6 files changed, 325 insertions, 97 deletions

diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index 87310ee..285c143 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -132,12 +132,46 @@ struct ipoib_cm_data {
 	__be32 mtu;
 };
 
+/*
+ * Quoting 10.3.1 Queue Pair and EE Context States:
+ *
+ * Note, for QPs that are associated with an SRQ, the Consumer should take the
+ * QP through the Error State before invoking a Destroy QP or a Modify QP to the
+ * Reset State.  The Consumer may invoke the Destroy QP without first performing
+ * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
+ * Last WQE Reached Event. However, if the Consumer does not wait for the
+ * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
+ * leakage may occur. Therefore, it is good programming practice to tear down a
+ * QP that is associated with an SRQ by using the following process:
+ *
+ * - Put the QP in the Error State
+ * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
+ * - either:
+ *       drain the CQ by invoking the Poll CQ verb and either wait for CQ
+ *       to be empty or the number of Poll CQ operations has exceeded
+ *       CQ capacity size;
+ * - or
+ *       post another WR that completes on the same CQ and wait for this
+ *       WR to return as a WC;
+ * - and then invoke a Destroy QP or Reset QP.
+ *
+ * We use the second option and wait for a completion on the
+ * same CQ before destroying QPs attached to our SRQ.
+ */
+
+enum ipoib_cm_state {
+	IPOIB_CM_RX_LIVE,
+	IPOIB_CM_RX_ERROR, /* Ignored by stale task */
+	IPOIB_CM_RX_FLUSH  /* Last WQE Reached event observed */
+};
+
 struct ipoib_cm_rx {
 	struct ib_cm_id     *id;
 	struct ib_qp        *qp;
 	struct list_head     list;
 	struct net_device   *dev;
 	unsigned long        jiffies;
+	enum ipoib_cm_state  state;
 };
 
 struct ipoib_cm_tx {
@@ -165,10 +199,15 @@ struct ipoib_cm_dev_priv {
 	struct ib_srq  	       *srq;
 	struct ipoib_cm_rx_buf *srq_ring;
 	struct ib_cm_id        *id;
-	struct list_head        passive_ids;
+	struct list_head        passive_ids;   /* state: LIVE */
+	struct list_head        rx_error_list; /* state: ERROR */
+	struct list_head        rx_flush_list; /* state: FLUSH, drain not started */
+	struct list_head        rx_drain_list; /* state: FLUSH, drain started */
+	struct list_head        rx_reap_list;  /* state: FLUSH, drain done */
 	struct work_struct      start_task;
 	struct work_struct      reap_task;
 	struct work_struct      skb_task;
+	struct work_struct      rx_reap_task;
 	struct delayed_work     stale_task;
 	struct sk_buff_head     skb_queue;
 	struct list_head        start_list;
@@ -201,15 +240,17 @@ struct ipoib_dev_priv {
 	struct list_head multicast_list;
 	struct rb_root multicast_tree;
 
-	struct delayed_work pkey_task;
+	struct delayed_work pkey_poll_task;
 	struct delayed_work mcast_task;
 	struct work_struct flush_task;
 	struct work_struct restart_task;
 	struct delayed_work ah_reap_task;
+	struct work_struct pkey_event_task;
 
 	struct ib_device *ca;
 	u8            	  port;
 	u16           	  pkey;
+	u16               pkey_index;
 	struct ib_pd  	 *pd;
 	struct ib_mr  	 *mr;
 	struct ib_cq  	 *cq;
@@ -333,12 +374,13 @@ struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
 
 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
 void ipoib_ib_dev_flush(struct work_struct *work);
+void ipoib_pkey_event(struct work_struct *work);
 void ipoib_ib_dev_cleanup(struct net_device *dev);
 
 int ipoib_ib_dev_open(struct net_device *dev);
 int ipoib_ib_dev_up(struct net_device *dev);
 int ipoib_ib_dev_down(struct net_device *dev, int flush);
-int ipoib_ib_dev_stop(struct net_device *dev);
+int ipoib_ib_dev_stop(struct net_device *dev, int flush);
 
 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
 void ipoib_dev_cleanup(struct net_device *dev);
@@ -386,6 +428,7 @@ int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
 
 void ipoib_pkey_poll(struct work_struct *work);
 int ipoib_pkey_dev_delay_open(struct net_device *dev);
+void ipoib_drain_cq(struct net_device *dev);
 
 #ifdef CONFIG_INFINIBAND_IPOIB_CM
 
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
index eec833b..076a0bb 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
@@ -37,6 +37,7 @@
 #include <net/dst.h>
 #include <net/icmp.h>
 #include <linux/icmpv6.h>
+#include <linux/delay.h>
 
 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
 static int data_debug_level;
@@ -62,6 +63,17 @@ struct ipoib_cm_id {
 	u32 remote_mtu;
 };
 
+static struct ib_qp_attr ipoib_cm_err_attr = {
+	.qp_state = IB_QPS_ERR
+};
+
+#define IPOIB_CM_RX_DRAIN_WRID 0x7fffffff
+
+static struct ib_send_wr ipoib_cm_rx_drain_wr = {
+	.wr_id = IPOIB_CM_RX_DRAIN_WRID,
+	.opcode = IB_WR_SEND,
+};
+
 static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id,
 			       struct ib_cm_event *event);
 
@@ -150,15 +162,54 @@ partial_error:
 	return NULL;
 }
 
+static void ipoib_cm_start_rx_drain(struct ipoib_dev_priv* priv)
+{
+	struct ib_send_wr *bad_wr;
+	struct ipoib_cm_rx *p;
+
+	/* We only reserved 1 extra slot in CQ for drain WRs, so
+	 * make sure we have at most 1 outstanding WR. */
+	if (list_empty(&priv->cm.rx_flush_list) ||
+	    !list_empty(&priv->cm.rx_drain_list))
+		return;
+
+	/*
+	 * QPs on flush list are error state.  This way, a "flush
+	 * error" WC will be immediately generated for each WR we post.
+	 */
+	p = list_entry(priv->cm.rx_flush_list.next, typeof(*p), list);
+	if (ib_post_send(p->qp, &ipoib_cm_rx_drain_wr, &bad_wr))
+		ipoib_warn(priv, "failed to post drain wr\n");
+
+	list_splice_init(&priv->cm.rx_flush_list, &priv->cm.rx_drain_list);
+}
+
+static void ipoib_cm_rx_event_handler(struct ib_event *event, void *ctx)
+{
+	struct ipoib_cm_rx *p = ctx;
+	struct ipoib_dev_priv *priv = netdev_priv(p->dev);
+	unsigned long flags;
+
+	if (event->event != IB_EVENT_QP_LAST_WQE_REACHED)
+		return;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	list_move(&p->list, &priv->cm.rx_flush_list);
+	p->state = IPOIB_CM_RX_FLUSH;
+	ipoib_cm_start_rx_drain(priv);
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
 static struct ib_qp *ipoib_cm_create_rx_qp(struct net_device *dev,
 					   struct ipoib_cm_rx *p)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ib_qp_init_attr attr = {
-		.send_cq = priv->cq, /* does not matter, we never send anything */
+		.event_handler = ipoib_cm_rx_event_handler,
+		.send_cq = priv->cq, /* For drain WR */
 		.recv_cq = priv->cq,
 		.srq = priv->cm.srq,
-		.cap.max_send_wr = 1, /* FIXME: 0 Seems not to work */
+		.cap.max_send_wr = 1, /* For drain WR */
 		.cap.max_send_sge = 1, /* FIXME: 0 Seems not to work */
 		.sq_sig_type = IB_SIGNAL_ALL_WR,
 		.qp_type = IB_QPT_RC,
@@ -198,6 +249,27 @@ static int ipoib_cm_modify_rx_qp(struct net_device *dev,
 		ipoib_warn(priv, "failed to modify QP to RTR: %d\n", ret);
 		return ret;
 	}
+
+	/*
+	 * Current Mellanox HCA firmware won't generate completions
+	 * with error for drain WRs unless the QP has been moved to
+	 * RTS first. This work-around leaves a window where a QP has
+	 * moved to error asynchronously, but this will eventually get
+	 * fixed in firmware, so let's not error out if modify QP
+	 * fails.
+	 */
+	qp_attr.qp_state = IB_QPS_RTS;
+	ret = ib_cm_init_qp_attr(cm_id, &qp_attr, &qp_attr_mask);
+	if (ret) {
+		ipoib_warn(priv, "failed to init QP attr for RTS: %d\n", ret);
+		return 0;
+	}
+	ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
+	if (ret) {
+		ipoib_warn(priv, "failed to modify QP to RTS: %d\n", ret);
+		return 0;
+	}
+
 	return 0;
 }
 
@@ -256,6 +328,7 @@ static int ipoib_cm_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *even
 
 	cm_id->context = p;
 	p->jiffies = jiffies;
+	p->state = IPOIB_CM_RX_LIVE;
 	spin_lock_irq(&priv->lock);
 	if (list_empty(&priv->cm.passive_ids))
 		queue_delayed_work(ipoib_workqueue,
@@ -277,7 +350,6 @@ static int ipoib_cm_rx_handler(struct ib_cm_id *cm_id,
 {
 	struct ipoib_cm_rx *p;
 	struct ipoib_dev_priv *priv;
-	int ret;
 
 	switch (event->event) {
 	case IB_CM_REQ_RECEIVED:
@@ -289,20 +361,9 @@ static int ipoib_cm_rx_handler(struct ib_cm_id *cm_id,
 	case IB_CM_REJ_RECEIVED:
 		p = cm_id->context;
 		priv = netdev_priv(p->dev);
-		spin_lock_irq(&priv->lock);
-		if (list_empty(&p->list))
-			ret = 0; /* Connection is going away already. */
-		else {
-			list_del_init(&p->list);
-			ret = -ECONNRESET;
-		}
-		spin_unlock_irq(&priv->lock);
-		if (ret) {
-			ib_destroy_qp(p->qp);
-			kfree(p);
-			return ret;
-		}
-		return 0;
+		if (ib_modify_qp(p->qp, &ipoib_cm_err_attr, IB_QP_STATE))
+			ipoib_warn(priv, "unable to move qp to error state\n");
+		/* Fall through */
 	default:
 		return 0;
 	}
@@ -354,8 +415,15 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
 		       wr_id, wc->status);
 
 	if (unlikely(wr_id >= ipoib_recvq_size)) {
-		ipoib_warn(priv, "cm recv completion event with wrid %d (> %d)\n",
-			   wr_id, ipoib_recvq_size);
+		if (wr_id == (IPOIB_CM_RX_DRAIN_WRID & ~IPOIB_CM_OP_SRQ)) {
+			spin_lock_irqsave(&priv->lock, flags);
+			list_splice_init(&priv->cm.rx_drain_list, &priv->cm.rx_reap_list);
+			ipoib_cm_start_rx_drain(priv);
+			queue_work(ipoib_workqueue, &priv->cm.rx_reap_task);
+			spin_unlock_irqrestore(&priv->lock, flags);
+		} else
+			ipoib_warn(priv, "cm recv completion event with wrid %d (> %d)\n",
+				   wr_id, ipoib_recvq_size);
 		return;
 	}
 
@@ -374,9 +442,9 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
 		if (p && time_after_eq(jiffies, p->jiffies + IPOIB_CM_RX_UPDATE_TIME)) {
 			spin_lock_irqsave(&priv->lock, flags);
 			p->jiffies = jiffies;
-			/* Move this entry to list head, but do
-			 * not re-add it if it has been removed. */
-			if (!list_empty(&p->list))
+			/* Move this entry to list head, but do not re-add it
+			 * if it has been moved out of list. */
+			if (p->state == IPOIB_CM_RX_LIVE)
 				list_move(&p->list, &priv->cm.passive_ids);
 			spin_unlock_irqrestore(&priv->lock, flags);
 		}
@@ -592,8 +660,7 @@ int ipoib_cm_dev_open(struct net_device *dev)
 	if (IS_ERR(priv->cm.id)) {
 		printk(KERN_WARNING "%s: failed to create CM ID\n", priv->ca->name);
 		ret = PTR_ERR(priv->cm.id);
-		priv->cm.id = NULL;
-		return ret;
+		goto err_cm;
 	}
 
 	ret = ib_cm_listen(priv->cm.id, cpu_to_be64(IPOIB_CM_IETF_ID | priv->qp->qp_num),
@@ -601,34 +668,76 @@ int ipoib_cm_dev_open(struct net_device *dev)
 	if (ret) {
 		printk(KERN_WARNING "%s: failed to listen on ID 0x%llx\n", priv->ca->name,
 		       IPOIB_CM_IETF_ID | priv->qp->qp_num);
-		ib_destroy_cm_id(priv->cm.id);
-		priv->cm.id = NULL;
-		return ret;
+		goto err_listen;
 	}
+
 	return 0;
+
+err_listen:
+	ib_destroy_cm_id(priv->cm.id);
+err_cm:
+	priv->cm.id = NULL;
+	return ret;
 }
 
 void ipoib_cm_dev_stop(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	struct ipoib_cm_rx *p;
+	struct ipoib_cm_rx *p, *n;
+	unsigned long begin;
+	LIST_HEAD(list);
+	int ret;
 
 	if (!IPOIB_CM_SUPPORTED(dev->dev_addr) || !priv->cm.id)
 		return;
 
 	ib_destroy_cm_id(priv->cm.id);
 	priv->cm.id = NULL;
+
 	spin_lock_irq(&priv->lock);
 	while (!list_empty(&priv->cm.passive_ids)) {
 		p = list_entry(priv->cm.passive_ids.next, typeof(*p), list);
-		list_del_init(&p->list);
+		list_move(&p->list, &priv->cm.rx_error_list);
+		p->state = IPOIB_CM_RX_ERROR;
+		spin_unlock_irq(&priv->lock);
+		ret = ib_modify_qp(p->qp, &ipoib_cm_err_attr, IB_QP_STATE);
+		if (ret)
+			ipoib_warn(priv, "unable to move qp to error state: %d\n", ret);
+		spin_lock_irq(&priv->lock);
+	}
+
+	/* Wait for all RX to be drained */
+	begin = jiffies;
+
+	while (!list_empty(&priv->cm.rx_error_list) ||
+	       !list_empty(&priv->cm.rx_flush_list) ||
+	       !list_empty(&priv->cm.rx_drain_list)) {
+		if (time_after(jiffies, begin + 5 * HZ)) {
+			ipoib_warn(priv, "RX drain timing out\n");
+
+			/*
+			 * assume the HW is wedged and just free up everything.
+			 */
+			list_splice_init(&priv->cm.rx_flush_list, &list);
+			list_splice_init(&priv->cm.rx_error_list, &list);
+			list_splice_init(&priv->cm.rx_drain_list, &list);
+			break;
+		}
 		spin_unlock_irq(&priv->lock);
+		msleep(1);
+		ipoib_drain_cq(dev);
+		spin_lock_irq(&priv->lock);
+	}
+
+	list_splice_init(&priv->cm.rx_reap_list, &list);
+
+	spin_unlock_irq(&priv->lock);
+
+	list_for_each_entry_safe(p, n, &list, list) {
 		ib_destroy_cm_id(p->id);
 		ib_destroy_qp(p->qp);
 		kfree(p);
-		spin_lock_irq(&priv->lock);
 	}
-	spin_unlock_irq(&priv->lock);
 
 	cancel_delayed_work(&priv->cm.stale_task);
 }
@@ -1079,24 +1188,44 @@ void ipoib_cm_skb_too_long(struct net_device* dev, struct sk_buff *skb,
 		queue_work(ipoib_workqueue, &priv->cm.skb_task);
 }
 
+static void ipoib_cm_rx_reap(struct work_struct *work)
+{
+	struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
+						   cm.rx_reap_task);
+	struct ipoib_cm_rx *p, *n;
+	LIST_HEAD(list);
+
+	spin_lock_irq(&priv->lock);
+	list_splice_init(&priv->cm.rx_reap_list, &list);
+	spin_unlock_irq(&priv->lock);
+
+	list_for_each_entry_safe(p, n, &list, list) {
+		ib_destroy_cm_id(p->id);
+		ib_destroy_qp(p->qp);
+		kfree(p);
+	}
+}
+
 static void ipoib_cm_stale_task(struct work_struct *work)
 {
 	struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
 						   cm.stale_task.work);
 	struct ipoib_cm_rx *p;
+	int ret;
 
 	spin_lock_irq(&priv->lock);
 	while (!list_empty(&priv->cm.passive_ids)) {
-		/* List if sorted by LRU, start from tail,
+		/* List is sorted by LRU, start from tail,
 		 * stop when we see a recently used entry */
 		p = list_entry(priv->cm.passive_ids.prev, typeof(*p), list);
 		if (time_before_eq(jiffies, p->jiffies + IPOIB_CM_RX_TIMEOUT))
 			break;
-		list_del_init(&p->list);
+		list_move(&p->list, &priv->cm.rx_error_list);
+		p->state = IPOIB_CM_RX_ERROR;
 		spin_unlock_irq(&priv->lock);
-		ib_destroy_cm_id(p->id);
-		ib_destroy_qp(p->qp);
-		kfree(p);
+		ret = ib_modify_qp(p->qp, &ipoib_cm_err_attr, IB_QP_STATE);
+		if (ret)
+			ipoib_warn(priv, "unable to move qp to error state: %d\n", ret);
 		spin_lock_irq(&priv->lock);
 	}
 
@@ -1164,9 +1293,14 @@ int ipoib_cm_dev_init(struct net_device *dev)
 	INIT_LIST_HEAD(&priv->cm.passive_ids);
 	INIT_LIST_HEAD(&priv->cm.reap_list);
 	INIT_LIST_HEAD(&priv->cm.start_list);
+	INIT_LIST_HEAD(&priv->cm.rx_error_list);
+	INIT_LIST_HEAD(&priv->cm.rx_flush_list);
+	INIT_LIST_HEAD(&priv->cm.rx_drain_list);
+	INIT_LIST_HEAD(&priv->cm.rx_reap_list);
 	INIT_WORK(&priv->cm.start_task, ipoib_cm_tx_start);
 	INIT_WORK(&priv->cm.reap_task, ipoib_cm_tx_reap);
 	INIT_WORK(&priv->cm.skb_task, ipoib_cm_skb_reap);
+	INIT_WORK(&priv->cm.rx_reap_task, ipoib_cm_rx_reap);
 	INIT_DELAYED_WORK(&priv->cm.stale_task, ipoib_cm_stale_task);
 
 	skb_queue_head_init(&priv->cm.skb_queue);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index 68d72c6..8404f05b 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -448,6 +448,13 @@ int ipoib_ib_dev_open(struct net_device *dev)
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	int ret;
 
+	if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &priv->pkey_index)) {
+		ipoib_warn(priv, "P_Key 0x%04x not found\n", priv->pkey);
+		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+		return -1;
+	}
+	set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+
 	ret = ipoib_init_qp(dev);
 	if (ret) {
 		ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret);
@@ -457,14 +464,14 @@ int ipoib_ib_dev_open(struct net_device *dev)
 	ret = ipoib_ib_post_receives(dev);
 	if (ret) {
 		ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret);
-		ipoib_ib_dev_stop(dev);
+		ipoib_ib_dev_stop(dev, 1);
 		return -1;
 	}
 
 	ret = ipoib_cm_dev_open(dev);
 	if (ret) {
-		ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret);
-		ipoib_ib_dev_stop(dev);
+		ipoib_warn(priv, "ipoib_cm_dev_open returned %d\n", ret);
+		ipoib_ib_dev_stop(dev, 1);
 		return -1;
 	}
 
@@ -516,7 +523,7 @@ int ipoib_ib_dev_down(struct net_device *dev, int flush)
 	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
 		mutex_lock(&pkey_mutex);
 		set_bit(IPOIB_PKEY_STOP, &priv->flags);
-		cancel_delayed_work(&priv->pkey_task);
+		cancel_delayed_work(&priv->pkey_poll_task);
 		mutex_unlock(&pkey_mutex);
 		if (flush)
 			flush_workqueue(ipoib_workqueue);
@@ -543,13 +550,30 @@ static int recvs_pending(struct net_device *dev)
 	return pending;
 }
 
-int ipoib_ib_dev_stop(struct net_device *dev)
+void ipoib_drain_cq(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	int i, n;
+	do {
+		n = ib_poll_cq(priv->cq, IPOIB_NUM_WC, priv->ibwc);
+		for (i = 0; i < n; ++i) {
+			if (priv->ibwc[i].wr_id & IPOIB_CM_OP_SRQ)
+				ipoib_cm_handle_rx_wc(dev, priv->ibwc + i);
+			else if (priv->ibwc[i].wr_id & IPOIB_OP_RECV)
+				ipoib_ib_handle_rx_wc(dev, priv->ibwc + i);
+			else
+				ipoib_ib_handle_tx_wc(dev, priv->ibwc + i);
+		}
+	} while (n == IPOIB_NUM_WC);
+}
+
+int ipoib_ib_dev_stop(struct net_device *dev, int flush)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ib_qp_attr qp_attr;
 	unsigned long begin;
 	struct ipoib_tx_buf *tx_req;
-	int i, n;
+	int i;
 
 	clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags);
 	netif_poll_disable(dev);
@@ -604,17 +628,7 @@ int ipoib_ib_dev_stop(struct net_device *dev)
 			goto timeout;
 		}
 
-		do {
-			n = ib_poll_cq(priv->cq, IPOIB_NUM_WC, priv->ibwc);
-			for (i = 0; i < n; ++i) {
-				if (priv->ibwc[i].wr_id & IPOIB_CM_OP_SRQ)
-					ipoib_cm_handle_rx_wc(dev, priv->ibwc + i);
-				else if (priv->ibwc[i].wr_id & IPOIB_OP_RECV)
-					ipoib_ib_handle_rx_wc(dev, priv->ibwc + i);
-				else
-					ipoib_ib_handle_tx_wc(dev, priv->ibwc + i);
-			}
-		} while (n == IPOIB_NUM_WC);
+		ipoib_drain_cq(dev);
 
 		msleep(1);
 	}
@@ -629,7 +643,8 @@ timeout:
 	/* Wait for all AHs to be reaped */
 	set_bit(IPOIB_STOP_REAPER, &priv->flags);
 	cancel_delayed_work(&priv->ah_reap_task);
-	flush_workqueue(ipoib_workqueue);
+	if (flush)
+		flush_workqueue(ipoib_workqueue);
 
 	begin = jiffies;
 
@@ -673,13 +688,24 @@ int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
 	return 0;
 }
 
-void ipoib_ib_dev_flush(struct work_struct *work)
+static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, int pkey_event)
 {
-	struct ipoib_dev_priv *cpriv, *priv =
-		container_of(work, struct ipoib_dev_priv, flush_task);
+	struct ipoib_dev_priv *cpriv;
 	struct net_device *dev = priv->dev;
+	u16 new_index;
+
+	mutex_lock(&priv->vlan_mutex);
+
+	/*
+	 * Flush any child interfaces too -- they might be up even if
+	 * the parent is down.
+	 */
+	list_for_each_entry(cpriv, &priv->child_intfs, list)
+		__ipoib_ib_dev_flush(cpriv, pkey_event);
 
-	if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags) ) {
+	mutex_unlock(&priv->vlan_mutex);
+
+	if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) {
 		ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n");
 		return;
 	}
@@ -689,10 +715,32 @@ void ipoib_ib_dev_flush(struct work_struct *work)
 		return;
 	}
 
+	if (pkey_event) {
+		if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &new_index)) {
+			clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+			ipoib_ib_dev_down(dev, 0);
+			ipoib_pkey_dev_delay_open(dev);
+			return;
+		}
+		set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+
+		/* restart QP only if P_Key index is changed */
+		if (new_index == priv->pkey_index) {
+			ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n");
+			return;
+		}
+		priv->pkey_index = new_index;
+	}
+
 	ipoib_dbg(priv, "flushing\n");
 
 	ipoib_ib_dev_down(dev, 0);
 
+	if (pkey_event) {
+		ipoib_ib_dev_stop(dev, 0);
+		ipoib_ib_dev_open(dev);
+	}
+
 	/*
 	 * The device could have been brought down between the start and when
 	 * we get here, don't bring it back up if it's not configured up
@@ -701,14 +749,24 @@ void ipoib_ib_dev_flush(struct work_struct *work)
 		ipoib_ib_dev_up(dev);
 		ipoib_mcast_restart_task(&priv->restart_task);
 	}
+}
 
-	mutex_lock(&priv->vlan_mutex);
+void ipoib_ib_dev_flush(struct work_struct *work)
+{
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, flush_task);
 
-	/* Flush any child interfaces too */
-	list_for_each_entry(cpriv, &priv->child_intfs, list)
-		ipoib_ib_dev_flush(&cpriv->flush_task);
+	ipoib_dbg(priv, "Flushing %s\n", priv->dev->name);
+	__ipoib_ib_dev_flush(priv, 0);
+}
 
-	mutex_unlock(&priv->vlan_mutex);
+void ipoib_pkey_event(struct work_struct *work)
+{
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, pkey_event_task);
+
+	ipoib_dbg(priv, "Flushing %s and restarting its QP\n", priv->dev->name);
+	__ipoib_ib_dev_flush(priv, 1);
 }
 
 void ipoib_ib_dev_cleanup(struct net_device *dev)
@@ -736,7 +794,7 @@ void ipoib_ib_dev_cleanup(struct net_device *dev)
 void ipoib_pkey_poll(struct work_struct *work)
 {
 	struct ipoib_dev_priv *priv =
-		container_of(work, struct ipoib_dev_priv, pkey_task.work);
+		container_of(work, struct ipoib_dev_priv, pkey_poll_task.work);
 	struct net_device *dev = priv->dev;
 
 	ipoib_pkey_dev_check_presence(dev);
@@ -747,7 +805,7 @@ void ipoib_pkey_poll(struct work_struct *work)
 		mutex_lock(&pkey_mutex);
 		if (!test_bit(IPOIB_PKEY_STOP, &priv->flags))
 			queue_delayed_work(ipoib_workqueue,
-					   &priv->pkey_task,
+					   &priv->pkey_poll_task,
 					   HZ);
 		mutex_unlock(&pkey_mutex);
 	}
@@ -766,7 +824,7 @@ int ipoib_pkey_dev_delay_open(struct net_device *dev)
 		mutex_lock(&pkey_mutex);
 		clear_bit(IPOIB_PKEY_STOP, &priv->flags);
 		queue_delayed_work(ipoib_workqueue,
-				   &priv->pkey_task,
+				   &priv->pkey_poll_task,
 				   HZ);
 		mutex_unlock(&pkey_mutex);
 		return 1;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index 0a428f2..894b1dcd 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -107,7 +107,7 @@ int ipoib_open(struct net_device *dev)
 		return -EINVAL;
 
 	if (ipoib_ib_dev_up(dev)) {
-		ipoib_ib_dev_stop(dev);
+		ipoib_ib_dev_stop(dev, 1);
 		return -EINVAL;
 	}
 
@@ -152,7 +152,7 @@ static int ipoib_stop(struct net_device *dev)
 	flush_workqueue(ipoib_workqueue);
 
 	ipoib_ib_dev_down(dev, 1);
-	ipoib_ib_dev_stop(dev);
+	ipoib_ib_dev_stop(dev, 1);
 
 	if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
 		struct ipoib_dev_priv *cpriv;
@@ -988,7 +988,8 @@ static void ipoib_setup(struct net_device *dev)
 	INIT_LIST_HEAD(&priv->dead_ahs);
 	INIT_LIST_HEAD(&priv->multicast_list);
 
-	INIT_DELAYED_WORK(&priv->pkey_task,    ipoib_pkey_poll);
+	INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll);
+	INIT_WORK(&priv->pkey_event_task, ipoib_pkey_event);
 	INIT_DELAYED_WORK(&priv->mcast_task,   ipoib_mcast_join_task);
 	INIT_WORK(&priv->flush_task,   ipoib_ib_dev_flush);
 	INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
index 54fbead..aae3670 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
@@ -524,7 +524,7 @@ void ipoib_mcast_join_task(struct work_struct *work)
 		return;
 
 	if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
-		ipoib_warn(priv, "ib_gid_entry_get() failed\n");
+		ipoib_warn(priv, "ib_query_gid() failed\n");
 	else
 		memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
 
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
index 5c3c6a4..982eb88 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
@@ -33,8 +33,6 @@
  * $Id: ipoib_verbs.c 1349 2004-12-16 21:09:43Z roland $
  */
 
-#include <rdma/ib_cache.h>
-
 #include "ipoib.h"
 
 int ipoib_mcast_attach(struct net_device *dev, u16 mlid, union ib_gid *mgid)
@@ -49,7 +47,7 @@ int ipoib_mcast_attach(struct net_device *dev, u16 mlid, union ib_gid *mgid)
 	if (!qp_attr)
 		goto out;
 
-	if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) {
+	if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) {
 		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
 		ret = -ENXIO;
 		goto out;
@@ -94,26 +92,16 @@ int ipoib_init_qp(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	int ret;
-	u16 pkey_index;
 	struct ib_qp_attr qp_attr;
 	int attr_mask;
 
-	/*
-	 * Search through the port P_Key table for the requested pkey value.
-	 * The port has to be assigned to the respective IB partition in
-	 * advance.
-	 */
-	ret = ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index);
-	if (ret) {
-		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
-		return ret;
-	}
-	set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
+		return -1;
 
 	qp_attr.qp_state = IB_QPS_INIT;
 	qp_attr.qkey = 0;
 	qp_attr.port_num = priv->port;
-	qp_attr.pkey_index = pkey_index;
+	qp_attr.pkey_index = priv->pkey_index;
 	attr_mask =
 	    IB_QP_QKEY |
 	    IB_QP_PORT |
@@ -185,7 +173,7 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 	size = ipoib_sendq_size + ipoib_recvq_size + 1;
 	ret = ipoib_cm_dev_init(dev);
 	if (!ret)
-		size += ipoib_recvq_size;
+		size += ipoib_recvq_size + 1 /* 1 extra for rx_drain_qp */;
 
 	priv->cq = ib_create_cq(priv->ca, ipoib_ib_completion, NULL, dev, size, 0);
 	if (IS_ERR(priv->cq)) {
@@ -259,14 +247,18 @@ void ipoib_event(struct ib_event_handler *handler,
 	struct ipoib_dev_priv *priv =
 		container_of(handler, struct ipoib_dev_priv, event_handler);
 
-	if ((record->event == IB_EVENT_PORT_ERR    ||
-	     record->event == IB_EVENT_PKEY_CHANGE ||
-	     record->event == IB_EVENT_PORT_ACTIVE ||
-	     record->event == IB_EVENT_LID_CHANGE  ||
-	     record->event == IB_EVENT_SM_CHANGE   ||
-	     record->event == IB_EVENT_CLIENT_REREGISTER) &&
-	    record->element.port_num == priv->port) {
+	if (record->element.port_num != priv->port)
+		return;
+
+	if (record->event == IB_EVENT_PORT_ERR    ||
+	    record->event == IB_EVENT_PORT_ACTIVE ||
+	    record->event == IB_EVENT_LID_CHANGE  ||
+	    record->event == IB_EVENT_SM_CHANGE   ||
+	    record->event == IB_EVENT_CLIENT_REREGISTER) {
 		ipoib_dbg(priv, "Port state change event\n");
 		queue_work(ipoib_workqueue, &priv->flush_task);
+	} else if (record->event == IB_EVENT_PKEY_CHANGE) {
+		ipoib_dbg(priv, "P_Key change event on port:%d\n", priv->port);
+		queue_work(ipoib_workqueue, &priv->pkey_event_task);
 	}
 }