RDMA/amso1100: Start of endianness annotation

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>

8 files changed, 166 insertions, 160 deletions

diff --git a/drivers/infiniband/hw/amso1100/c2.c b/drivers/infiniband/hw/amso1100/c2.c
index c50533b5999..113f3c03c5b 100644
--- a/drivers/infiniband/hw/amso1100/c2.c
+++ b/drivers/infiniband/hw/amso1100/c2.c
@@ -130,10 +130,10 @@ static int c2_tx_ring_alloc(struct c2_ring *tx_ring, void *vaddr,
 		tx_desc->status = 0;
 
 		/* Set TXP_HTXD_UNINIT */
-		__raw_writeq(cpu_to_be64(0x1122334455667788ULL),
+		__raw_writeq((__force u64) cpu_to_be64(0x1122334455667788ULL),
 			     (void __iomem *) txp_desc + C2_TXP_ADDR);
 		__raw_writew(0, (void __iomem *) txp_desc + C2_TXP_LEN);
-		__raw_writew(cpu_to_be16(TXP_HTXD_UNINIT),
+		__raw_writew((__force u16) cpu_to_be16(TXP_HTXD_UNINIT),
 			     (void __iomem *) txp_desc + C2_TXP_FLAGS);
 
 		elem->skb = NULL;
@@ -179,13 +179,13 @@ static int c2_rx_ring_alloc(struct c2_ring *rx_ring, void *vaddr,
 		rx_desc->status = 0;
 
 		/* Set RXP_HRXD_UNINIT */
-		__raw_writew(cpu_to_be16(RXP_HRXD_OK),
+		__raw_writew((__force u16) cpu_to_be16(RXP_HRXD_OK),
 		       (void __iomem *) rxp_desc + C2_RXP_STATUS);
 		__raw_writew(0, (void __iomem *) rxp_desc + C2_RXP_COUNT);
 		__raw_writew(0, (void __iomem *) rxp_desc + C2_RXP_LEN);
-		__raw_writeq(cpu_to_be64(0x99aabbccddeeffULL),
+		__raw_writeq((__force u64) cpu_to_be64(0x99aabbccddeeffULL),
 			     (void __iomem *) rxp_desc + C2_RXP_ADDR);
-		__raw_writew(cpu_to_be16(RXP_HRXD_UNINIT),
+		__raw_writew((__force u16) cpu_to_be16(RXP_HRXD_UNINIT),
 			     (void __iomem *) rxp_desc + C2_RXP_FLAGS);
 
 		elem->skb = NULL;
@@ -239,10 +239,11 @@ static inline int c2_rx_alloc(struct c2_port *c2_port, struct c2_element *elem)
 	rxp_hdr->flags = RXP_HRXD_READY;
 
 	__raw_writew(0, elem->hw_desc + C2_RXP_STATUS);
-	__raw_writew(cpu_to_be16((u16) maplen - sizeof(*rxp_hdr)),
+	__raw_writew((__force u16) cpu_to_be16((u16) maplen - sizeof(*rxp_hdr)),
 		     elem->hw_desc + C2_RXP_LEN);
-	__raw_writeq(cpu_to_be64(mapaddr), elem->hw_desc + C2_RXP_ADDR);
-	__raw_writew(cpu_to_be16(RXP_HRXD_READY), elem->hw_desc + C2_RXP_FLAGS);
+	__raw_writeq((__force u64) cpu_to_be64(mapaddr), elem->hw_desc + C2_RXP_ADDR);
+	__raw_writew((__force u16) cpu_to_be16(RXP_HRXD_READY),
+		     elem->hw_desc + C2_RXP_FLAGS);
 
 	elem->skb = skb;
 	elem->mapaddr = mapaddr;
@@ -290,9 +291,9 @@ static void c2_rx_clean(struct c2_port *c2_port)
 		__raw_writew(0, elem->hw_desc + C2_RXP_STATUS);
 		__raw_writew(0, elem->hw_desc + C2_RXP_COUNT);
 		__raw_writew(0, elem->hw_desc + C2_RXP_LEN);
-		__raw_writeq(cpu_to_be64(0x99aabbccddeeffULL),
+		__raw_writeq((__force u64) cpu_to_be64(0x99aabbccddeeffULL),
 			     elem->hw_desc + C2_RXP_ADDR);
-		__raw_writew(cpu_to_be16(RXP_HRXD_UNINIT),
+		__raw_writew((__force u16) cpu_to_be16(RXP_HRXD_UNINIT),
 			     elem->hw_desc + C2_RXP_FLAGS);
 
 		if (elem->skb) {
@@ -346,16 +347,16 @@ static void c2_tx_clean(struct c2_port *c2_port)
 					     elem->hw_desc + C2_TXP_LEN);
 				__raw_writeq(0,
 					     elem->hw_desc + C2_TXP_ADDR);
-				__raw_writew(cpu_to_be16(TXP_HTXD_DONE),
+				__raw_writew((__force u16) cpu_to_be16(TXP_HTXD_DONE),
 					     elem->hw_desc + C2_TXP_FLAGS);
 				c2_port->netstats.tx_dropped++;
 				break;
 			} else {
 				__raw_writew(0,
 					     elem->hw_desc + C2_TXP_LEN);
-				__raw_writeq(cpu_to_be64(0x1122334455667788ULL),
+				__raw_writeq((__force u64) cpu_to_be64(0x1122334455667788ULL),
 					     elem->hw_desc + C2_TXP_ADDR);
-				__raw_writew(cpu_to_be16(TXP_HTXD_UNINIT),
+				__raw_writew((__force u16) cpu_to_be16(TXP_HTXD_UNINIT),
 					     elem->hw_desc + C2_TXP_FLAGS);
 			}
 
@@ -390,7 +391,7 @@ static void c2_tx_interrupt(struct net_device *netdev)
 	for (elem = tx_ring->to_clean; elem != tx_ring->to_use;
 	     elem = elem->next) {
 		txp_htxd.flags =
-		    be16_to_cpu(readw(elem->hw_desc + C2_TXP_FLAGS));
+		    be16_to_cpu((__force __be16) readw(elem->hw_desc + C2_TXP_FLAGS));
 
 		if (txp_htxd.flags != TXP_HTXD_DONE)
 			break;
@@ -398,7 +399,7 @@ static void c2_tx_interrupt(struct net_device *netdev)
 		if (netif_msg_tx_done(c2_port)) {
 			/* PCI reads are expensive in fast path */
 			txp_htxd.len =
-			    be16_to_cpu(readw(elem->hw_desc + C2_TXP_LEN));
+			    be16_to_cpu((__force __be16) readw(elem->hw_desc + C2_TXP_LEN));
 			pr_debug("%s: tx done slot %3Zu status 0x%x len "
 				"%5u bytes\n",
 				netdev->name, elem - tx_ring->start,
@@ -448,10 +449,12 @@ static void c2_rx_error(struct c2_port *c2_port, struct c2_element *elem)
 	/* Write the descriptor to the adapter's rx ring */
 	__raw_writew(0, elem->hw_desc + C2_RXP_STATUS);
 	__raw_writew(0, elem->hw_desc + C2_RXP_COUNT);
-	__raw_writew(cpu_to_be16((u16) elem->maplen - sizeof(*rxp_hdr)),
+	__raw_writew((__force u16) cpu_to_be16((u16) elem->maplen - sizeof(*rxp_hdr)),
 		     elem->hw_desc + C2_RXP_LEN);
-	__raw_writeq(cpu_to_be64(elem->mapaddr), elem->hw_desc + C2_RXP_ADDR);
-	__raw_writew(cpu_to_be16(RXP_HRXD_READY), elem->hw_desc + C2_RXP_FLAGS);
+	__raw_writeq((__force u64) cpu_to_be64(elem->mapaddr),
+		     elem->hw_desc + C2_RXP_ADDR);
+	__raw_writew((__force u16) cpu_to_be16(RXP_HRXD_READY),
+		     elem->hw_desc + C2_RXP_FLAGS);
 
 	pr_debug("packet dropped\n");
 	c2_port->netstats.rx_dropped++;
@@ -653,7 +656,7 @@ static int c2_up(struct net_device *netdev)
 	     i++, elem++) {
 		rxp_hdr = (struct c2_rxp_hdr *) elem->skb->data;
 		rxp_hdr->flags = 0;
-		__raw_writew(cpu_to_be16(RXP_HRXD_READY),
+		__raw_writew((__force u16) cpu_to_be16(RXP_HRXD_READY),
 			     elem->hw_desc + C2_RXP_FLAGS);
 	}
 
@@ -787,9 +790,12 @@ static int c2_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 	elem->maplen = maplen;
 
 	/* Tell HW to xmit */
-	__raw_writeq(cpu_to_be64(mapaddr), elem->hw_desc + C2_TXP_ADDR);
-	__raw_writew(cpu_to_be16(maplen), elem->hw_desc + C2_TXP_LEN);
-	__raw_writew(cpu_to_be16(TXP_HTXD_READY), elem->hw_desc + C2_TXP_FLAGS);
+	__raw_writeq((__force u64) cpu_to_be64(mapaddr),
+		     elem->hw_desc + C2_TXP_ADDR);
+	__raw_writew((__force u16) cpu_to_be16(maplen),
+		     elem->hw_desc + C2_TXP_LEN);
+	__raw_writew((__force u16) cpu_to_be16(TXP_HTXD_READY),
+		     elem->hw_desc + C2_TXP_FLAGS);
 
 	c2_port->netstats.tx_packets++;
 	c2_port->netstats.tx_bytes += maplen;
@@ -810,11 +816,11 @@ static int c2_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 			elem->maplen = maplen;
 
 			/* Tell HW to xmit */
-			__raw_writeq(cpu_to_be64(mapaddr),
+			__raw_writeq((__force u64) cpu_to_be64(mapaddr),
 				     elem->hw_desc + C2_TXP_ADDR);
-			__raw_writew(cpu_to_be16(maplen),
+			__raw_writew((__force u16) cpu_to_be16(maplen),
 				     elem->hw_desc + C2_TXP_LEN);
-			__raw_writew(cpu_to_be16(TXP_HTXD_READY),
+			__raw_writew((__force u16) cpu_to_be16(TXP_HTXD_READY),
 				     elem->hw_desc + C2_TXP_FLAGS);
 
 			c2_port->netstats.tx_packets++;
@@ -1029,10 +1035,10 @@ static int __devinit c2_probe(struct pci_dev *pcidev,
 	}
 
 	/* Validate the adapter version */
-	if (be32_to_cpu(readl(mmio_regs + C2_REGS_VERS)) != C2_VERSION) {
+	if (be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_VERS)) != C2_VERSION) {
 		printk(KERN_ERR PFX "Version mismatch "
 			"[fw=%u, c2=%u], Adapter not claimed\n",
-			be32_to_cpu(readl(mmio_regs + C2_REGS_VERS)),
+			be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_VERS)),
 			C2_VERSION);
 		ret = -EINVAL;
 		iounmap(mmio_regs);
@@ -1040,12 +1046,12 @@ static int __devinit c2_probe(struct pci_dev *pcidev,
 	}
 
 	/* Validate the adapter IVN */
-	if (be32_to_cpu(readl(mmio_regs + C2_REGS_IVN)) != C2_IVN) {
+	if (be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_IVN)) != C2_IVN) {
 		printk(KERN_ERR PFX "Downlevel FIrmware level. You should be using "
 		       "the OpenIB device support kit. "
 		       "[fw=0x%x, c2=0x%x], Adapter not claimed\n",
-			be32_to_cpu(readl(mmio_regs + C2_REGS_IVN)),
-			C2_IVN);
+		       be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_IVN)),
+		       C2_IVN);
 		ret = -EINVAL;
 		iounmap(mmio_regs);
 		goto bail2;
@@ -1068,7 +1074,7 @@ static int __devinit c2_probe(struct pci_dev *pcidev,
 
 	/* Get the last RX index */
 	c2dev->cur_rx =
-	    (be32_to_cpu(readl(mmio_regs + C2_REGS_HRX_CUR)) -
+	    (be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_HRX_CUR)) -
 	     0xffffc000) / sizeof(struct c2_rxp_desc);
 
 	/* Request an interrupt line for the driver */
@@ -1090,7 +1096,7 @@ static int __devinit c2_probe(struct pci_dev *pcidev,
 	}
 
 	/* Save off the actual size prior to unmapping mmio_regs */
-	kva_map_size = be32_to_cpu(readl(mmio_regs + C2_REGS_PCI_WINSIZE));
+	kva_map_size = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_PCI_WINSIZE));
 
 	/* Unmap the adapter PCI registers in BAR4 */
 	iounmap(mmio_regs);
diff --git a/drivers/infiniband/hw/amso1100/c2.h b/drivers/infiniband/hw/amso1100/c2.h
index fa58200217a..21bcb3ada7f 100644
--- a/drivers/infiniband/hw/amso1100/c2.h
+++ b/drivers/infiniband/hw/amso1100/c2.h
@@ -346,7 +346,7 @@ struct c2_dev {
 	//	spinlock_t aeq_lock;
 	//	spinlock_t rnic_lock;
 
-	u16 *hint_count;
+	__be16 *hint_count;
 	dma_addr_t hint_count_dma;
 	u16 hints_read;
 
@@ -425,10 +425,10 @@ static inline void __raw_writeq(u64 val, void __iomem * addr)
 #endif
 
 #define C2_SET_CUR_RX(c2dev, cur_rx) \
-	__raw_writel(cpu_to_be32(cur_rx), c2dev->mmio_txp_ring + 4092)
+	__raw_writel((__force u32) cpu_to_be32(cur_rx), c2dev->mmio_txp_ring + 4092)
 
 #define C2_GET_CUR_RX(c2dev) \
-	be32_to_cpu(readl(c2dev->mmio_txp_ring + 4092))
+	be32_to_cpu((__force __be32) readl(c2dev->mmio_txp_ring + 4092))
 
 static inline struct c2_dev *to_c2dev(struct ib_device *ibdev)
 {
@@ -485,8 +485,8 @@ extern void c2_unregister_device(struct c2_dev *c2dev);
 extern int c2_rnic_init(struct c2_dev *c2dev);
 extern void c2_rnic_term(struct c2_dev *c2dev);
 extern void c2_rnic_interrupt(struct c2_dev *c2dev);
-extern int c2_del_addr(struct c2_dev *c2dev, u32 inaddr, u32 inmask);
-extern int c2_add_addr(struct c2_dev *c2dev, u32 inaddr, u32 inmask);
+extern int c2_del_addr(struct c2_dev *c2dev, __be32 inaddr, __be32 inmask);
+extern int c2_add_addr(struct c2_dev *c2dev, __be32 inaddr, __be32 inmask);
 
 /* QPs */
 extern int c2_alloc_qp(struct c2_dev *c2dev, struct c2_pd *pd,
diff --git a/drivers/infiniband/hw/amso1100/c2_ae.c b/drivers/infiniband/hw/amso1100/c2_ae.c
index a31439bd3b6..07ce952699b 100644
--- a/drivers/infiniband/hw/amso1100/c2_ae.c
+++ b/drivers/infiniband/hw/amso1100/c2_ae.c
@@ -195,7 +195,7 @@ void c2_ae_event(struct c2_dev *c2dev, u32 mq_index)
 			"resource=%x, qp_state=%s\n",
 			__FUNCTION__,
 			to_event_str(event_id),
-			(unsigned long long) be64_to_cpu(wr->ae.ae_generic.user_context),
+			(unsigned long long) wr->ae.ae_generic.user_context,
 			be32_to_cpu(wr->ae.ae_generic.resource_type),
 			be32_to_cpu(wr->ae.ae_generic.resource),
 			to_qp_state_str(be32_to_cpu(wr->ae.ae_generic.qp_state)));
diff --git a/drivers/infiniband/hw/amso1100/c2_mm.c b/drivers/infiniband/hw/amso1100/c2_mm.c
index 1e4f46493fc..b506fe22b4d 100644
--- a/drivers/infiniband/hw/amso1100/c2_mm.c
+++ b/drivers/infiniband/hw/amso1100/c2_mm.c
@@ -45,7 +45,7 @@
  *	  Reply buffer _is_ freed by this function.
  */
 static int
-send_pbl_messages(struct c2_dev *c2dev, u32 stag_index,
+send_pbl_messages(struct c2_dev *c2dev, __be32 stag_index,
 		  unsigned long va, u32 pbl_depth,
 		  struct c2_vq_req *vq_req, int pbl_type)
 {
diff --git a/drivers/infiniband/hw/amso1100/c2_mq.c b/drivers/infiniband/hw/amso1100/c2_mq.c
index b88a7559210..0cddc49beae 100644
--- a/drivers/infiniband/hw/amso1100/c2_mq.c
+++ b/drivers/infiniband/hw/amso1100/c2_mq.c
@@ -64,7 +64,7 @@ void c2_mq_produce(struct c2_mq *q)
 		q->priv = (q->priv + 1) % q->q_size;
 		q->hint_count++;
 		/* Update peer's offset. */
-		__raw_writew(cpu_to_be16(q->priv), &q->peer->shared);
+		__raw_writew((__force u16) cpu_to_be16(q->priv), &q->peer->shared);
 	}
 }
 
@@ -105,7 +105,7 @@ void c2_mq_free(struct c2_mq *q)
 #endif
 		q->priv = (q->priv + 1) % q->q_size;
 		/* Update peer's offset. */
-		__raw_writew(cpu_to_be16(q->priv), &q->peer->shared);
+		__raw_writew((__force u16) cpu_to_be16(q->priv), &q->peer->shared);
 	}
 }
 
diff --git a/drivers/infiniband/hw/amso1100/c2_mq.h b/drivers/infiniband/hw/amso1100/c2_mq.h
index 9185bbb2165..acede007b94 100644
--- a/drivers/infiniband/hw/amso1100/c2_mq.h
+++ b/drivers/infiniband/hw/amso1100/c2_mq.h
@@ -75,7 +75,7 @@ struct c2_mq {
 	u16 hint_count;
 	u16 priv;
 	struct c2_mq_shared __iomem *peer;
-	u16 *shared;
+	__be16 *shared;
 	dma_addr_t shared_dma;
 	u32 q_size;
 	u32 msg_size;
diff --git a/drivers/infiniband/hw/amso1100/c2_rnic.c b/drivers/infiniband/hw/amso1100/c2_rnic.c
index 1687c511cb2..7be1f87dcfa 100644
--- a/drivers/infiniband/hw/amso1100/c2_rnic.c
+++ b/drivers/infiniband/hw/amso1100/c2_rnic.c
@@ -208,7 +208,7 @@ static int c2_rnic_query(struct c2_dev *c2dev, struct ib_device_attr *props)
 /*
  * Add an IP address to the RNIC interface
  */
-int c2_add_addr(struct c2_dev *c2dev, u32 inaddr, u32 inmask)
+int c2_add_addr(struct c2_dev *c2dev, __be32 inaddr, __be32 inmask)
 {
 	struct c2_vq_req *vq_req;
 	struct c2wr_rnic_setconfig_req *wr;
@@ -270,7 +270,7 @@ int c2_add_addr(struct c2_dev *c2dev, u32 inaddr, u32 inmask)
 /*
  * Delete an IP address from the RNIC interface
  */
-int c2_del_addr(struct c2_dev *c2dev, u32 inaddr, u32 inmask)
+int c2_del_addr(struct c2_dev *c2dev, __be32 inaddr, __be32 inmask)
 {
 	struct c2_vq_req *vq_req;
 	struct c2wr_rnic_setconfig_req *wr;
@@ -506,17 +506,17 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
 	mmio_regs = c2dev->kva;
 	/* Initialize the Verbs Request Queue */
 	c2_mq_req_init(&c2dev->req_vq, 0,
-		       be32_to_cpu(readl(mmio_regs + C2_REGS_Q0_QSIZE)),
-		       be32_to_cpu(readl(mmio_regs + C2_REGS_Q0_MSGSIZE)),
+		       be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q0_QSIZE)),
+		       be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q0_MSGSIZE)),
 		       mmio_regs +
-		       be32_to_cpu(readl(mmio_regs + C2_REGS_Q0_POOLSTART)),
+		       be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q0_POOLSTART)),
 		       mmio_regs +
-		       be32_to_cpu(readl(mmio_regs + C2_REGS_Q0_SHARED)),
+		       be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q0_SHARED)),
 		       C2_MQ_ADAPTER_TARGET);
 
 	/* Initialize the Verbs Reply Queue */
-	qsize = be32_to_cpu(readl(mmio_regs + C2_REGS_Q1_QSIZE));
-	msgsize = be32_to_cpu(readl(mmio_regs + C2_REGS_Q1_MSGSIZE));
+	qsize = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q1_QSIZE));
+	msgsize = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q1_MSGSIZE));
 	q1_pages = dma_alloc_coherent(&c2dev->pcidev->dev, qsize * msgsize,
 				      &c2dev->rep_vq.host_dma, GFP_KERNEL);
 	if (!q1_pages) {
@@ -532,12 +532,12 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
 		   msgsize,
 		   q1_pages,
 		   mmio_regs +
-		   be32_to_cpu(readl(mmio_regs + C2_REGS_Q1_SHARED)),
+		   be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q1_SHARED)),
 		   C2_MQ_HOST_TARGET);
 
 	/* Initialize the Asynchronus Event Queue */
-	qsize = be32_to_cpu(readl(mmio_regs + C2_REGS_Q2_QSIZE));
-	msgsize = be32_to_cpu(readl(mmio_regs + C2_REGS_Q2_MSGSIZE));
+	qsize = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q2_QSIZE));
+	msgsize = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q2_MSGSIZE));
 	q2_pages = dma_alloc_coherent(&c2dev->pcidev->dev, qsize * msgsize,
 				      &c2dev->aeq.host_dma, GFP_KERNEL);
 	if (!q2_pages) {
@@ -553,7 +553,7 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
 		       msgsize,
 		       q2_pages,
 		       mmio_regs +
-		       be32_to_cpu(readl(mmio_regs + C2_REGS_Q2_SHARED)),
+		       be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q2_SHARED)),
 		       C2_MQ_HOST_TARGET);
 
 	/* Initialize the verbs request allocator */
diff --git a/drivers/infiniband/hw/amso1100/c2_wr.h b/drivers/infiniband/hw/amso1100/c2_wr.h
index 3ec6c43bb0e..c65fbdd6e46 100644
--- a/drivers/infiniband/hw/amso1100/c2_wr.h
+++ b/drivers/infiniband/hw/amso1100/c2_wr.h
@@ -180,8 +180,8 @@ enum c2_wr_type {
 };
 
 struct c2_netaddr {
-	u32 ip_addr;
-	u32 netmask;
+	__be32 ip_addr;
+	__be32 netmask;
 	u32 mtu;
 };
 
@@ -199,9 +199,9 @@ struct c2_route {
  * A Scatter Gather Entry.
  */
 struct c2_data_addr {
-	u32 stag;
-	u32 length;
-	u64 to;
+	__be32 stag;
+	__be32 length;
+	__be64 to;
 };
 
 /*
@@ -274,7 +274,7 @@ struct c2wr_hdr {
 	 * from the host to adapter by libccil, but we copy it anyway
 	 * to make the memcpy to the adapter better aligned.
 	 */
-	u32 wqe_count;
+	__be32 wqe_count;
 
 	/* Put these fields next so that later 32- and 64-bit
 	 * quantities are naturally aligned.
@@ -316,8 +316,8 @@ enum c2_rnic_flags {
 struct c2wr_rnic_open_req {
 	struct c2wr_hdr hdr;
 	u64 user_context;
-	u16 flags;		/* See enum c2_rnic_flags */
-	u16 port_num;
+	__be16 flags;		/* See enum c2_rnic_flags */
+	__be16 port_num;
 } __attribute__((packed));
 
 struct c2wr_rnic_open_rep {
@@ -341,30 +341,30 @@ struct c2wr_rnic_query_req {
 struct c2wr_rnic_query_rep {
 	struct c2wr_hdr hdr;
 	u64 user_context;
-	u32 vendor_id;
-	u32 part_number;
-	u32 hw_version;
-	u32 fw_ver_major;
-	u32 fw_ver_minor;
-	u32 fw_ver_patch;
+	__be32 vendor_id;
+	__be32 part_number;
+	__be32 hw_version;
+	__be32 fw_ver_major;
+	__be32 fw_ver_minor;
+	__be32 fw_ver_patch;
 	char fw_ver_build_str[WR_BUILD_STR_LEN];
-	u32 max_qps;
-	u32 max_qp_depth;
+	__be32 max_qps;
+	__be32 max_qp_depth;
 	u32 max_srq_depth;
 	u32 max_send_sgl_depth;
 	u32 max_rdma_sgl_depth;
-	u32 max_cqs;
-	u32 max_cq_depth;
+	__be32 max_cqs;
+	__be32 max_cq_depth;
 	u32 max_cq_event_handlers;
-	u32 max_mrs;
+	__be32 max_mrs;
 	u32 max_pbl_depth;
-	u32 max_pds;
-	u32 max_global_ird;
+	__be32 max_pds;
+	__be32 max_global_ird;
 	u32 max_global_ord;
-	u32 max_qp_ird;
-	u32 max_qp_ord;
+	__be32 max_qp_ird;
+	__be32 max_qp_ord;
 	u32 flags;
-	u32 max_mws;
+	__be32 max_mws;
 	u32 pbe_range_low;
 	u32 pbe_range_high;
 	u32 max_srqs;
@@ -405,7 +405,7 @@ union c2wr_rnic_getconfig {
 struct c2wr_rnic_setconfig_req {
 	struct c2wr_hdr hdr;
 	u32 rnic_handle;
-	u32 option;		/* See c2_setconfig_cmd_t */
+	__be32 option;		/* See c2_setconfig_cmd_t */
 	/* variable data and pad. See c2_netaddr and c2_route */
 	u8 data[0];
 } __attribute__((packed)) ;
@@ -441,18 +441,18 @@ union c2wr_rnic_close {
  */
 struct c2wr_cq_create_req {
 	struct c2wr_hdr hdr;
-	u64 shared_ht;
+	__be64 shared_ht;
 	u64 user_context;
-	u64 msg_pool;
+	__be64 msg_pool;
 	u32 rnic_handle;
-	u32 msg_size;
-	u32 depth;
+	__be32 msg_size;
+	__be32 depth;
 } __attribute__((packed)) ;
 
 struct c2wr_cq_create_rep {
 	struct c2wr_hdr hdr;
-	u32 mq_index;
-	u32 adapter_shared;
+	__be32 mq_index;
+	__be32 adapter_shared;
 	u32 cq_handle;
 } __attribute__((packed)) ;
 
@@ -585,40 +585,40 @@ enum c2wr_qp_flags {
 
 struct c2wr_qp_create_req {
 	struct c2wr_hdr hdr;
-	u64 shared_sq_ht;
-	u64 shared_rq_ht;
+	__be64 shared_sq_ht;
+	__be64 shared_rq_ht;
 	u64 user_context;
 	u32 rnic_handle;
 	u32 sq_cq_handle;
 	u32 rq_cq_handle;
-	u32 sq_depth;
-	u32 rq_depth;
+	__be32 sq_depth;
+	__be32 rq_depth;
 	u32 srq_handle;
 	u32 srq_limit;
-	u32 flags;		/* see enum c2wr_qp_flags */
-	u32 send_sgl_depth;
-	u32 recv_sgl_depth;
-	u32 rdma_write_sgl_depth;
-	u32 ord;
-	u32 ird;
+	__be32 flags;		/* see enum c2wr_qp_flags */
+	__be32 send_sgl_depth;
+	__be32 recv_sgl_depth;
+	__be32 rdma_write_sgl_depth;
+	__be32 ord;
+	__be32 ird;
 	u32 pd_id;
 } __attribute__((packed)) ;
 
 struct c2wr_qp_create_rep {
 	struct c2wr_hdr hdr;
-	u32 sq_depth;
-	u32 rq_depth;
+	__be32 sq_depth;
+	__be32 rq_depth;
 	u32 send_sgl_depth;
 	u32 recv_sgl_depth;
 	u32 rdma_write_sgl_depth;
 	u32 ord;
 	u32 ird;
-	u32 sq_msg_size;
-	u32 sq_mq_index;
-	u32 sq_mq_start;
-	u32 rq_msg_size;
-	u32 rq_mq_index;
-	u32 rq_mq_start;
+	__be32 sq_msg_size;
+	__be32 sq_mq_index;
+	__be32 sq_mq_start;
+	__be32 rq_msg_size;
+	__be32 rq_mq_index;
+	__be32 rq_mq_start;
 	u32 qp_handle;
 } __attribute__((packed)) ;
 
@@ -667,11 +667,11 @@ struct c2wr_qp_modify_req {
 	u32 stream_msg_length;
 	u32 rnic_handle;
 	u32 qp_handle;
-	u32 next_qp_state;
-	u32 ord;
-	u32 ird;
-	u32 sq_depth;
-	u32 rq_depth;
+	__be32 next_qp_state;
+	__be32 ord;
+	__be32 ird;
+	__be32 sq_depth;
+	__be32 rq_depth;
 	u32 llp_ep_handle;
 } __attribute__((packed)) ;
 
@@ -721,10 +721,10 @@ struct c2wr_qp_connect_req {
 	struct c2wr_hdr hdr;
 	u32 rnic_handle;
 	u32 qp_handle;
-	u32 remote_addr;
-	u16 remote_port;
+	__be32 remote_addr;
+	__be16 remote_port;
 	u16 pad;
-	u32 private_data_length;
+	__be32 private_data_length;
 	u8 private_data[0];	/* Private data in-line. */
 } __attribute__((packed)) ;
 
@@ -759,25 +759,25 @@ union c2wr_nsmr_stag_alloc {
 
 struct c2wr_nsmr_register_req {
 	struct c2wr_hdr hdr;
-	u64 va;
+	__be64 va;
 	u32 rnic_handle;
-	u16 flags;
+	__be16 flags;
 	u8 stag_key;
 	u8 pad;
 	u32 pd_id;
-	u32 pbl_depth;
-	u32 pbe_size;
-	u32 fbo;
-	u32 length;
-	u32 addrs_length;
+	__be32 pbl_depth;
+	__be32 pbe_size;
+	__be32 fbo;
+	__be32 length;
+	__be32 addrs_length;
 	/* array of paddrs (must be aligned on a 64bit boundary) */
-	u64 paddrs[0];
+	__be64 paddrs[0];
 } __attribute__((packed)) ;
 
 struct c2wr_nsmr_register_rep {
 	struct c2wr_hdr hdr;
 	u32 pbl_depth;
-	u32 stag_index;
+	__be32 stag_index;
 } __attribute__((packed)) ;
 
 union c2wr_nsmr_register {
@@ -788,11 +788,11 @@ union c2wr_nsmr_register {
 struct c2wr_nsmr_pbl_req {
 	struct c2wr_hdr hdr;
 	u32 rnic_handle;
-	u32 flags;
-	u32 stag_index;
-	u32 addrs_length;
+	__be32 flags;
+	__be32 stag_index;
+	__be32 addrs_length;
 	/* array of paddrs (must be aligned on a 64bit boundary) */
-	u64 paddrs[0];
+	__be64 paddrs[0];
 } __attribute__((packed)) ;
 
 struct c2wr_nsmr_pbl_rep {
@@ -847,7 +847,7 @@ union c2wr_mw_query {
 struct c2wr_stag_dealloc_req {
 	struct c2wr_hdr hdr;
 	u32 rnic_handle;
-	u32 stag_index;
+	__be32 stag_index;
 } __attribute__((packed)) ;
 
 struct c2wr_stag_dealloc_rep {
@@ -949,7 +949,7 @@ struct c2wr_ce {
 	u64 qp_user_context;	/* c2_user_qp_t * */
 	u32 qp_state;		/* Current QP State */
 	u32 handle;		/* QPID or EP Handle */
-	u32 bytes_rcvd;		/* valid for RECV WCs */
+	__be32 bytes_rcvd;		/* valid for RECV WCs */
 	u32 stag;
 } __attribute__((packed)) ;
 
@@ -984,8 +984,8 @@ struct c2_rq_hdr {
  */
 struct c2wr_send_req {
 	struct c2_sq_hdr sq_hdr;
-	u32 sge_len;
-	u32 remote_stag;
+	__be32 sge_len;
+	__be32 remote_stag;
 	u8 data[0];		/* SGE array */
 } __attribute__((packed));
 
@@ -996,9 +996,9 @@ union c2wr_send {
 
 struct c2wr_rdma_write_req {
 	struct c2_sq_hdr sq_hdr;
-	u64 remote_to;
-	u32 remote_stag;
-	u32 sge_len;
+	__be64 remote_to;
+	__be32 remote_stag;
+	__be32 sge_len;
 	u8 data[0];		/* SGE array */
 } __attribute__((packed));
 
@@ -1009,11 +1009,11 @@ union c2wr_rdma_write {
 
 struct c2wr_rdma_read_req {
 	struct c2_sq_hdr sq_hdr;
-	u64 local_to;
-	u64 remote_to;
-	u32 local_stag;
-	u32 remote_stag;
-	u32 length;
+	__be64 local_to;
+	__be64 remote_to;
+	__be32 local_stag;
+	__be32 remote_stag;
+	__be32 length;
 } __attribute__((packed));
 
 union c2wr_rdma_read {
@@ -1113,9 +1113,9 @@ union c2wr_recv {
 struct c2wr_ae_hdr {
 	struct c2wr_hdr hdr;
 	u64 user_context;	/* user context for this res. */
-	u32 resource_type;	/* see enum c2_resource_indicator */
-	u32 resource;		/* handle for resource */
-	u32 qp_state;		/* current QP State */
+	__be32 resource_type;	/* see enum c2_resource_indicator */
+	__be32 resource;	/* handle for resource */
+	__be32 qp_state;	/* current QP State */
 } __attribute__((packed));
 
 /*
@@ -1124,11 +1124,11 @@ struct c2wr_ae_hdr {
  */
 struct c2wr_ae_active_connect_results {
 	struct c2wr_ae_hdr ae_hdr;
-	u32 laddr;
-	u32 raddr;
-	u16 lport;
-	u16 rport;
-	u32 private_data_length;
+	__be32 laddr;
+	__be32 raddr;
+	__be16 lport;
+	__be16 rport;
+	__be32 private_data_length;
 	u8 private_data[0];	/* data is in-line in the msg. */
 } __attribute__((packed));
 
@@ -1142,11 +1142,11 @@ struct c2wr_ae_active_connect_results {
 struct c2wr_ae_connection_request {
 	struct c2wr_ae_hdr ae_hdr;
 	u32 cr_handle;		/* connreq handle (sock ptr) */
-	u32 laddr;
-	u32 raddr;
-	u16 lport;
-	u16 rport;
-	u32 private_data_length;
+	__be32 laddr;
+	__be32 raddr;
+	__be16 lport;
+	__be16 rport;
+	__be32 private_data_length;
 	u8 private_data[0];	/* data is in-line in the msg. */
 } __attribute__((packed));
 
@@ -1158,12 +1158,12 @@ union c2wr_ae {
 
 struct c2wr_init_req {
 	struct c2wr_hdr hdr;
-	u64 hint_count;
-	u64 q0_host_shared;
-	u64 q1_host_shared;
-	u64 q1_host_msg_pool;
-	u64 q2_host_shared;
-	u64 q2_host_msg_pool;
+	__be64 hint_count;
+	__be64 q0_host_shared;
+	__be64 q1_host_shared;
+	__be64 q1_host_msg_pool;
+	__be64 q2_host_shared;
+	__be64 q2_host_msg_pool;
 } __attribute__((packed));
 
 struct c2wr_init_rep {
@@ -1276,10 +1276,10 @@ struct c2wr_ep_listen_create_req {
 	struct c2wr_hdr hdr;
 	u64 user_context;	/* returned in AEs. */
 	u32 rnic_handle;
-	u32 local_addr;		/* local addr, or 0  */
-	u16 local_port;		/* 0 means "pick one" */
+	__be32 local_addr;		/* local addr, or 0  */
+	__be16 local_port;		/* 0 means "pick one" */
 	u16 pad;
-	u32 backlog;		/* tradional tcp listen bl */
+	__be32 backlog;		/* tradional tcp listen bl */
 } __attribute__((packed));
 
 struct c2wr_ep_listen_create_rep {
@@ -1340,7 +1340,7 @@ struct c2wr_cr_accept_req {
 	u32 rnic_handle;
 	u32 qp_handle;		/* QP to bind to this LLP conn */
 	u32 ep_handle;		/* LLP  handle to accept */
-	u32 private_data_length;
+	__be32 private_data_length;
 	u8 private_data[0];	/* data in-line in msg. */
 } __attribute__((packed));
 
@@ -1508,7 +1508,7 @@ static __inline__ void c2_wr_set_sge_count(void *wr, u8 sge_count)
 {
 	((struct c2wr_hdr *) wr)->sge_count = sge_count;
 }
-static __inline__ u32 c2_wr_get_wqe_count(void *wr)
+static __inline__ __be32 c2_wr_get_wqe_count(void *wr)
 {
 	return ((struct c2wr_hdr *) wr)->wqe_count;
 }