forcedeth: TX handler changes (experimental)

@@ -87,6 +87,8 @@
  *	0.35: 26 Jun 2005: Support for MCP55 added.
  *	0.36: 28 Jun 2005: Add jumbo frame support.
  *	0.37: 10 Jul 2005: Additional ethtool support, cleanup of pci id list
+ *	0.38: 16 Jul 2005: tx irq rewrite: Use global flags instead of
+ *			   per-packet flags.
  *
  * Known bugs:
  * We suspect that on some hardware no TX done interrupts are generated.

@@ -98,7 +100,7 @@
  * DEV_NEED_TIMERIRQ will not harm you on sane hardware, only generating a few
  * superfluous timer interrupts from the nic.
  */
-#define FORCEDETH_VERSION		"0.36"
+#define FORCEDETH_VERSION		"0.38"
 #define DRV_NAME			"forcedeth"
 
 #include <linux/module.h>

@@ -133,12 +135,9 @@
  * Hardware access:
  */
 
-#define DEV_NEED_LASTPACKET1	0x0001	/* set LASTPACKET1 in tx flags */
-#define DEV_IRQMASK_1		0x0002  /* use NVREG_IRQMASK_WANTED_1 for irq mask */
-#define DEV_IRQMASK_2		0x0004  /* use NVREG_IRQMASK_WANTED_2 for irq mask */
-#define DEV_NEED_TIMERIRQ	0x0008  /* set the timer irq flag in the irq mask */
-#define DEV_NEED_LINKTIMER	0x0010	/* poll link settings. Relies on the timer irq */
-#define DEV_HAS_LARGEDESC	0x0020	/* device supports jumbo frames and needs packet format 2 */
+#define DEV_NEED_TIMERIRQ	0x0001  /* set the timer irq flag in the irq mask */
+#define DEV_NEED_LINKTIMER	0x0002	/* poll link settings. Relies on the timer irq */
+#define DEV_HAS_LARGEDESC	0x0003	/* device supports jumbo frames and needs packet format 2 */
 
 enum {
 	NvRegIrqStatus = 0x000,

@@ -149,13 +148,16 @@
 #define NVREG_IRQ_RX			0x0002
 #define NVREG_IRQ_RX_NOBUF		0x0004
 #define NVREG_IRQ_TX_ERR		0x0008
-#define NVREG_IRQ_TX2			0x0010
+#define NVREG_IRQ_TX_OK			0x0010
 #define NVREG_IRQ_TIMER			0x0020
 #define NVREG_IRQ_LINK			0x0040
+#define NVREG_IRQ_TX_ERROR		0x0080
 #define NVREG_IRQ_TX1			0x0100
-#define NVREG_IRQMASK_WANTED_1		0x005f
-#define NVREG_IRQMASK_WANTED_2		0x0147
-#define NVREG_IRQ_UNKNOWN		(~(NVREG_IRQ_RX_ERROR|NVREG_IRQ_RX|NVREG_IRQ_RX_NOBUF|NVREG_IRQ_TX_ERR|NVREG_IRQ_TX2|NVREG_IRQ_TIMER|NVREG_IRQ_LINK|NVREG_IRQ_TX1))
+#define NVREG_IRQMASK_WANTED		0x00df
+
+#define NVREG_IRQ_UNKNOWN	(~(NVREG_IRQ_RX_ERROR|NVREG_IRQ_RX|NVREG_IRQ_RX_NOBUF|NVREG_IRQ_TX_ERR| \
+					NVREG_IRQ_TX_OK|NVREG_IRQ_TIMER|NVREG_IRQ_LINK|NVREG_IRQ_TX_ERROR| \
+					NVREG_IRQ_TX1))
 
 	NvRegUnknownSetupReg6 = 0x008,
 #define NVREG_UNKSETUP6_VAL		3

@@ -296,7 +298,7 @@
 
 #define NV_TX_LASTPACKET	(1<<16)
 #define NV_TX_RETRYERROR	(1<<19)
-#define NV_TX_LASTPACKET1	(1<<24)
+#define NV_TX_FORCED_INTERRUPT	(1<<24)
 #define NV_TX_DEFERRED		(1<<26)
 #define NV_TX_CARRIERLOST	(1<<27)
 #define NV_TX_LATECOLLISION	(1<<28)

@@ -306,7 +308,7 @@
 
 #define NV_TX2_LASTPACKET	(1<<29)
 #define NV_TX2_RETRYERROR	(1<<18)
-#define NV_TX2_LASTPACKET1	(1<<23)
+#define NV_TX2_FORCED_INTERRUPT	(1<<30)
 #define NV_TX2_DEFERRED		(1<<25)
 #define NV_TX2_CARRIERLOST	(1<<26)
 #define NV_TX2_LATECOLLISION	(1<<27)

@@ -1013,9 +1015,39 @@
 	struct fe_priv *np = get_nvpriv(dev);
 	u8 __iomem *base = get_hwbase(dev);
 
-	dprintk(KERN_DEBUG "%s: Got tx_timeout. irq: %08x\n", dev->name,
+	printk(KERN_INFO "%s: Got tx_timeout. irq: %08x\n", dev->name,
 			readl(base + NvRegIrqStatus) & NVREG_IRQSTAT_MASK);
 
+	{
+		int i;
+
+		printk(KERN_INFO "%s: Ring at %lx: next %d nic %d\n",
+				dev->name, (unsigned long)np->ring_addr,
+				np->next_tx, np->nic_tx);
+		printk(KERN_INFO "%s: Dumping tx registers\n", dev->name);
+		for (i=0;i<0x400;i+= 32) {
+			printk(KERN_INFO "%3x: %08x %08x %08x %08x %08x %08x %08x %08x\n",
+					i,
+					readl(base + i + 0), readl(base + i + 4),
+					readl(base + i + 8), readl(base + i + 12),
+					readl(base + i + 16), readl(base + i + 20),
+					readl(base + i + 24), readl(base + i + 28));
+		}
+		printk(KERN_INFO "%s: Dumping tx ring\n", dev->name);
+		for (i=0;i<TX_RING;i+= 4) {
+			printk(KERN_INFO "%03x: %08x %08x // %08x %08x // %08x %08x // %08x %08x\n",
+					i, 
+					le32_to_cpu(np->tx_ring[i].PacketBuffer),
+					le32_to_cpu(np->tx_ring[i].FlagLen),
+					le32_to_cpu(np->tx_ring[i+1].PacketBuffer),
+					le32_to_cpu(np->tx_ring[i+1].FlagLen),
+					le32_to_cpu(np->tx_ring[i+2].PacketBuffer),
+					le32_to_cpu(np->tx_ring[i+2].FlagLen),
+					le32_to_cpu(np->tx_ring[i+3].PacketBuffer),
+					le32_to_cpu(np->tx_ring[i+3].FlagLen));
+		}
+	}
+
 	spin_lock_irq(&np->lock);
 
 	/* 1) stop tx engine */

@@ -1557,7 +1589,7 @@
 		if (!(events & np->irqmask))
 			break;
 
-		if (events & (NVREG_IRQ_TX1|NVREG_IRQ_TX2|NVREG_IRQ_TX_ERR)) {
+		if (events & (NVREG_IRQ_TX1|NVREG_IRQ_TX_OK|NVREG_IRQ_TX_ERROR|NVREG_IRQ_TX_ERR)) {
 			spin_lock(&np->lock);
 			nv_tx_done(dev);
 			spin_unlock(&np->lock);

@@ -2213,17 +2245,10 @@
 
 	if (np->desc_ver == DESC_VER_1) {
 		np->tx_flags = NV_TX_LASTPACKET|NV_TX_VALID;
-		if (id->driver_data & DEV_NEED_LASTPACKET1)
-			np->tx_flags |= NV_TX_LASTPACKET1;
 	} else {
 		np->tx_flags = NV_TX2_LASTPACKET|NV_TX2_VALID;
-		if (id->driver_data & DEV_NEED_LASTPACKET1)
-			np->tx_flags |= NV_TX2_LASTPACKET1;
 	}
-	if (id->driver_data & DEV_IRQMASK_1)
-		np->irqmask = NVREG_IRQMASK_WANTED_1;
-	if (id->driver_data & DEV_IRQMASK_2)
-		np->irqmask = NVREG_IRQMASK_WANTED_2;
+	np->irqmask = NVREG_IRQMASK_WANTED;
 	if (id->driver_data & DEV_NEED_TIMERIRQ)
 		np->irqmask |= NVREG_IRQ_TIMER;
 	if (id->driver_data & DEV_NEED_LINKTIMER) {

@@ -2329,73 +2354,63 @@
 static struct pci_device_id pci_tbl[] = {
 	{	/* nForce Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_1),
-		.driver_data = DEV_IRQMASK_1|DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
 	},
 	{	/* nForce2 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_2),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
 	},
 	{	/* nForce3 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_3),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
 	},
 	{	/* nForce3 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_4),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{	/* nForce3 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_5),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{	/* nForce3 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_6),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{	/* nForce3 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_7),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{	/* CK804 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_8),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{	/* CK804 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_9),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{	/* MCP04 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_10),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{	/* MCP04 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_11),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{	/* MCP51 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_12),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
 	},
 	{	/* MCP51 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_13),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER,
 	},
 	{	/* MCP55 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_14),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{	/* MCP55 Ethernet Controller */
 		PCI_DEVICE(PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NVENET_15),
-		.driver_data = DEV_NEED_LASTPACKET1|DEV_IRQMASK_2|DEV_NEED_TIMERIRQ|
-			DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
+		.driver_data = DEV_NEED_TIMERIRQ|DEV_NEED_LINKTIMER|DEV_HAS_LARGEDESC,
 	},
 	{0,},
 };

@@ -85,6 +85,7 @@
  *	0.33: 16 May 2005: Support for MCP51 added.
  *	0.34: 18 Jun 2005: Add DEV_NEED_LINKTIMER to all nForce nics.
  *	0.35: 26 Jun 2005: Support for MCP55 added.
+ *	0.36: 28 Jul 2005: Add jumbo frame support.
  *
  * Known bugs:
  * We suspect that on some hardware no TX done interrupts are generated.

@@ -96,7 +97,7 @@
  * DEV_NEED_TIMERIRQ will not harm you on sane hardware, only generating a few
  * superfluous timer interrupts from the nic.
  */
-#define FORCEDETH_VERSION		"0.35"
+#define FORCEDETH_VERSION		"0.36"
 #define DRV_NAME			"forcedeth"
 
 #include <linux/module.h>

@@ -379,9 +380,13 @@
 #define TX_LIMIT_START	62
 
 /* rx/tx mac addr + type + vlan + align + slack*/
-#define RX_NIC_BUFSIZE		(ETH_DATA_LEN + 64)
-/* even more slack */
-#define RX_ALLOC_BUFSIZE	(ETH_DATA_LEN + 128)
+#define NV_RX_HEADERS		(64)
+/* even more slack. */
+#define NV_RX_ALLOC_PAD		(64)
+
+/* maximum mtu size */
+#define NV_PKTLIMIT_1	ETH_DATA_LEN	/* hard limit not known */
+#define NV_PKTLIMIT_2	9100	/* Actual limit according to NVidia: 9202 */
 
 #define OOM_REFILL	(1+HZ/20)
 #define POLL_WAIT	(1+HZ/100)

@@ -473,6 +478,7 @@
 	struct sk_buff *rx_skbuff[RX_RING];
 	dma_addr_t rx_dma[RX_RING];
 	unsigned int rx_buf_sz;
+	unsigned int pkt_limit;
 	struct timer_list oom_kick;
 	struct timer_list nic_poll;
 

@@ -792,7 +798,7 @@
 		nr = refill_rx % RX_RING;
 		if (np->rx_skbuff[nr] == NULL) {
 
-			skb = dev_alloc_skb(RX_ALLOC_BUFSIZE);
+			skb = dev_alloc_skb(np->rx_buf_sz + NV_RX_ALLOC_PAD);
 			if (!skb)
 				break;
 

@@ -805,7 +811,7 @@
 						PCI_DMA_FROMDEVICE);
 		np->rx_ring[nr].PacketBuffer = cpu_to_le32(np->rx_dma[nr]);
 		wmb();
-		np->rx_ring[nr].FlagLen = cpu_to_le32(RX_NIC_BUFSIZE | NV_RX_AVAIL);
+		np->rx_ring[nr].FlagLen = cpu_to_le32(np->rx_buf_sz | NV_RX_AVAIL);
 		dprintk(KERN_DEBUG "%s: nv_alloc_rx: Packet %d marked as Available\n",
 					dev->name, refill_rx);
 		refill_rx++;

@@ -831,19 +837,31 @@
 	enable_irq(dev->irq);
 }
 
-static int nv_init_ring(struct net_device *dev)
+static void nv_init_rx(struct net_device *dev) 
 {
 	struct fe_priv *np = get_nvpriv(dev);
 	int i;
 
-	np->next_tx = np->nic_tx = 0;
-	for (i = 0; i < TX_RING; i++)
-		np->tx_ring[i].FlagLen = 0;
-
 	np->cur_rx = RX_RING;
 	np->refill_rx = 0;
 	for (i = 0; i < RX_RING; i++)
 		np->rx_ring[i].FlagLen = 0;
+}
+
+static void nv_init_tx(struct net_device *dev)
+{
+	struct fe_priv *np = get_nvpriv(dev);
+	int i;
+
+	np->next_tx = np->nic_tx = 0;
+	for (i = 0; i < TX_RING; i++)
+		np->tx_ring[i].FlagLen = 0;
+}
+
+static int nv_init_ring(struct net_device *dev)
+{
+	nv_init_tx(dev);
+	nv_init_rx(dev);
 	return nv_alloc_rx(dev);
 }
 

@@ -1207,15 +1225,82 @@
 	}
 }
 
+static void set_bufsize(struct net_device *dev)
+{
+	struct fe_priv *np = netdev_priv(dev);
+
+	if (dev->mtu <= ETH_DATA_LEN)
+		np->rx_buf_sz = ETH_DATA_LEN + NV_RX_HEADERS;
+	else
+		np->rx_buf_sz = dev->mtu + NV_RX_HEADERS;
+}
+
 /*
  * nv_change_mtu: dev->change_mtu function
  * Called with dev_base_lock held for read.
  */
 static int nv_change_mtu(struct net_device *dev, int new_mtu)
 {
-	if (new_mtu > ETH_DATA_LEN)
+	struct fe_priv *np = get_nvpriv(dev);
+	int old_mtu;
+
+	if (new_mtu < 64 || new_mtu > np->pkt_limit)
 		return -EINVAL;
+
+	old_mtu = dev->mtu;
 	dev->mtu = new_mtu;
+
+	/* return early if the buffer sizes will not change */
+	if (old_mtu <= ETH_DATA_LEN && new_mtu <= ETH_DATA_LEN)
+		return 0;
+	if (old_mtu == new_mtu)
+		return 0;
+
+	/* synchronized against open : rtnl_lock() held by caller */
+	if (netif_running(dev)) {
+		u8 *base = get_hwbase(dev);
+		/*
+		 * It seems that the nic preloads valid ring entries into an
+		 * internal buffer. The procedure for flushing everything is
+		 * guessed, there is probably a simpler approach.
+		 * Changing the MTU is a rare event, it shouldn't matter.
+		 */
+		disable_irq(dev->irq);
+		spin_lock_bh(&dev->xmit_lock);
+		spin_lock(&np->lock);
+		/* stop engines */
+		nv_stop_rx(dev);
+		nv_stop_tx(dev);
+		nv_txrx_reset(dev);
+		/* drain rx queue */
+		nv_drain_rx(dev);
+		nv_drain_tx(dev);
+		/* reinit driver view of the rx queue */
+		nv_init_rx(dev);
+		nv_init_tx(dev);
+		/* alloc new rx buffers */
+		set_bufsize(dev);
+		if (nv_alloc_rx(dev)) {
+			if (!np->in_shutdown)
+				mod_timer(&np->oom_kick, jiffies + OOM_REFILL);
+		}
+		/* reinit nic view of the rx queue */
+		writel(np->rx_buf_sz, base + NvRegOffloadConfig);
+		writel((u32) np->ring_addr, base + NvRegRxRingPhysAddr);
+		writel((u32) (np->ring_addr + RX_RING*sizeof(struct ring_desc)), base + NvRegTxRingPhysAddr);
+		writel( ((RX_RING-1) << NVREG_RINGSZ_RXSHIFT) + ((TX_RING-1) << NVREG_RINGSZ_TXSHIFT),
+			base + NvRegRingSizes);
+		pci_push(base);
+		writel(NVREG_TXRXCTL_KICK|np->desc_ver, get_hwbase(dev) + NvRegTxRxControl);
+		pci_push(base);
+
+		/* restart rx engine */
+		nv_start_rx(dev);
+		nv_start_tx(dev);
+		spin_unlock(&np->lock);
+		spin_unlock_bh(&dev->xmit_lock);
+		enable_irq(dev->irq);
+	}
 	return 0;
 }
 

@@ -1792,6 +1877,7 @@
 	writel(0, base + NvRegAdapterControl);
 
 	/* 2) initialize descriptor rings */
+	set_bufsize(dev);
 	oom = nv_init_ring(dev);
 
 	writel(0, base + NvRegLinkSpeed);

@@ -1837,7 +1923,7 @@
 	writel(NVREG_MISC1_FORCE | NVREG_MISC1_HD, base + NvRegMisc1);
 	writel(readl(base + NvRegTransmitterStatus), base + NvRegTransmitterStatus);
 	writel(NVREG_PFF_ALWAYS, base + NvRegPacketFilterFlags);
-	writel(NVREG_OFFLOAD_NORMAL, base + NvRegOffloadConfig);
+	writel(np->rx_buf_sz, base + NvRegOffloadConfig);
 
 	writel(readl(base + NvRegReceiverStatus), base + NvRegReceiverStatus);
 	get_random_bytes(&i, sizeof(i));

@@ -2007,13 +2093,16 @@
 
 	/* handle different descriptor versions */
 	if (pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_1 ||
-		pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_2 ||
-		pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_3 ||    
-		pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_12 ||
-		pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_13)
+			pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_2 ||
+			pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_3 ||    
+			pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_12 ||
+			pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_13) {
 		np->desc_ver = DESC_VER_1;
-	else
+ 		np->pkt_limit = NV_PKTLIMIT_1;
+	} else {
 		np->desc_ver = DESC_VER_2;
+ 		np->pkt_limit = NV_PKTLIMIT_1;
+	}
 
 	err = -ENOMEM;
 	np->base = ioremap(addr, NV_PCI_REGSZ);

@@ -85,6 +85,7 @@
  *	0.33: 16 May 2005: Support for MCP51 added.
  *	0.34: 18 Jun 2005: Add DEV_NEED_LINKTIMER to all nForce nics.
  *	0.35: 26 Jun 2005: Support for MCP55 added.
+ *	0.36: 28 Jul 2005: Add jumbo frame support.
  *
  * Known bugs:
  * We suspect that on some hardware no TX done interrupts are generated.

@@ -96,7 +97,7 @@
  * DEV_NEED_TIMERIRQ will not harm you on sane hardware, only generating a few
  * superfluous timer interrupts from the nic.
  */
-#define FORCEDETH_VERSION		"0.35"
+#define FORCEDETH_VERSION		"0.36"
 #define DRV_NAME			"forcedeth"
 
 #include <linux/module.h>

@@ -379,9 +380,13 @@
 #define TX_LIMIT_START	62
 
 /* rx/tx mac addr + type + vlan + align + slack*/
-#define RX_NIC_BUFSIZE		(ETH_DATA_LEN + 64)
-/* even more slack */
-#define RX_ALLOC_BUFSIZE	(ETH_DATA_LEN + 128)
+#define NV_RX_HEADERS		(64)
+/* even more slack. */
+#define NV_RX_ALLOC_PAD		(64)
+
+/* maximum mtu size */
+#define NV_PKTLIMIT_1	ETH_DATA_LEN	/* hard limit not known */
+#define NV_PKTLIMIT_2	9100	/* Actual limit according to NVidia: 9202 */
 
 #define OOM_REFILL	(1+HZ/20)
 #define POLL_WAIT	(1+HZ/100)

@@ -473,6 +478,7 @@
 	struct sk_buff *rx_skbuff[RX_RING];
 	dma_addr_t rx_dma[RX_RING];
 	unsigned int rx_buf_sz;
+	unsigned int pkt_limit;
 	struct timer_list oom_kick;
 	struct timer_list nic_poll;
 

@@ -792,7 +798,7 @@
 		nr = refill_rx % RX_RING;
 		if (np->rx_skbuff[nr] == NULL) {
 
-			skb = dev_alloc_skb(RX_ALLOC_BUFSIZE);
+			skb = dev_alloc_skb(np->rx_buf_sz + NV_RX_ALLOC_PAD);
 			if (!skb)
 				break;
 

@@ -805,7 +811,7 @@
 						PCI_DMA_FROMDEVICE);
 		np->rx_ring[nr].PacketBuffer = cpu_to_le32(np->rx_dma[nr]);
 		wmb();
-		np->rx_ring[nr].FlagLen = cpu_to_le32(RX_NIC_BUFSIZE | NV_RX_AVAIL);
+		np->rx_ring[nr].FlagLen = cpu_to_le32(np->rx_buf_sz | NV_RX_AVAIL);
 		dprintk(KERN_DEBUG "%s: nv_alloc_rx: Packet %d marked as Available\n",
 					dev->name, refill_rx);
 		refill_rx++;

@@ -831,19 +837,31 @@
 	enable_irq(dev->irq);
 }
 
-static int nv_init_ring(struct net_device *dev)
+static void nv_init_rx(struct net_device *dev) 
 {
 	struct fe_priv *np = get_nvpriv(dev);
 	int i;
 
-	np->next_tx = np->nic_tx = 0;
-	for (i = 0; i < TX_RING; i++)
-		np->tx_ring[i].FlagLen = 0;
-
 	np->cur_rx = RX_RING;
 	np->refill_rx = 0;
 	for (i = 0; i < RX_RING; i++)
 		np->rx_ring[i].FlagLen = 0;
+}
+
+static void nv_init_tx(struct net_device *dev)
+{
+	struct fe_priv *np = get_nvpriv(dev);
+	int i;
+
+	np->next_tx = np->nic_tx = 0;
+	for (i = 0; i < TX_RING; i++)
+		np->tx_ring[i].FlagLen = 0;
+}
+
+static int nv_init_ring(struct net_device *dev)
+{
+	nv_init_tx(dev);
+	nv_init_rx(dev);
 	return nv_alloc_rx(dev);
 }
 

@@ -1207,15 +1225,82 @@
 	}
 }
 
+static void set_bufsize(struct net_device *dev)
+{
+	struct fe_priv *np = netdev_priv(dev);
+
+	if (dev->mtu <= ETH_DATA_LEN)
+		np->rx_buf_sz = ETH_DATA_LEN + NV_RX_HEADERS;
+	else
+		np->rx_buf_sz = dev->mtu + NV_RX_HEADERS;
+}
+
 /*
  * nv_change_mtu: dev->change_mtu function
  * Called with dev_base_lock held for read.
  */
 static int nv_change_mtu(struct net_device *dev, int new_mtu)
 {
-	if (new_mtu > ETH_DATA_LEN)
+	struct fe_priv *np = get_nvpriv(dev);
+	int old_mtu;
+
+	if (new_mtu < 64 || new_mtu > np->pkt_limit)
 		return -EINVAL;
+
+	old_mtu = dev->mtu;
 	dev->mtu = new_mtu;
+
+	/* return early if the buffer sizes will not change */
+	if (old_mtu <= ETH_DATA_LEN && new_mtu <= ETH_DATA_LEN)
+		return 0;
+	if (old_mtu == new_mtu)
+		return 0;
+
+	/* synchronized against open : rtnl_lock() held by caller */
+	if (netif_running(dev)) {
+		u8 *base = get_hwbase(dev);
+		/*
+		 * It seems that the nic preloads valid ring entries into an
+		 * internal buffer. The procedure for flushing everything is
+		 * guessed, there is probably a simpler approach.
+		 * Changing the MTU is a rare event, it shouldn't matter.
+		 */
+		disable_irq(dev->irq);
+		spin_lock_bh(&dev->xmit_lock);
+		spin_lock(&np->lock);
+		/* stop engines */
+		nv_stop_rx(dev);
+		nv_stop_tx(dev);
+		nv_txrx_reset(dev);
+		/* drain rx queue */
+		nv_drain_rx(dev);
+		nv_drain_tx(dev);
+		/* reinit driver view of the rx queue */
+		nv_init_rx(dev);
+		nv_init_tx(dev);
+		/* alloc new rx buffers */
+		set_bufsize(dev);
+		if (nv_alloc_rx(dev)) {
+			if (!np->in_shutdown)
+				mod_timer(&np->oom_kick, jiffies + OOM_REFILL);
+		}
+		/* reinit nic view of the rx queue */
+		writel(np->rx_buf_sz, base + NvRegOffloadConfig);
+		writel((u32) np->ring_addr, base + NvRegRxRingPhysAddr);
+		writel((u32) (np->ring_addr + RX_RING*sizeof(struct ring_desc)), base + NvRegTxRingPhysAddr);
+		writel( ((RX_RING-1) << NVREG_RINGSZ_RXSHIFT) + ((TX_RING-1) << NVREG_RINGSZ_TXSHIFT),
+			base + NvRegRingSizes);
+		pci_push(base);
+		writel(NVREG_TXRXCTL_KICK|np->desc_ver, get_hwbase(dev) + NvRegTxRxControl);
+		pci_push(base);
+
+		/* restart rx engine */
+		nv_start_rx(dev);
+		nv_start_tx(dev);
+		spin_unlock(&np->lock);
+		spin_unlock_bh(&dev->xmit_lock);
+		enable_irq(dev->irq);
+	}
 	return 0;
 }
 

@@ -1792,6 +1877,7 @@
 	writel(0, base + NvRegAdapterControl);
 
 	/* 2) initialize descriptor rings */
+	set_bufsize(dev);
 	oom = nv_init_ring(dev);
 
 	writel(0, base + NvRegLinkSpeed);

@@ -1837,7 +1923,7 @@
 	writel(NVREG_MISC1_FORCE | NVREG_MISC1_HD, base + NvRegMisc1);
 	writel(readl(base + NvRegTransmitterStatus), base + NvRegTransmitterStatus);
 	writel(NVREG_PFF_ALWAYS, base + NvRegPacketFilterFlags);
-	writel(NVREG_OFFLOAD_NORMAL, base + NvRegOffloadConfig);
+	writel(np->rx_buf_sz, base + NvRegOffloadConfig);
 
 	writel(readl(base + NvRegReceiverStatus), base + NvRegReceiverStatus);
 	get_random_bytes(&i, sizeof(i));

@@ -2007,13 +2093,16 @@
 
 	/* handle different descriptor versions */
 	if (pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_1 ||
-		pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_2 ||
-		pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_3 ||    
-		pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_12 ||
-		pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_13)
+			pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_2 ||
+			pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_3 ||    
+			pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_12 ||
+			pci_dev->device == PCI_DEVICE_ID_NVIDIA_NVENET_13) {
 		np->desc_ver = DESC_VER_1;
-	else
+ 		np->pkt_limit = NV_PKTLIMIT_1;
+	} else {
 		np->desc_ver = DESC_VER_2;
+ 		np->pkt_limit = NV_PKTLIMIT_1;
+	}
 
 	err = -ENOMEM;
 	np->base = ioremap(addr, NV_PCI_REGSZ);