X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=drivers%2Finfiniband%2Fulp%2Fipoib%2Fipoib_ib.c;h=59d9594ed6d97e0fc243066f42f94c8f8f3c0ca6;hb=refs%2Fheads%2Fvserver;hp=8238766746b2c1c34d70768167541d7dd113460a;hpb=f7f1b0f1e2fbadeab12d24236000e778aa9b1ead;p=linux-2.6.git

diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index 823876674..59d9594ed 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -1,5 +1,8 @@
 /*
  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
+ * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
+ * Copyright (c) 2004, 2005 Voltaire, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -35,7 +38,7 @@
 #include <linux/delay.h>
 #include <linux/dma-mapping.h>
 
-#include <ib_cache.h>
+#include <rdma/ib_cache.h>
 
 #include "ipoib.h"
 
@@ -49,7 +52,7 @@ MODULE_PARM_DESC(data_debug_level,
 
 #define	IPOIB_OP_RECV	(1ul << 31)
 
-static DECLARE_MUTEX(pkey_sem);
+static DEFINE_MUTEX(pkey_mutex);
 
 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
 				 struct ib_pd *pd, struct ib_ah_attr *attr)
@@ -81,68 +84,68 @@ void ipoib_free_ah(struct kref *kref)
 
 	unsigned long flags;
 
-	if (ah->last_send <= priv->tx_tail) {
-		ipoib_dbg(priv, "Freeing ah %p\n", ah->ah);
-		ib_destroy_ah(ah->ah);
-		kfree(ah);
-	} else {
-		spin_lock_irqsave(&priv->lock, flags);
-		list_add_tail(&ah->list, &priv->dead_ahs);
-		spin_unlock_irqrestore(&priv->lock, flags);
-	}
+	spin_lock_irqsave(&priv->lock, flags);
+	list_add_tail(&ah->list, &priv->dead_ahs);
+	spin_unlock_irqrestore(&priv->lock, flags);
 }
 
-static inline int ipoib_ib_receive(struct ipoib_dev_priv *priv,
-				   unsigned int wr_id,
-				   dma_addr_t addr)
+static int ipoib_ib_post_receive(struct net_device *dev, int id)
 {
-	struct ib_sge list = {
-		.addr    = addr,
-		.length  = IPOIB_BUF_SIZE,
-		.lkey    = priv->mr->lkey,
-	};
-	struct ib_recv_wr param = {
-		.wr_id 	    = wr_id | IPOIB_OP_RECV,
-		.sg_list    = &list,
-		.num_sge    = 1,
-	};
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ib_sge list;
+	struct ib_recv_wr param;
 	struct ib_recv_wr *bad_wr;
+	int ret;
 
-	return ib_post_recv(priv->qp, &param, &bad_wr);
+	list.addr     = priv->rx_ring[id].mapping;
+	list.length   = IPOIB_BUF_SIZE;
+	list.lkey     = priv->mr->lkey;
+
+	param.next    = NULL;
+	param.wr_id   = id | IPOIB_OP_RECV;
+	param.sg_list = &list;
+	param.num_sge = 1;
+
+	ret = ib_post_recv(priv->qp, &param, &bad_wr);
+	if (unlikely(ret)) {
+		ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret);
+		ib_dma_unmap_single(priv->ca, priv->rx_ring[id].mapping,
+				    IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
+		dev_kfree_skb_any(priv->rx_ring[id].skb);
+		priv->rx_ring[id].skb = NULL;
+	}
+
+	return ret;
 }
 
-static int ipoib_ib_post_receive(struct net_device *dev, int id)
+static int ipoib_alloc_rx_skb(struct net_device *dev, int id)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct sk_buff *skb;
-	dma_addr_t addr;
-	int ret;
+	u64 addr;
 
 	skb = dev_alloc_skb(IPOIB_BUF_SIZE + 4);
-	if (!skb) {
-		ipoib_warn(priv, "failed to allocate receive buffer\n");
-
-		priv->rx_ring[id].skb = NULL;
+	if (!skb)
 		return -ENOMEM;
-	}
-	skb_reserve(skb, 4);	/* 16 byte align IP header */
-	priv->rx_ring[id].skb = skb;
-	addr = dma_map_single(priv->ca->dma_device,
-			      skb->data, IPOIB_BUF_SIZE,
-			      DMA_FROM_DEVICE);
-	pci_unmap_addr_set(&priv->rx_ring[id], mapping, addr);
-
-	ret = ipoib_ib_receive(priv, id, addr);
-	if (ret) {
-		ipoib_warn(priv, "ipoib_ib_receive failed for buf %d (%d)\n",
-			   id, ret);
-		dma_unmap_single(priv->ca->dma_device, addr,
-				 IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
+
+	/*
+	 * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte
+	 * header.  So we need 4 more bytes to get to 48 and align the
+	 * IP header to a multiple of 16.
+	 */
+	skb_reserve(skb, 4);
+
+	addr = ib_dma_map_single(priv->ca, skb->data, IPOIB_BUF_SIZE,
+				 DMA_FROM_DEVICE);
+	if (unlikely(ib_dma_mapping_error(priv->ca, addr))) {
 		dev_kfree_skb_any(skb);
-		priv->rx_ring[id].skb = NULL;
+		return -EIO;
 	}
 
-	return ret;
+	priv->rx_ring[id].skb     = skb;
+	priv->rx_ring[id].mapping = addr;
+
+	return 0;
 }
 
 static int ipoib_ib_post_receives(struct net_device *dev)
@@ -150,7 +153,11 @@ static int ipoib_ib_post_receives(struct net_device *dev)
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	int i;
 
-	for (i = 0; i < IPOIB_RX_RING_SIZE; ++i) {
+	for (i = 0; i < ipoib_recvq_size; ++i) {
+		if (ipoib_alloc_rx_skb(dev, i)) {
+			ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
+			return -ENOMEM;
+		}
 		if (ipoib_ib_post_receive(dev, i)) {
 			ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i);
 			return -EIO;
@@ -160,108 +167,126 @@ static int ipoib_ib_post_receives(struct net_device *dev)
 	return 0;
 }
 
-static void ipoib_ib_handle_wc(struct net_device *dev,
-			       struct ib_wc *wc)
+static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	unsigned int wr_id = wc->wr_id;
+	unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV;
+	struct sk_buff *skb;
+	u64 addr;
 
-	ipoib_dbg_data(priv, "called: id %d, op %d, status: %d\n",
+	ipoib_dbg_data(priv, "recv completion: id %d, op %d, status: %d\n",
 		       wr_id, wc->opcode, wc->status);
 
-	if (wr_id & IPOIB_OP_RECV) {
-		wr_id &= ~IPOIB_OP_RECV;
+	if (unlikely(wr_id >= ipoib_recvq_size)) {
+		ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n",
+			   wr_id, ipoib_recvq_size);
+		return;
+	}
 
-		if (wr_id < IPOIB_RX_RING_SIZE) {
-			struct sk_buff *skb = priv->rx_ring[wr_id].skb;
+	skb  = priv->rx_ring[wr_id].skb;
+	addr = priv->rx_ring[wr_id].mapping;
 
-			priv->rx_ring[wr_id].skb = NULL;
+	if (unlikely(wc->status != IB_WC_SUCCESS)) {
+		if (wc->status != IB_WC_WR_FLUSH_ERR)
+			ipoib_warn(priv, "failed recv event "
+				   "(status=%d, wrid=%d vend_err %x)\n",
+				   wc->status, wr_id, wc->vendor_err);
+		ib_dma_unmap_single(priv->ca, addr,
+				    IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
+		dev_kfree_skb_any(skb);
+		priv->rx_ring[wr_id].skb = NULL;
+		return;
+	}
 
-			dma_unmap_single(priv->ca->dma_device,
-					 pci_unmap_addr(&priv->rx_ring[wr_id],
-							mapping),
-					 IPOIB_BUF_SIZE,
-					 DMA_FROM_DEVICE);
+	/*
+	 * If we can't allocate a new RX buffer, dump
+	 * this packet and reuse the old buffer.
+	 */
+	if (unlikely(ipoib_alloc_rx_skb(dev, wr_id))) {
+		++priv->stats.rx_dropped;
+		goto repost;
+	}
 
-			if (wc->status != IB_WC_SUCCESS) {
-				if (wc->status != IB_WC_WR_FLUSH_ERR)
-					ipoib_warn(priv, "failed recv event "
-						   "(status=%d, wrid=%d vend_err %x)\n",
-						   wc->status, wr_id, wc->vendor_err);
-				dev_kfree_skb_any(skb);
-				return;
-			}
+	ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
+		       wc->byte_len, wc->slid);
 
-			ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
-				       wc->byte_len, wc->slid);
-
-			skb_put(skb, wc->byte_len);
-			skb_pull(skb, IB_GRH_BYTES);
-
-			if (wc->slid != priv->local_lid ||
-			    wc->src_qp != priv->qp->qp_num) {
-				skb->protocol = ((struct ipoib_header *) skb->data)->proto;
-				skb->mac.raw = skb->data;
-				skb_pull(skb, IPOIB_ENCAP_LEN);
-
-				dev->last_rx = jiffies;
-				++priv->stats.rx_packets;
-				priv->stats.rx_bytes += skb->len;
-
-				skb->dev = dev;
-				/* XXX get correct PACKET_ type here */
-				skb->pkt_type = PACKET_HOST;
-				netif_rx_ni(skb);
-			} else {
-				ipoib_dbg_data(priv, "dropping loopback packet\n");
-				dev_kfree_skb_any(skb);
-			}
+	ib_dma_unmap_single(priv->ca, addr, IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
+
+	skb_put(skb, wc->byte_len);
+	skb_pull(skb, IB_GRH_BYTES);
 
-			/* repost receive */
-			if (ipoib_ib_post_receive(dev, wr_id))
-				ipoib_warn(priv, "ipoib_ib_post_receive failed "
-					   "for buf %d\n", wr_id);
-		} else
-			ipoib_warn(priv, "completion event with wrid %d\n",
-				   wr_id);
+	if (wc->slid != priv->local_lid ||
+	    wc->src_qp != priv->qp->qp_num) {
+		skb->protocol = ((struct ipoib_header *) skb->data)->proto;
+		skb->mac.raw = skb->data;
+		skb_pull(skb, IPOIB_ENCAP_LEN);
 
+		dev->last_rx = jiffies;
+		++priv->stats.rx_packets;
+		priv->stats.rx_bytes += skb->len;
+
+		skb->dev = dev;
+		/* XXX get correct PACKET_ type here */
+		skb->pkt_type = PACKET_HOST;
+		netif_rx_ni(skb);
 	} else {
-		struct ipoib_buf *tx_req;
-		unsigned long flags;
+		ipoib_dbg_data(priv, "dropping loopback packet\n");
+		dev_kfree_skb_any(skb);
+	}
 
-		if (wr_id >= IPOIB_TX_RING_SIZE) {
-			ipoib_warn(priv, "completion event with wrid %d (> %d)\n",
-				   wr_id, IPOIB_TX_RING_SIZE);
-			return;
-		}
+repost:
+	if (unlikely(ipoib_ib_post_receive(dev, wr_id)))
+		ipoib_warn(priv, "ipoib_ib_post_receive failed "
+			   "for buf %d\n", wr_id);
+}
 
-		ipoib_dbg_data(priv, "send complete, wrid %d\n", wr_id);
+static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	unsigned int wr_id = wc->wr_id;
+	struct ipoib_tx_buf *tx_req;
+	unsigned long flags;
 
-		tx_req = &priv->tx_ring[wr_id];
+	ipoib_dbg_data(priv, "send completion: id %d, op %d, status: %d\n",
+		       wr_id, wc->opcode, wc->status);
 
-		dma_unmap_single(priv->ca->dma_device,
-				 pci_unmap_addr(tx_req, mapping),
-				 tx_req->skb->len,
-				 DMA_TO_DEVICE);
+	if (unlikely(wr_id >= ipoib_sendq_size)) {
+		ipoib_warn(priv, "send completion event with wrid %d (> %d)\n",
+			   wr_id, ipoib_sendq_size);
+		return;
+	}
 
-		++priv->stats.tx_packets;
-		priv->stats.tx_bytes += tx_req->skb->len;
+	tx_req = &priv->tx_ring[wr_id];
 
-		dev_kfree_skb_any(tx_req->skb);
+	ib_dma_unmap_single(priv->ca, tx_req->mapping,
+			    tx_req->skb->len, DMA_TO_DEVICE);
 
-		spin_lock_irqsave(&priv->tx_lock, flags);
-		++priv->tx_tail;
-		if (netif_queue_stopped(dev) &&
-		    priv->tx_head - priv->tx_tail <= IPOIB_TX_RING_SIZE / 2)
-			netif_wake_queue(dev);
-		spin_unlock_irqrestore(&priv->tx_lock, flags);
+	++priv->stats.tx_packets;
+	priv->stats.tx_bytes += tx_req->skb->len;
 
-		if (wc->status != IB_WC_SUCCESS &&
-		    wc->status != IB_WC_WR_FLUSH_ERR)
-			ipoib_warn(priv, "failed send event "
-				   "(status=%d, wrid=%d vend_err %x)\n",
-				   wc->status, wr_id, wc->vendor_err);
-	}
+	dev_kfree_skb_any(tx_req->skb);
+
+	spin_lock_irqsave(&priv->tx_lock, flags);
+	++priv->tx_tail;
+	if (netif_queue_stopped(dev) &&
+	    test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags) &&
+	    priv->tx_head - priv->tx_tail <= ipoib_sendq_size >> 1)
+		netif_wake_queue(dev);
+	spin_unlock_irqrestore(&priv->tx_lock, flags);
+
+	if (wc->status != IB_WC_SUCCESS &&
+	    wc->status != IB_WC_WR_FLUSH_ERR)
+		ipoib_warn(priv, "failed send event "
+			   "(status=%d, wrid=%d vend_err %x)\n",
+			   wc->status, wr_id, wc->vendor_err);
+}
+
+static void ipoib_ib_handle_wc(struct net_device *dev, struct ib_wc *wc)
+{
+	if (wc->wr_id & IPOIB_OP_RECV)
+		ipoib_ib_handle_rx_wc(dev, wc);
+	else
+		ipoib_ib_handle_tx_wc(dev, wc);
 }
 
 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr)
@@ -281,7 +306,7 @@ void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr)
 static inline int post_send(struct ipoib_dev_priv *priv,
 			    unsigned int wr_id,
 			    struct ib_ah *address, u32 qpn,
-			    dma_addr_t addr, int len)
+			    u64 addr, int len)
 {
 	struct ib_send_wr *bad_wr;
 
@@ -299,10 +324,10 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 		struct ipoib_ah *address, u32 qpn)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	struct ipoib_buf *tx_req;
-	dma_addr_t addr;
+	struct ipoib_tx_buf *tx_req;
+	u64 addr;
 
-	if (skb->len > dev->mtu + INFINIBAND_ALEN) {
+	if (unlikely(skb->len > dev->mtu + INFINIBAND_ALEN)) {
 		ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n",
 			   skb->len, dev->mtu + INFINIBAND_ALEN);
 		++priv->stats.tx_dropped;
@@ -321,18 +346,22 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 	 * means we have to make sure everything is properly recorded and
 	 * our state is consistent before we call post_send().
 	 */
-	tx_req = &priv->tx_ring[priv->tx_head & (IPOIB_TX_RING_SIZE - 1)];
+	tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)];
 	tx_req->skb = skb;
-	addr = dma_map_single(priv->ca->dma_device, skb->data, skb->len,
-			      DMA_TO_DEVICE);
-	pci_unmap_addr_set(tx_req, mapping, addr);
+	addr = ib_dma_map_single(priv->ca, skb->data, skb->len,
+				 DMA_TO_DEVICE);
+	if (unlikely(ib_dma_mapping_error(priv->ca, addr))) {
+		++priv->stats.tx_errors;
+		dev_kfree_skb_any(skb);
+		return;
+	}
+	tx_req->mapping = addr;
 
-	if (unlikely(post_send(priv, priv->tx_head & (IPOIB_TX_RING_SIZE - 1),
+	if (unlikely(post_send(priv, priv->tx_head & (ipoib_sendq_size - 1),
 			       address->ah, qpn, addr, skb->len))) {
 		ipoib_warn(priv, "post_send failed\n");
 		++priv->stats.tx_errors;
-		dma_unmap_single(priv->ca->dma_device, addr, skb->len,
-				 DMA_TO_DEVICE);
+		ib_dma_unmap_single(priv->ca, addr, skb->len, DMA_TO_DEVICE);
 		dev_kfree_skb_any(skb);
 	} else {
 		dev->trans_start = jiffies;
@@ -340,7 +369,7 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 		address->last_send = priv->tx_head;
 		++priv->tx_head;
 
-		if (priv->tx_head - priv->tx_tail == IPOIB_TX_RING_SIZE) {
+		if (priv->tx_head - priv->tx_tail == ipoib_sendq_size) {
 			ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n");
 			netif_stop_queue(dev);
 		}
@@ -353,25 +382,23 @@ static void __ipoib_reap_ah(struct net_device *dev)
 	struct ipoib_ah *ah, *tah;
 	LIST_HEAD(remove_list);
 
-	spin_lock_irq(&priv->lock);
+	spin_lock_irq(&priv->tx_lock);
+	spin_lock(&priv->lock);
 	list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list)
-		if (ah->last_send <= priv->tx_tail) {
+		if ((int) priv->tx_tail - (int) ah->last_send >= 0) {
 			list_del(&ah->list);
-			list_add_tail(&ah->list, &remove_list);
+			ib_destroy_ah(ah->ah);
+			kfree(ah);
 		}
-	spin_unlock_irq(&priv->lock);
-
-	list_for_each_entry_safe(ah, tah, &remove_list, list) {
-		ipoib_dbg(priv, "Reaping ah %p\n", ah->ah);
-		ib_destroy_ah(ah->ah);
-		kfree(ah);
-	}
+	spin_unlock(&priv->lock);
+	spin_unlock_irq(&priv->tx_lock);
 }
 
-void ipoib_reap_ah(void *dev_ptr)
+void ipoib_reap_ah(struct work_struct *work)
 {
-	struct net_device *dev = dev_ptr;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, ah_reap_task.work);
+	struct net_device *dev = priv->dev;
 
 	__ipoib_reap_ah(dev);
 
@@ -384,34 +411,55 @@ int ipoib_ib_dev_open(struct net_device *dev)
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	int ret;
 
-	ret = ipoib_qp_create(dev);
+	ret = ipoib_init_qp(dev);
 	if (ret) {
-		ipoib_warn(priv, "ipoib_qp_create returned %d\n", ret);
+		ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret);
 		return -1;
 	}
 
 	ret = ipoib_ib_post_receives(dev);
 	if (ret) {
 		ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret);
+		ipoib_ib_dev_stop(dev);
 		return -1;
 	}
 
 	clear_bit(IPOIB_STOP_REAPER, &priv->flags);
 	queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ);
 
+	set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags);
+
 	return 0;
 }
 
+static void ipoib_pkey_dev_check_presence(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	u16 pkey_index = 0;
+
+	if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index))
+		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+	else
+		set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+}
+
 int ipoib_ib_dev_up(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
+	ipoib_pkey_dev_check_presence(dev);
+
+	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
+		ipoib_dbg(priv, "PKEY is not assigned.\n");
+		return 0;
+	}
+
 	set_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
 
 	return ipoib_mcast_start_thread(dev);
 }
 
-int ipoib_ib_dev_down(struct net_device *dev)
+int ipoib_ib_dev_down(struct net_device *dev, int flush)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
@@ -422,25 +470,17 @@ int ipoib_ib_dev_down(struct net_device *dev)
 
 	/* Shutdown the P_Key thread if still active */
 	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
-		down(&pkey_sem);
+		mutex_lock(&pkey_mutex);
 		set_bit(IPOIB_PKEY_STOP, &priv->flags);
 		cancel_delayed_work(&priv->pkey_task);
-		up(&pkey_sem);
-		flush_workqueue(ipoib_workqueue);
+		mutex_unlock(&pkey_mutex);
+		if (flush)
+			flush_workqueue(ipoib_workqueue);
 	}
 
-	ipoib_mcast_stop_thread(dev);
-
-	/*
-	 * Flush the multicast groups first so we stop any multicast joins. The
-	 * completion thread may have already died and we may deadlock waiting
-	 * for the completion thread to finish some multicast joins.
-	 */
+	ipoib_mcast_stop_thread(dev, flush);
 	ipoib_mcast_dev_flush(dev);
 
-	/* Delete broadcast and local addresses since they will be recreated */
-	ipoib_mcast_dev_down(dev);
-
 	ipoib_flush_paths(dev);
 
 	return 0;
@@ -452,7 +492,7 @@ static int recvs_pending(struct net_device *dev)
 	int pending = 0;
 	int i;
 
-	for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
+	for (i = 0; i < ipoib_recvq_size; ++i)
 		if (priv->rx_ring[i].skb)
 			++pending;
 
@@ -463,15 +503,18 @@ int ipoib_ib_dev_stop(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ib_qp_attr qp_attr;
-	int attr_mask;
 	unsigned long begin;
-	struct ipoib_buf *tx_req;
+	struct ipoib_tx_buf *tx_req;
 	int i;
 
-	/* Kill the existing QP and allocate a new one */
+	clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags);
+
+	/*
+	 * Move our QP to the error state and then reinitialize in
+	 * when all work requests have completed or have been flushed.
+	 */
 	qp_attr.qp_state = IB_QPS_ERR;
-	attr_mask        = IB_QP_STATE;
-	if (ib_modify_qp(priv->qp, &qp_attr, attr_mask))
+	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
 		ipoib_warn(priv, "Failed to modify QP to ERROR state\n");
 
 	/* Wait for all sends and receives to complete */
@@ -486,27 +529,30 @@ int ipoib_ib_dev_stop(struct net_device *dev)
 			 * assume the HW is wedged and just free up
 			 * all our pending work requests.
 			 */
-			while (priv->tx_tail < priv->tx_head) {
+			while ((int) priv->tx_tail - (int) priv->tx_head < 0) {
 				tx_req = &priv->tx_ring[priv->tx_tail &
-							(IPOIB_TX_RING_SIZE - 1)];
-				dma_unmap_single(priv->ca->dma_device,
-						 pci_unmap_addr(tx_req, mapping),
-						 tx_req->skb->len,
-						 DMA_TO_DEVICE);
+							(ipoib_sendq_size - 1)];
+				ib_dma_unmap_single(priv->ca,
+						    tx_req->mapping,
+						    tx_req->skb->len,
+						    DMA_TO_DEVICE);
 				dev_kfree_skb_any(tx_req->skb);
 				++priv->tx_tail;
 			}
 
-			for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
-				if (priv->rx_ring[i].skb) {
-					dma_unmap_single(priv->ca->dma_device,
-							 pci_unmap_addr(&priv->rx_ring[i],
-									mapping),
-							 IPOIB_BUF_SIZE,
-							 DMA_FROM_DEVICE);
-					dev_kfree_skb_any(priv->rx_ring[i].skb);
-					priv->rx_ring[i].skb = NULL;
-				}
+			for (i = 0; i < ipoib_recvq_size; ++i) {
+				struct ipoib_rx_buf *rx_req;
+
+				rx_req = &priv->rx_ring[i];
+				if (!rx_req->skb)
+					continue;
+				ib_dma_unmap_single(priv->ca,
+						    rx_req->mapping,
+						    IPOIB_BUF_SIZE,
+						    DMA_FROM_DEVICE);
+				dev_kfree_skb_any(rx_req->skb);
+				rx_req->skb = NULL;
+			}
 
 			goto timeout;
 		}
@@ -518,8 +564,7 @@ int ipoib_ib_dev_stop(struct net_device *dev)
 
 timeout:
 	qp_attr.qp_state = IB_QPS_RESET;
-	attr_mask        = IB_QP_STATE;
-	if (ib_modify_qp(priv->qp, &qp_attr, attr_mask))
+	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
 		ipoib_warn(priv, "Failed to modify QP to RESET state\n");
 
 	/* Wait for all AHs to be reaped */
@@ -566,28 +611,42 @@ int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
 	return 0;
 }
 
-void ipoib_ib_dev_flush(void *_dev)
+void ipoib_ib_dev_flush(struct work_struct *work)
 {
-	struct net_device *dev = (struct net_device *)_dev;
-	struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv;
+	struct ipoib_dev_priv *cpriv, *priv =
+		container_of(work, struct ipoib_dev_priv, flush_task);
+	struct net_device *dev = priv->dev;
 
-	if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
+	if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags) ) {
+		ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n");
 		return;
+	}
+
+	if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) {
+		ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n");
+		return;
+	}
 
 	ipoib_dbg(priv, "flushing\n");
 
-	ipoib_ib_dev_down(dev);
+	ipoib_ib_dev_down(dev, 0);
 
 	/*
 	 * 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
 	 */
-	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
+	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) {
 		ipoib_ib_dev_up(dev);
+		ipoib_mcast_restart_task(&priv->restart_task);
+	}
+
+	mutex_lock(&priv->vlan_mutex);
 
 	/* Flush any child interfaces too */
 	list_for_each_entry(cpriv, &priv->child_intfs, list)
-		ipoib_ib_dev_flush(&cpriv->dev);
+		ipoib_ib_dev_flush(&cpriv->flush_task);
+
+	mutex_unlock(&priv->vlan_mutex);
 }
 
 void ipoib_ib_dev_cleanup(struct net_device *dev)
@@ -596,10 +655,8 @@ void ipoib_ib_dev_cleanup(struct net_device *dev)
 
 	ipoib_dbg(priv, "cleaning up ib_dev\n");
 
-	ipoib_mcast_stop_thread(dev);
-
-	/* Delete the broadcast address and the local address */
-	ipoib_mcast_dev_down(dev);
+	ipoib_mcast_stop_thread(dev, 1);
+	ipoib_mcast_dev_flush(dev);
 
 	ipoib_transport_dev_cleanup(dev);
 }
@@ -613,35 +670,24 @@ void ipoib_ib_dev_cleanup(struct net_device *dev)
  * Bug #2507. This implementation will probably be removed when the P_Key
  * change async notification is available.
  */
-int ipoib_open(struct net_device *dev);
-
-static void ipoib_pkey_dev_check_presence(struct net_device *dev)
-{
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	u16 pkey_index = 0;
 
-	if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index))
-		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
-	else
-		set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
-}
-
-void ipoib_pkey_poll(void *dev_ptr)
+void ipoib_pkey_poll(struct work_struct *work)
 {
-	struct net_device *dev = dev_ptr;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, pkey_task.work);
+	struct net_device *dev = priv->dev;
 
 	ipoib_pkey_dev_check_presence(dev);
 
 	if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
 		ipoib_open(dev);
 	else {
-		down(&pkey_sem);
+		mutex_lock(&pkey_mutex);
 		if (!test_bit(IPOIB_PKEY_STOP, &priv->flags))
 			queue_delayed_work(ipoib_workqueue,
 					   &priv->pkey_task,
 					   HZ);
-		up(&pkey_sem);
+		mutex_unlock(&pkey_mutex);
 	}
 }
 
@@ -655,12 +701,12 @@ int ipoib_pkey_dev_delay_open(struct net_device *dev)
 
 	/* P_Key value not assigned yet - start polling */
 	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
-		down(&pkey_sem);
+		mutex_lock(&pkey_mutex);
 		clear_bit(IPOIB_PKEY_STOP, &priv->flags);
 		queue_delayed_work(ipoib_workqueue,
 				   &priv->pkey_task,
 				   HZ);
-		up(&pkey_sem);
+		mutex_unlock(&pkey_mutex);
 		return 1;
 	}