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

Pull networking updates from David Miller:
 "Highlights:

   1) Steady transitioning of the BPF instructure to a generic spot so
      all kernel subsystems can make use of it, from Alexei Starovoitov.

   2) SFC driver supports busy polling, from Alexandre Rames.

   3) Take advantage of hash table in UDP multicast delivery, from David
      Held.

   4) Lighten locking, in particular by getting rid of the LRU lists, in
      inet frag handling.  From Florian Westphal.

   5) Add support for various RFC6458 control messages in SCTP, from
      Geir Ola Vaagland.

   6) Allow to filter bridge forwarding database dumps by device, from
      Jamal Hadi Salim.

   7) virtio-net also now supports busy polling, from Jason Wang.

   8) Some low level optimization tweaks in pktgen from Jesper Dangaard
      Brouer.

   9) Add support for ipv6 address generation modes, so that userland
      can have some input into the process.  From Jiri Pirko.

  10) Consolidate common TCP connection request code in ipv4 and ipv6,
      from Octavian Purdila.

  11) New ARP packet logger in netfilter, from Pablo Neira Ayuso.

  12) Generic resizable RCU hash table, with intial users in netlink and
      nftables.  From Thomas Graf.

  13) Maintain a name assignment type so that userspace can see where a
      network device name came from (enumerated by kernel, assigned
      explicitly by userspace, etc.) From Tom Gundersen.

  14) Automatic flow label generation on transmit in ipv6, from Tom
      Herbert.

  15) New packet timestamping facilities from Willem de Bruijn, meant to
      assist in measuring latencies going into/out-of the packet
      scheduler, latency from TCP data transmission to ACK, etc"

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1536 commits)
  cxgb4 : Disable recursive mailbox commands when enabling vi
  net: reduce USB network driver config options.
  tg3: Modify tg3_tso_bug() to handle multiple TX rings
  amd-xgbe: Perform phy connect/disconnect at dev open/stop
  amd-xgbe: Use dma_set_mask_and_coherent to set DMA mask
  net: sun4i-emac: fix memory leak on bad packet
  sctp: fix possible seqlock seadlock in sctp_packet_transmit()
  Revert "net: phy: Set the driver when registering an MDIO bus device"
  cxgb4vf: Turn off SGE RX/TX Callback Timers and interrupts in PCI shutdown routine
  team: Simplify return path of team_newlink
  bridge: Update outdated comment on promiscuous mode
  net-timestamp: ACK timestamp for bytestreams
  net-timestamp: TCP timestamping
  net-timestamp: SCHED timestamp on entering packet scheduler
  net-timestamp: add key to disambiguate concurrent datagrams
  net-timestamp: move timestamp flags out of sk_flags
  net-timestamp: extend SCM_TIMESTAMPING ancillary data struct
  cxgb4i : Move stray CPL definitions to cxgb4 driver
  tcp: reduce spurious retransmits due to transient SACK reneging
  qlcnic: Initialize dcbnl_ops before register_netdev
  ...

1 files changed, 95 insertions, 15 deletions

diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c
index c65b636bcab9..aa2093325be1 100644
--- a/drivers/net/xen-netback/netback.c
+++ b/drivers/net/xen-netback/netback.c
@@ -1869,8 +1869,7 @@ void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
 static inline int rx_work_todo(struct xenvif_queue *queue)
 {
 	return (!skb_queue_empty(&queue->rx_queue) &&
-	       xenvif_rx_ring_slots_available(queue, queue->rx_last_skb_slots)) ||
-	       queue->rx_queue_purge;
+	       xenvif_rx_ring_slots_available(queue, queue->rx_last_skb_slots));
 }
 
 static inline int tx_work_todo(struct xenvif_queue *queue)
@@ -1935,6 +1934,75 @@ static void xenvif_start_queue(struct xenvif_queue *queue)
 		xenvif_wake_queue(queue);
 }
 
+/* Only called from the queue's thread, it handles the situation when the guest
+ * doesn't post enough requests on the receiving ring.
+ * First xenvif_start_xmit disables QDisc and start a timer, and then either the
+ * timer fires, or the guest send an interrupt after posting new request. If it
+ * is the timer, the carrier is turned off here.
+ * */
+static void xenvif_rx_purge_event(struct xenvif_queue *queue)
+{
+	/* Either the last unsuccesful skb or at least 1 slot should fit */
+	int needed = queue->rx_last_skb_slots ?
+		     queue->rx_last_skb_slots : 1;
+
+	/* It is assumed that if the guest post new slots after this, the RX
+	 * interrupt will set the QUEUE_STATUS_RX_PURGE_EVENT bit and wake up
+	 * the thread again
+	 */
+	set_bit(QUEUE_STATUS_RX_STALLED, &queue->status);
+	if (!xenvif_rx_ring_slots_available(queue, needed)) {
+		rtnl_lock();
+		if (netif_carrier_ok(queue->vif->dev)) {
+			/* Timer fired and there are still no slots. Turn off
+			 * everything except the interrupts
+			 */
+			netif_carrier_off(queue->vif->dev);
+			skb_queue_purge(&queue->rx_queue);
+			queue->rx_last_skb_slots = 0;
+			if (net_ratelimit())
+				netdev_err(queue->vif->dev, "Carrier off due to lack of guest response on queue %d\n", queue->id);
+		} else {
+			/* Probably an another queue already turned the carrier
+			 * off, make sure nothing is stucked in the internal
+			 * queue of this queue
+			 */
+			skb_queue_purge(&queue->rx_queue);
+			queue->rx_last_skb_slots = 0;
+		}
+		rtnl_unlock();
+	} else if (!netif_carrier_ok(queue->vif->dev)) {
+		unsigned int num_queues = queue->vif->num_queues;
+		unsigned int i;
+		/* The carrier was down, but an interrupt kicked
+		 * the thread again after new requests were
+		 * posted
+		 */
+		clear_bit(QUEUE_STATUS_RX_STALLED,
+			  &queue->status);
+		rtnl_lock();
+		netif_carrier_on(queue->vif->dev);
+		netif_tx_wake_all_queues(queue->vif->dev);
+		rtnl_unlock();
+
+		for (i = 0; i < num_queues; i++) {
+			struct xenvif_queue *temp = &queue->vif->queues[i];
+
+			xenvif_napi_schedule_or_enable_events(temp);
+		}
+		if (net_ratelimit())
+			netdev_err(queue->vif->dev, "Carrier on again\n");
+	} else {
+		/* Queuing were stopped, but the guest posted
+		 * new requests and sent an interrupt
+		 */
+		clear_bit(QUEUE_STATUS_RX_STALLED,
+			  &queue->status);
+		del_timer_sync(&queue->rx_stalled);
+		xenvif_start_queue(queue);
+	}
+}
+
 int xenvif_kthread_guest_rx(void *data)
 {
 	struct xenvif_queue *queue = data;
@@ -1944,8 +2012,12 @@ int xenvif_kthread_guest_rx(void *data)
 		wait_event_interruptible(queue->wq,
 					 rx_work_todo(queue) ||
 					 queue->vif->disabled ||
+					 test_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status) ||
 					 kthread_should_stop());
 
+		if (kthread_should_stop())
+			break;
+
 		/* This frontend is found to be rogue, disable it in
 		 * kthread context. Currently this is only set when
 		 * netback finds out frontend sends malformed packet,
@@ -1953,26 +2025,23 @@ int xenvif_kthread_guest_rx(void *data)
 		 * context so we defer it here, if this thread is
 		 * associated with queue 0.
 		 */
-		if (unlikely(queue->vif->disabled && netif_carrier_ok(queue->vif->dev) && queue->id == 0))
+		if (unlikely(queue->vif->disabled && queue->id == 0))
 			xenvif_carrier_off(queue->vif);
-
-		if (kthread_should_stop())
-			break;
-
-		if (queue->rx_queue_purge) {
+		else if (unlikely(test_and_clear_bit(QUEUE_STATUS_RX_PURGE_EVENT,
+						     &queue->status))) {
+			xenvif_rx_purge_event(queue);
+		} else if (!netif_carrier_ok(queue->vif->dev)) {
+			/* Another queue stalled and turned the carrier off, so
+			 * purge the internal queue of queues which were not
+			 * blocked
+			 */
 			skb_queue_purge(&queue->rx_queue);
-			queue->rx_queue_purge = false;
+			queue->rx_last_skb_slots = 0;
 		}
 
 		if (!skb_queue_empty(&queue->rx_queue))
 			xenvif_rx_action(queue);
 
-		if (skb_queue_empty(&queue->rx_queue) &&
-		    xenvif_queue_stopped(queue)) {
-			del_timer_sync(&queue->wake_queue);
-			xenvif_start_queue(queue);
-		}
-
 		cond_resched();
 	}
 
@@ -2027,6 +2096,13 @@ static int __init netback_init(void)
 
 	rx_drain_timeout_jiffies = msecs_to_jiffies(rx_drain_timeout_msecs);
 
+#ifdef CONFIG_DEBUG_FS
+	xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL);
+	if (IS_ERR_OR_NULL(xen_netback_dbg_root))
+		pr_warn("Init of debugfs returned %ld!\n",
+			PTR_ERR(xen_netback_dbg_root));
+#endif /* CONFIG_DEBUG_FS */
+
 	return 0;
 
 failed_init:
@@ -2037,6 +2113,10 @@ module_init(netback_init);
 
 static void __exit netback_fini(void)
 {
+#ifdef CONFIG_DEBUG_FS
+	if (!IS_ERR_OR_NULL(xen_netback_dbg_root))
+		debugfs_remove_recursive(xen_netback_dbg_root);
+#endif /* CONFIG_DEBUG_FS */
 	xenvif_xenbus_fini();
 }
 module_exit(netback_fini);