Merge branch 'dma_rmb_wmb'

Alexander Duyck says:

====================
Replace wmb()/rmb() with dma_wmb()/dma_rmb() where appropriate

This is a start of a side project cleaning up the drivers that can make use
of the dma_wmb and dma_rmb calls.  The general idea is to start removing
the unnecessary wmb/rmb calls from a number of drivers and to make use of
the lighter weight dma_wmb/dma_rmb calls as this should allow for an
overall improvement in performance as each barrier can cost a significant
number of cycles and on architectures such as x86 this is unnecessary.

These changes are what I would consider low hanging fruit.  The likelihood
of the changes introducing an error should be low since the use of the
barriers in these cases are fairly obvious.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/intel/e1000/e1000_main.c

@@ -3856,7 +3856,7 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
 	while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 	       (count < tx_ring->count)) {
 		bool cleaned = false;
-		rmb();	/* read buffer_info after eop_desc */
+		dma_rmb();	/* read buffer_info after eop_desc */
 		for ( ; !cleaned; count++) {
 			tx_desc = E1000_TX_DESC(*tx_ring, i);
 			buffer_info = &tx_ring->buffer_info[i];
@@ -4154,7 +4154,7 @@ static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 		if (*work_done >= work_to_do)
 			break;
 		(*work_done)++;
-		rmb(); /* read descriptor and rx_buffer_info after status DD */
+		dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
 
 		status = rx_desc->status;
 
@@ -4375,7 +4375,7 @@ static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 		if (*work_done >= work_to_do)
 			break;
 		(*work_done)++;
-		rmb(); /* read descriptor and rx_buffer_info after status DD */
+		dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
 
 		status = rx_desc->status;
 		length = le16_to_cpu(rx_desc->length);

drivers/net/ethernet/intel/e1000e/netdev.c

@@ -947,7 +947,7 @@ static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
 		if (*work_done >= work_to_do)
 			break;
 		(*work_done)++;
-		rmb();	/* read descriptor and rx_buffer_info after status DD */
+		dma_rmb();	/* read descriptor and rx_buffer_info after status DD */
 
 		skb = buffer_info->skb;
 		buffer_info->skb = NULL;
@@ -1232,7 +1232,7 @@ static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
 	       (count < tx_ring->count)) {
 		bool cleaned = false;
 
-		rmb();		/* read buffer_info after eop_desc */
+		dma_rmb();		/* read buffer_info after eop_desc */
 		for (; !cleaned; count++) {
 			tx_desc = E1000_TX_DESC(*tx_ring, i);
 			buffer_info = &tx_ring->buffer_info[i];
@@ -1332,7 +1332,7 @@ static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
 			break;
 		(*work_done)++;
 		skb = buffer_info->skb;
-		rmb();	/* read descriptor and rx_buffer_info after status DD */
+		dma_rmb();	/* read descriptor and rx_buffer_info after status DD */
 
 		/* in the packet split case this is header only */
 		prefetch(skb->data - NET_IP_ALIGN);
@@ -1536,7 +1536,7 @@ static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
 		if (*work_done >= work_to_do)
 			break;
 		(*work_done)++;
-		rmb();	/* read descriptor and rx_buffer_info after status DD */
+		dma_rmb();	/* read descriptor and rx_buffer_info after status DD */
 
 		skb = buffer_info->skb;
 		buffer_info->skb = NULL;

drivers/net/ethernet/neterion/s2io.c

@@ -2520,7 +2520,7 @@ static int fill_rx_buffers(struct s2io_nic *nic, struct ring_info *ring,
 			DBG_PRINT(INFO_DBG, "%s: Could not allocate skb\n",
 				  ring->dev->name);
 			if (first_rxdp) {
-				wmb();
+				dma_wmb();
 				first_rxdp->Control_1 |= RXD_OWN_XENA;
 			}
 			swstats->mem_alloc_fail_cnt++;
@@ -2634,7 +2634,7 @@ static int fill_rx_buffers(struct s2io_nic *nic, struct ring_info *ring,
 		rxdp->Control_2 |= SET_RXD_MARKER;
 		if (!(alloc_tab & ((1 << rxsync_frequency) - 1))) {
 			if (first_rxdp) {
-				wmb();
+				dma_wmb();
 				first_rxdp->Control_1 |= RXD_OWN_XENA;
 			}
 			first_rxdp = rxdp;
@@ -2649,7 +2649,7 @@ end:
 	 * and other fields are seen by adapter correctly.
 	 */
 	if (first_rxdp) {
-		wmb();
+		dma_wmb();
 		first_rxdp->Control_1 |= RXD_OWN_XENA;
 	}
 
@@ -6950,7 +6950,7 @@ static  int rxd_owner_bit_reset(struct s2io_nic *sp)
 				}
 
 				set_rxd_buffer_size(sp, rxdp, size);
-				wmb();
+				dma_wmb();
 				/* flip the Ownership bit to Hardware */
 				rxdp->Control_1 |= RXD_OWN_XENA;
 			}

drivers/net/ethernet/sun/sungem.c

@@ -718,7 +718,7 @@ static __inline__ void gem_post_rxds(struct gem *gp, int limit)
 	cluster_start = curr = (gp->rx_new & ~(4 - 1));
 	count = 0;
 	kick = -1;
-	wmb();
+	dma_wmb();
 	while (curr != limit) {
 		curr = NEXT_RX(curr);
 		if (++count == 4) {
@@ -1038,7 +1038,7 @@ static netdev_tx_t gem_start_xmit(struct sk_buff *skb,
 		if (gem_intme(entry))
 			ctrl |= TXDCTRL_INTME;
 		txd->buffer = cpu_to_le64(mapping);
-		wmb();
+		dma_wmb();
 		txd->control_word = cpu_to_le64(ctrl);
 		entry = NEXT_TX(entry);
 	} else {
@@ -1076,7 +1076,7 @@ static netdev_tx_t gem_start_xmit(struct sk_buff *skb,
 
 			txd = &gp->init_block->txd[entry];
 			txd->buffer = cpu_to_le64(mapping);
-			wmb();
+			dma_wmb();
 			txd->control_word = cpu_to_le64(this_ctrl | len);
 
 			if (gem_intme(entry))
@@ -1086,7 +1086,7 @@ static netdev_tx_t gem_start_xmit(struct sk_buff *skb,
 		}
 		txd = &gp->init_block->txd[first_entry];
 		txd->buffer = cpu_to_le64(first_mapping);
-		wmb();
+		dma_wmb();
 		txd->control_word =
 			cpu_to_le64(ctrl | TXDCTRL_SOF | intme | first_len);
 	}
@@ -1585,7 +1585,7 @@ static void gem_clean_rings(struct gem *gp)
 			gp->rx_skbs[i] = NULL;
 		}
 		rxd->status_word = 0;
-		wmb();
+		dma_wmb();
 		rxd->buffer = 0;
 	}
 
@@ -1647,7 +1647,7 @@ static void gem_init_rings(struct gem *gp)
 					RX_BUF_ALLOC_SIZE(gp),
 					PCI_DMA_FROMDEVICE);
 		rxd->buffer = cpu_to_le64(dma_addr);
-		wmb();
+		dma_wmb();
 		rxd->status_word = cpu_to_le64(RXDCTRL_FRESH(gp));
 		skb_reserve(skb, RX_OFFSET);
 	}
@@ -1656,7 +1656,7 @@ static void gem_init_rings(struct gem *gp)
 		struct gem_txd *txd = &gb->txd[i];
 
 		txd->control_word = 0;
-		wmb();
+		dma_wmb();
 		txd->buffer = 0;
 	}
 	wmb();

drivers/net/ethernet/sun/sunhme.c

@@ -196,14 +196,14 @@ static u32 sbus_hme_read32(void __iomem *reg)
 static void sbus_hme_write_rxd(struct happy_meal_rxd *rxd, u32 flags, u32 addr)
 {
 	rxd->rx_addr = (__force hme32)addr;
-	wmb();
+	dma_wmb();
 	rxd->rx_flags = (__force hme32)flags;
 }
 
 static void sbus_hme_write_txd(struct happy_meal_txd *txd, u32 flags, u32 addr)
 {
 	txd->tx_addr = (__force hme32)addr;
-	wmb();
+	dma_wmb();
 	txd->tx_flags = (__force hme32)flags;
 }
 
@@ -225,14 +225,14 @@ static u32 pci_hme_read32(void __iomem *reg)
 static void pci_hme_write_rxd(struct happy_meal_rxd *rxd, u32 flags, u32 addr)
 {
 	rxd->rx_addr = (__force hme32)cpu_to_le32(addr);
-	wmb();
+	dma_wmb();
 	rxd->rx_flags = (__force hme32)cpu_to_le32(flags);
 }
 
 static void pci_hme_write_txd(struct happy_meal_txd *txd, u32 flags, u32 addr)
 {
 	txd->tx_addr = (__force hme32)cpu_to_le32(addr);
-	wmb();
+	dma_wmb();
 	txd->tx_flags = (__force hme32)cpu_to_le32(flags);
 }
 
@@ -268,12 +268,12 @@ static u32 pci_hme_read_desc32(hme32 *p)
 	sbus_readl(__reg)
 #define hme_write_rxd(__hp, __rxd, __flags, __addr) \
 do {	(__rxd)->rx_addr = (__force hme32)(u32)(__addr); \
-	wmb(); \
+	dma_wmb(); \
 	(__rxd)->rx_flags = (__force hme32)(u32)(__flags); \
 } while(0)
 #define hme_write_txd(__hp, __txd, __flags, __addr) \
 do {	(__txd)->tx_addr = (__force hme32)(u32)(__addr); \
-	wmb(); \
+	dma_wmb(); \
 	(__txd)->tx_flags = (__force hme32)(u32)(__flags); \
 } while(0)
 #define hme_read_desc32(__hp, __p)	((__force u32)(hme32)*(__p))
@@ -293,12 +293,12 @@ do {	(__txd)->tx_addr = (__force hme32)(u32)(__addr); \
 	readl(__reg)
 #define hme_write_rxd(__hp, __rxd, __flags, __addr) \
 do {	(__rxd)->rx_addr = (__force hme32)cpu_to_le32(__addr); \
-	wmb(); \
+	dma_wmb(); \
 	(__rxd)->rx_flags = (__force hme32)cpu_to_le32(__flags); \
 } while(0)
 #define hme_write_txd(__hp, __txd, __flags, __addr) \
 do {	(__txd)->tx_addr = (__force hme32)cpu_to_le32(__addr); \
-	wmb(); \
+	dma_wmb(); \
 	(__txd)->tx_flags = (__force hme32)cpu_to_le32(__flags); \
 } while(0)
 static inline u32 hme_read_desc32(struct happy_meal *hp, hme32 *p)

drivers/net/ethernet/sun/sunvnet.c

@@ -519,7 +519,7 @@ static int vnet_walk_rx_one(struct vnet_port *port,
 	if (desc->hdr.state != VIO_DESC_READY)
 		return 1;
 
-	rmb();
+	dma_rmb();
 
 	viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
 	       desc->hdr.state, desc->hdr.ack,
@@ -1380,7 +1380,7 @@ static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	/* This has to be a non-SMP write barrier because we are writing
 	 * to memory which is shared with the peer LDOM.
 	 */
-	wmb();
+	dma_wmb();
 
 	d->hdr.state = VIO_DESC_READY;
 
@@ -1395,7 +1395,7 @@ static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	 * is marked READY, but start_cons was false.
 	 * If so, vnet_ack() should send out the missed "start" trigger.
 	 *
-	 * Note that the wmb() above makes sure the cookies et al. are
+	 * Note that the dma_wmb() above makes sure the cookies et al. are
 	 * not globally visible before the VIO_DESC_READY, and that the
 	 * stores are ordered correctly by the compiler. The consumer will
 	 * not proceed until the VIO_DESC_READY is visible assuring that