command
  net/mlx5_core: Managing root flow table
  ...

1 files changed, 130 insertions, 49 deletions

diff --git a/drivers/net/ethernet/chelsio/cxgb4/l2t.c b/drivers/net/ethernet/chelsio/cxgb4/l2t.c
index ac27898c6ab0..5b0f3ef348e9 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/l2t.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/l2t.c
@@ -66,7 +66,7 @@ struct l2t_data {
 
 static inline unsigned int vlan_prio(const struct l2t_entry *e)
 {
-	return e->vlan >> 13;
+	return e->vlan >> VLAN_PRIO_SHIFT;
 }
 
 static inline void l2t_hold(struct l2t_data *d, struct l2t_entry *e)
@@ -161,8 +161,7 @@ static int write_l2e(struct adapter *adap, struct l2t_entry *e, int sync)
 		memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
 	memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
 
-	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
-	t4_ofld_send(adap, skb);
+	t4_mgmt_tx(adap, skb);
 
 	if (sync && e->state != L2T_STATE_SWITCHING)
 		e->state = L2T_STATE_SYNC_WRITE;
@@ -175,14 +174,10 @@ static int write_l2e(struct adapter *adap, struct l2t_entry *e, int sync)
  */
 static void send_pending(struct adapter *adap, struct l2t_entry *e)
 {
-	while (e->arpq_head) {
-		struct sk_buff *skb = e->arpq_head;
+	struct sk_buff *skb;
 
-		e->arpq_head = skb->next;
-		skb->next = NULL;
+	while ((skb = __skb_dequeue(&e->arpq)) != NULL)
 		t4_ofld_send(adap, skb);
-	}
-	e->arpq_tail = NULL;
 }
 
 /*
@@ -222,12 +217,7 @@ void do_l2t_write_rpl(struct adapter *adap, const struct cpl_l2t_write_rpl *rpl)
  */
 static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
 {
-	skb->next = NULL;
-	if (e->arpq_head)
-		e->arpq_tail->next = skb;
-	else
-		e->arpq_head = skb;
-	e->arpq_tail = skb;
+	__skb_queue_tail(&e->arpq, skb);
 }
 
 int cxgb4_l2t_send(struct net_device *dev, struct sk_buff *skb,
@@ -259,7 +249,8 @@ again:
 		if (e->state == L2T_STATE_RESOLVING &&
 		    !neigh_event_send(e->neigh, NULL)) {
 			spin_lock_bh(&e->lock);
-			if (e->state == L2T_STATE_RESOLVING && e->arpq_head)
+			if (e->state == L2T_STATE_RESOLVING &&
+			    !skb_queue_empty(&e->arpq))
 				write_l2e(adap, e, 1);
 			spin_unlock_bh(&e->lock);
 		}
@@ -305,12 +296,82 @@ found:
 	return e;
 }
 
-/*
- * Called when an L2T entry has no more users.
+static struct l2t_entry *find_or_alloc_l2e(struct l2t_data *d, u16 vlan,
+					   u8 port, u8 *dmac)
+{
+	struct l2t_entry *end, *e, **p;
+	struct l2t_entry *first_free = NULL;
+
+	for (e = &d->l2tab[0], end = &d->l2tab[d->l2t_size]; e != end; ++e) {
+		if (atomic_read(&e->refcnt) == 0) {
+			if (!first_free)
+				first_free = e;
+		} else {
+			if (e->state == L2T_STATE_SWITCHING) {
+				if (ether_addr_equal(e->dmac, dmac) &&
+				    (e->vlan == vlan) && (e->lport == port))
+					goto exists;
+			}
+		}
+	}
+
+	if (first_free) {
+		e = first_free;
+		goto found;
+	}
+
+	return NULL;
+
+found:
+	/* The entry we found may be an inactive entry that is
+	 * presently in the hash table.  We need to remove it.
+	 */
+	if (e->state < L2T_STATE_SWITCHING)
+		for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
+			if (*p == e) {
+				*p = e->next;
+				e->next = NULL;
+				break;
+			}
+	e->state = L2T_STATE_UNUSED;
+
+exists:
+	return e;
+}
+
+/* Called when an L2T entry has no more users.  The entry is left in the hash
+ * table since it is likely to be reused but we also bump nfree to indicate
+ * that the entry can be reallocated for a different neighbor.  We also drop
+ * the existing neighbor reference in case the neighbor is going away and is
+ * waiting on our reference.
+ *
+ * Because entries can be reallocated to other neighbors once their ref count
+ * drops to 0 we need to take the entry's lock to avoid races with a new
+ * incarnation.
  */
+static void _t4_l2e_free(struct l2t_entry *e)
+{
+	struct l2t_data *d;
+	struct sk_buff *skb;
+
+	if (atomic_read(&e->refcnt) == 0) {  /* hasn't been recycled */
+		if (e->neigh) {
+			neigh_release(e->neigh);
+			e->neigh = NULL;
+		}
+		while ((skb = __skb_dequeue(&e->arpq)) != NULL)
+			kfree_skb(skb);
+	}
+
+	d = container_of(e, struct l2t_data, l2tab[e->idx]);
+	atomic_inc(&d->nfree);
+}
+
+/* Locked version of _t4_l2e_free */
 static void t4_l2e_free(struct l2t_entry *e)
 {
 	struct l2t_data *d;
+	struct sk_buff *skb;
 
 	spin_lock_bh(&e->lock);
 	if (atomic_read(&e->refcnt) == 0) {  /* hasn't been recycled */
@@ -318,13 +379,8 @@ static void t4_l2e_free(struct l2t_entry *e)
 			neigh_release(e->neigh);
 			e->neigh = NULL;
 		}
-		while (e->arpq_head) {
-			struct sk_buff *skb = e->arpq_head;
-
-			e->arpq_head = skb->next;
+		while ((skb = __skb_dequeue(&e->arpq)) != NULL)
 			kfree_skb(skb);
-		}
-		e->arpq_tail = NULL;
 	}
 	spin_unlock_bh(&e->lock);
 
@@ -457,18 +513,19 @@ EXPORT_SYMBOL(cxgb4_select_ntuple);
  * on the arpq head.  If a packet specifies a failure handler it is invoked,
  * otherwise the packet is sent to the device.
  */
-static void handle_failed_resolution(struct adapter *adap, struct sk_buff *arpq)
+static void handle_failed_resolution(struct adapter *adap, struct l2t_entry *e)
 {
-	while (arpq) {
-		struct sk_buff *skb = arpq;
+	struct sk_buff *skb;
+
+	while ((skb = __skb_dequeue(&e->arpq)) != NULL) {
 		const struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
 
-		arpq = skb->next;
-		skb->next = NULL;
+		spin_unlock(&e->lock);
 		if (cb->arp_err_handler)
 			cb->arp_err_handler(cb->handle, skb);
 		else
 			t4_ofld_send(adap, skb);
+		spin_lock(&e->lock);
 	}
 }
 
@@ -479,7 +536,7 @@ static void handle_failed_resolution(struct adapter *adap, struct sk_buff *arpq)
 void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
 {
 	struct l2t_entry *e;
-	struct sk_buff *arpq = NULL;
+	struct sk_buff_head *arpq = NULL;
 	struct l2t_data *d = adap->l2t;
 	int addr_len = neigh->tbl->key_len;
 	u32 *addr = (u32 *) neigh->primary_key;
@@ -506,10 +563,9 @@ void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
 
 	if (e->state == L2T_STATE_RESOLVING) {
 		if (neigh->nud_state & NUD_FAILED) {
-			arpq = e->arpq_head;
-			e->arpq_head = e->arpq_tail = NULL;
+			arpq = &e->arpq;
 		} else if ((neigh->nud_state & (NUD_CONNECTED | NUD_STALE)) &&
-			   e->arpq_head) {
+			   !skb_queue_empty(&e->arpq)) {
 			write_l2e(adap, e, 1);
 		}
 	} else {
@@ -519,43 +575,66 @@ void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
 			write_l2e(adap, e, 0);
 	}
 
-	spin_unlock_bh(&e->lock);
-
 	if (arpq)
-		handle_failed_resolution(adap, arpq);
+		handle_failed_resolution(adap, e);
+	spin_unlock_bh(&e->lock);
 }
 
 /* Allocate an L2T entry for use by a switching rule.  Such need to be
  * explicitly freed and while busy they are not on any hash chain, so normal
  * address resolution updates do not see them.
  */
-struct l2t_entry *t4_l2t_alloc_switching(struct l2t_data *d)
+struct l2t_entry *t4_l2t_alloc_switching(struct adapter *adap, u16 vlan,
+					 u8 port, u8 *eth_addr)
 {
+	struct l2t_data *d = adap->l2t;
 	struct l2t_entry *e;
+	int ret;
 
 	write_lock_bh(&d->lock);
-	e = alloc_l2e(d);
+	e = find_or_alloc_l2e(d, vlan, port, eth_addr);
 	if (e) {
 		spin_lock(&e->lock);          /* avoid race with t4_l2t_free */
-		e->state = L2T_STATE_SWITCHING;
-		atomic_set(&e->refcnt, 1);
+		if (!atomic_read(&e->refcnt)) {
+			e->state = L2T_STATE_SWITCHING;
+			e->vlan = vlan;
+			e->lport = port;
+			ether_addr_copy(e->dmac, eth_addr);
+			atomic_set(&e->refcnt, 1);
+			ret = write_l2e(adap, e, 0);
+			if (ret < 0) {
+				_t4_l2e_free(e);
+				spin_unlock(&e->lock);
+				write_unlock_bh(&d->lock);
+				return NULL;
+			}
+		} else {
+			atomic_inc(&e->refcnt);
+		}
+
 		spin_unlock(&e->lock);
 	}
 	write_unlock_bh(&d->lock);
 	return e;
 }
 
-/* Sets/updates the contents of a switching L2T entry that has been allocated
- * with an earlier call to @t4_l2t_alloc_switching.
+/**
+ * @dev: net_device pointer
+ * @vlan: VLAN Id
+ * @port: Associated port
+ * @dmac: Destination MAC address to add to L2T
+ * Returns pointer to the allocated l2t entry
+ *
+ * Allocates an L2T entry for use by switching rule of a filter
  */
-int t4_l2t_set_switching(struct adapter *adap, struct l2t_entry *e, u16 vlan,
-		u8 port, u8 *eth_addr)
+struct l2t_entry *cxgb4_l2t_alloc_switching(struct net_device *dev, u16 vlan,
+					    u8 port, u8 *dmac)
 {
-	e->vlan = vlan;
-	e->lport = port;
-	memcpy(e->dmac, eth_addr, ETH_ALEN);
-	return write_l2e(adap, e, 0);
+	struct adapter *adap = netdev2adap(dev);
+
+	return t4_l2t_alloc_switching(adap, vlan, port, dmac);
 }
+EXPORT_SYMBOL(cxgb4_l2t_alloc_switching);
 
 struct l2t_data *t4_init_l2t(unsigned int l2t_start, unsigned int l2t_end)
 {
@@ -585,6 +664,7 @@ struct l2t_data *t4_init_l2t(unsigned int l2t_start, unsigned int l2t_end)
 		d->l2tab[i].state = L2T_STATE_UNUSED;
 		spin_lock_init(&d->l2tab[i].lock);
 		atomic_set(&d->l2tab[i].refcnt, 0);
+		skb_queue_head_init(&d->l2tab[i].arpq);
 	}
 	return d;
 }
@@ -619,7 +699,8 @@ static char l2e_state(const struct l2t_entry *e)
 	case L2T_STATE_VALID: return 'V';
 	case L2T_STATE_STALE: return 'S';
 	case L2T_STATE_SYNC_WRITE: return 'W';
-	case L2T_STATE_RESOLVING: return e->arpq_head ? 'A' : 'R';
+	case L2T_STATE_RESOLVING:
+		return skb_queue_empty(&e->arpq) ? 'R' : 'A';
 	case L2T_STATE_SWITCHING: return 'X';
 	default:
 		return 'U';