ramips: raeth: add tx_info structure

This helps to reduce code size on TX path:

add/remove: 0/0 grow/shrink: 1/4 up/down: 56/-92 (-36)
function                                     old     new   delta
ramips_eth_open                             1048    1104     +56
ramips_eth_tx_housekeeping                   156     148      -8
ramips_ring_free                             224     212     -12
ramips_eth_stop                              420     400     -20
ramips_eth_hard_start_xmit                   480     428     -52

SVN-Revision: 30731

target/linux/ramips/files/drivers/net/ethernet/ramips/ramips_eth.h

@@ -213,6 +213,11 @@ struct ramips_tx_dma {
 	unsigned int txd4;
 } __packed __aligned(4);
 
+struct raeth_tx_info {
+	struct ramips_tx_dma	*tx_desc;
+	struct sk_buff		*tx_skb;
+};
+
 struct raeth_int_stats {
 	unsigned long		rx_delayed;
 	unsigned long		tx_delayed;
@@ -247,10 +252,10 @@ struct raeth_priv
 	struct sk_buff		*rx_skb[NUM_RX_DESC];
 	dma_addr_t		rx_dma[NUM_RX_DESC];
 
+	struct raeth_tx_info	*tx_info;
 	dma_addr_t		tx_desc_dma;
 	struct tasklet_struct	tx_housekeeping_tasklet;
 	struct ramips_tx_dma	*tx;
-	struct sk_buff		*tx_skb[NUM_TX_DESC];
 
 	unsigned int		skb_free_idx;
 

target/linux/ramips/files/drivers/net/ethernet/ramips/ramips_main.c

@@ -108,20 +108,25 @@ ramips_ring_setup(struct raeth_priv *re)
 	int len;
 	int i;
 
+	memset(re->tx_info, 0, NUM_TX_DESC * sizeof(struct raeth_tx_info));
+
 	len = NUM_TX_DESC * sizeof(struct ramips_tx_dma);
 	memset(re->tx, 0, len);
 
 	for (i = 0; i < NUM_TX_DESC; i++) {
+		struct raeth_tx_info *txi;
 		struct ramips_tx_dma *txd;
 
 		txd = &re->tx[i];
 		txd->txd4 = TX_DMA_QN(3) | TX_DMA_PN(1);
 		txd->txd2 = TX_DMA_LSO | TX_DMA_DONE;
 
-		if (re->tx_skb[i] != NULL) {
+		txi = &re->tx_info[i];
+		txi->tx_desc = txd;
+		if (txi->tx_skb != NULL) {
 			netdev_warn(re->netdev,
 				    "dirty skb for TX desc %d\n", i);
-			re->tx_skb[i] = NULL;
+			txi->tx_skb = NULL;
 		}
 	}
 
@@ -153,10 +158,14 @@ ramips_ring_cleanup(struct raeth_priv *re)
 			dma_unmap_single(&re->netdev->dev, re->rx_dma[i],
 					 MAX_RX_LENGTH, DMA_FROM_DEVICE);
 
-	for (i = 0; i < NUM_TX_DESC; i++)
-		if (re->tx_skb[i]) {
-			dev_kfree_skb_any(re->tx_skb[i]);
-			re->tx_skb[i] = NULL;
+	for (i = 0; i < NUM_TX_DESC; i++) {
+		struct raeth_tx_info *txi;
+
+		txi = &re->tx_info[i];
+		if (txi->tx_skb) {
+			dev_kfree_skb_any(txi->tx_skb);
+			txi->tx_skb = NULL;
+		}
 	}
 }
 
@@ -559,6 +568,8 @@ ramips_ring_free(struct raeth_priv *re)
 		dma_free_coherent(&re->netdev->dev, len, re->tx,
 				  re->tx_desc_dma);
 	}
+
+	kfree(re->tx_info);
 }
 
 static int
@@ -568,6 +579,11 @@ ramips_ring_alloc(struct raeth_priv *re)
 	int err = -ENOMEM;
 	int i;
 
+	re->tx_info = kzalloc(NUM_TX_DESC * sizeof(struct raeth_tx_info),
+			      GFP_ATOMIC);
+	if (!re->tx_info)
+		goto err_cleanup;
+
 	/* allocate tx ring */
 	len = NUM_TX_DESC * sizeof(struct ramips_tx_dma);
 	re->tx = dma_alloc_coherent(&re->netdev->dev, len,
@@ -617,6 +633,8 @@ static int
 ramips_eth_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
 {
 	struct raeth_priv *re = netdev_priv(dev);
+	struct raeth_tx_info *txi, *txi_next;
+	struct ramips_tx_dma *txd, *txd_next;
 	unsigned long tx;
 	unsigned int tx_next;
 	dma_addr_t mapped_addr;
@@ -641,18 +659,23 @@ ramips_eth_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	tx = ramips_fe_rr(RAMIPS_TX_CTX_IDX0);
 	tx_next = (tx + 1) % NUM_TX_DESC;
 
-	if ((re->tx_skb[tx]) || (re->tx_skb[tx_next]) ||
-	    !(re->tx[tx].txd2 & TX_DMA_DONE) ||
-	    !(re->tx[tx_next].txd2 & TX_DMA_DONE))
+	txi = &re->tx_info[tx];
+	txd = txi->tx_desc;
+	txi_next = &re->tx_info[tx_next];
+	txd_next = txi_next->tx_desc;
+
+	if ((txi->tx_skb) || (txi_next->tx_skb) ||
+	    !(txd->txd2 & TX_DMA_DONE) ||
+	    !(txd_next->txd2 & TX_DMA_DONE))
 		goto out;
 
-	re->tx[tx].txd1 = (unsigned int) mapped_addr;
-	re->tx[tx].txd2 &= ~(TX_DMA_PLEN0_MASK | TX_DMA_DONE);
-	re->tx[tx].txd2 |= TX_DMA_PLEN0(skb->len);
+	txi->tx_skb = skb;
+
+	txd->txd1 = (unsigned int) mapped_addr;
+	wmb();
+	txd->txd2 = TX_DMA_LSO | TX_DMA_PLEN0(skb->len);
 	dev->stats.tx_packets++;
 	dev->stats.tx_bytes += skb->len;
-	re->tx_skb[tx] = skb;
-	wmb();
 	ramips_fe_wr(tx_next, RAMIPS_TX_CTX_IDX0);
 	spin_unlock(&re->page_lock);
 	return NETDEV_TX_OK;
@@ -730,10 +753,18 @@ ramips_eth_tx_housekeeping(unsigned long ptr)
 	struct raeth_priv *re = netdev_priv(dev);
 
 	spin_lock(&re->page_lock);
-	while ((re->tx[re->skb_free_idx].txd2 & TX_DMA_DONE) &&
-	       (re->tx_skb[re->skb_free_idx])) {
-		dev_kfree_skb_irq(re->tx_skb[re->skb_free_idx]);
-		re->tx_skb[re->skb_free_idx] = 0;
+	while (1) {
+		struct raeth_tx_info *txi;
+		struct ramips_tx_dma *txd;
+
+		txi = &re->tx_info[re->skb_free_idx];
+		txd = txi->tx_desc;
+
+		if (!(txd->txd2 & TX_DMA_DONE) || !(txi->tx_skb))
+			break;
+
+		dev_kfree_skb_irq(txi->tx_skb);
+		txi->tx_skb = NULL;
 		re->skb_free_idx++;
 		if (re->skb_free_idx >= NUM_TX_DESC)
 			re->skb_free_idx = 0;