net: via-rhine: add BQL support

Add Byte Queue Limits (BQL) support to via-rhine driver.

[edumazet] tweaked patch and changed TX_RING_SIZE from 16 to 64
Signed-off-by: Tino Reichardt <milky-kernel@mcmilk.de>
Tested-by: Jamie Gloudon <jamie.gloudon@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/via/via-rhine.c

@@ -70,12 +70,14 @@ static const int multicast_filter_limit = 32;
 /* Operational parameters that are set at compile time. */
 
 /* Keep the ring sizes a power of two for compile efficiency.
-   The compiler will convert <unsigned>'%'<2^N> into a bit mask.
-   Making the Tx ring too large decreases the effectiveness of channel
-   bonding and packet priority.
-   There are no ill effects from too-large receive rings. */
-#define TX_RING_SIZE	16
-#define TX_QUEUE_LEN	10	/* Limit ring entries actually used. */
+ * The compiler will convert <unsigned>'%'<2^N> into a bit mask.
+ * Making the Tx ring too large decreases the effectiveness of channel
+ * bonding and packet priority.
+ * With BQL support, we can increase TX ring safely.
+ * There are no ill effects from too-large receive rings.
+ */
+#define TX_RING_SIZE	64
+#define TX_QUEUE_LEN	(TX_RING_SIZE - 6)	/* Limit ring entries actually used. */
 #define RX_RING_SIZE	64
 
 /* Operational parameters that usually are not changed. */
@@ -1295,6 +1297,7 @@ static void alloc_tbufs(struct net_device* dev)
 	}
 	rp->tx_ring[i-1].next_desc = cpu_to_le32(rp->tx_ring_dma);
 
+	netdev_reset_queue(dev);
 }
 
 static void free_tbufs(struct net_device* dev)
@@ -1795,6 +1798,7 @@ static netdev_tx_t rhine_start_tx(struct sk_buff *skb,
 	else
 		rp->tx_ring[entry].tx_status = 0;
 
+	netdev_sent_queue(dev, skb->len);
 	/* lock eth irq */
 	wmb();
 	rp->tx_ring[entry].tx_status |= cpu_to_le32(DescOwn);
@@ -1863,6 +1867,8 @@ static void rhine_tx(struct net_device *dev)
 	struct rhine_private *rp = netdev_priv(dev);
 	struct device *hwdev = dev->dev.parent;
 	int txstatus = 0, entry = rp->dirty_tx % TX_RING_SIZE;
+	unsigned int pkts_compl = 0, bytes_compl = 0;
+	struct sk_buff *skb;
 
 	/* find and cleanup dirty tx descriptors */
 	while (rp->dirty_tx != rp->cur_tx) {
@@ -1871,6 +1877,7 @@ static void rhine_tx(struct net_device *dev)
 			  entry, txstatus);
 		if (txstatus & DescOwn)
 			break;
+		skb = rp->tx_skbuff[entry];
 		if (txstatus & 0x8000) {
 			netif_dbg(rp, tx_done, dev,
 				  "Transmit error, Tx status %08x\n", txstatus);
@@ -1899,7 +1906,7 @@ static void rhine_tx(struct net_device *dev)
 				  (txstatus >> 3) & 0xF, txstatus & 0xF);
 
 			u64_stats_update_begin(&rp->tx_stats.syncp);
-			rp->tx_stats.bytes += rp->tx_skbuff[entry]->len;
+			rp->tx_stats.bytes += skb->len;
 			rp->tx_stats.packets++;
 			u64_stats_update_end(&rp->tx_stats.syncp);
 		}
@@ -1907,13 +1914,17 @@ static void rhine_tx(struct net_device *dev)
 		if (rp->tx_skbuff_dma[entry]) {
 			dma_unmap_single(hwdev,
 					 rp->tx_skbuff_dma[entry],
-					 rp->tx_skbuff[entry]->len,
+					 skb->len,
 					 DMA_TO_DEVICE);
 		}
-		dev_consume_skb_any(rp->tx_skbuff[entry]);
+		bytes_compl += skb->len;
+		pkts_compl++;
+		dev_consume_skb_any(skb);
 		rp->tx_skbuff[entry] = NULL;
 		entry = (++rp->dirty_tx) % TX_RING_SIZE;
 	}
+
+	netdev_completed_queue(dev, pkts_compl, bytes_compl);
 	if ((rp->cur_tx - rp->dirty_tx) < TX_QUEUE_LEN - 4)
 		netif_wake_queue(dev);
 }