ns83820.c

Linux Kernel 2.6.9 for OMAP1710

		);
#endif

	sg = dev->rx_info.descs + (next_empty * DESC_SIZE);
	if (unlikely(NULL != dev->rx_info.skbs[next_empty]))
		BUG();
	dev->rx_info.skbs[next_empty] = skb;

	dev->rx_info.next_empty = (next_empty + 1) % NR_RX_DESC;
	cmdsts = REAL_RX_BUF_SIZE | CMDSTS_INTR;
	buf = pci_map_single(dev->pci_dev, skb->tail,
			     REAL_RX_BUF_SIZE, PCI_DMA_FROMDEVICE);
	build_rx_desc(dev, sg, 0, buf, cmdsts, 0);
	/* update link of previous rx */
	if (likely(next_empty != dev->rx_info.next_rx))
		dev->rx_info.descs[((NR_RX_DESC + next_empty - 1) % NR_RX_DESC) * DESC_SIZE] = cpu_to_le32(dev->rx_info.phy_descs + (next_empty * DESC_SIZE * 4));

	return 0;
}

static inline int rx_refill(struct net_device *ndev, int gfp)
{
	struct ns83820 *dev = PRIV(ndev);
	unsigned i;
	unsigned long flags = 0;

	if (unlikely(nr_rx_empty(dev) <= 2))
		return 0;

	dprintk("rx_refill(%p)\n", ndev);
	if (gfp == GFP_ATOMIC)
		spin_lock_irqsave(&dev->rx_info.lock, flags);
	for (i=0; i<NR_RX_DESC; i++) {
		struct sk_buff *skb;
		long res;
		/* extra 16 bytes for alignment */
		skb = __dev_alloc_skb(REAL_RX_BUF_SIZE+16, gfp);
		if (unlikely(!skb))
			break;

		res = (long)skb->tail & 0xf;
		res = 0x10 - res;
		res &= 0xf;
		skb_reserve(skb, res);

		skb->dev = ndev;
		if (gfp != GFP_ATOMIC)
			spin_lock_irqsave(&dev->rx_info.lock, flags);
		res = ns83820_add_rx_skb(dev, skb);
		if (gfp != GFP_ATOMIC)
			spin_unlock_irqrestore(&dev->rx_info.lock, flags);
		if (res) {
			i = 1;
			break;
		}
	}
	if (gfp == GFP_ATOMIC)
		spin_unlock_irqrestore(&dev->rx_info.lock, flags);

	return i ? 0 : -ENOMEM;
}

static void FASTCALL(rx_refill_atomic(struct net_device *ndev));
static void fastcall rx_refill_atomic(struct net_device *ndev)
{
	rx_refill(ndev, GFP_ATOMIC);
}

/* REFILL */
static inline void queue_refill(void *_dev)
{
	struct net_device *ndev = _dev;
	struct ns83820 *dev = PRIV(ndev);

	rx_refill(ndev, GFP_KERNEL);
	if (dev->rx_info.up)
		kick_rx(ndev);
}

static inline void clear_rx_desc(struct ns83820 *dev, unsigned i)
{
	build_rx_desc(dev, dev->rx_info.descs + (DESC_SIZE * i), 0, 0, CMDSTS_OWN, 0);
}

static void FASTCALL(phy_intr(struct net_device *ndev));
static void fastcall phy_intr(struct net_device *ndev)
{
	struct ns83820 *dev = PRIV(ndev);
	static char *speeds[] = { "10", "100", "1000", "1000(?)", "1000F" };
	u32 cfg, new_cfg;
	u32 tbisr, tanar, tanlpar;
	int speed, fullduplex, newlinkstate;

	cfg = readl(dev->base + CFG) ^ SPDSTS_POLARITY;

	if (dev->CFG_cache & CFG_TBI_EN) {
		/* we have an optical transceiver */
		tbisr = readl(dev->base + TBISR);
		tanar = readl(dev->base + TANAR);
		tanlpar = readl(dev->base + TANLPAR);
		dprintk("phy_intr: tbisr=%08x, tanar=%08x, tanlpar=%08x\n",
			tbisr, tanar, tanlpar);

		if ( (fullduplex = (tanlpar & TANAR_FULL_DUP)
		      && (tanar & TANAR_FULL_DUP)) ) {

			/* both of us are full duplex */
			writel(readl(dev->base + TXCFG)
			       | TXCFG_CSI | TXCFG_HBI | TXCFG_ATP,
			       dev->base + TXCFG);
			writel(readl(dev->base + RXCFG) | RXCFG_RX_FD,
			       dev->base + RXCFG);
			/* Light up full duplex LED */
			writel(readl(dev->base + GPIOR) | GPIOR_GP1_OUT,
			       dev->base + GPIOR);

		} else if(((tanlpar & TANAR_HALF_DUP)
			   && (tanar & TANAR_HALF_DUP))
			|| ((tanlpar & TANAR_FULL_DUP)
			    && (tanar & TANAR_HALF_DUP))
			|| ((tanlpar & TANAR_HALF_DUP)
			    && (tanar & TANAR_FULL_DUP))) {

			/* one or both of us are half duplex */
			writel((readl(dev->base + TXCFG)
				& ~(TXCFG_CSI | TXCFG_HBI)) | TXCFG_ATP,
			       dev->base + TXCFG);
			writel(readl(dev->base + RXCFG) & ~RXCFG_RX_FD,
			       dev->base + RXCFG);
			/* Turn off full duplex LED */
			writel(readl(dev->base + GPIOR) & ~GPIOR_GP1_OUT,
			       dev->base + GPIOR);
		}

		speed = 4; /* 1000F */

	} else {
		/* we have a copper transceiver */
		new_cfg = dev->CFG_cache & ~(CFG_SB | CFG_MODE_1000 | CFG_SPDSTS);

		if (cfg & CFG_SPDSTS1)
			new_cfg |= CFG_MODE_1000;
		else
			new_cfg &= ~CFG_MODE_1000;

		speed = ((cfg / CFG_SPDSTS0) & 3);
		fullduplex = (cfg & CFG_DUPSTS);

		if (fullduplex)
			new_cfg |= CFG_SB;

		if ((cfg & CFG_LNKSTS) &&
		    ((new_cfg ^ dev->CFG_cache) & CFG_MODE_1000)) {
			writel(new_cfg, dev->base + CFG);
			dev->CFG_cache = new_cfg;
		}

		dev->CFG_cache &= ~CFG_SPDSTS;
		dev->CFG_cache |= cfg & CFG_SPDSTS;
	}

	newlinkstate = (cfg & CFG_LNKSTS) ? LINK_UP : LINK_DOWN;

	if (newlinkstate & LINK_UP
	    && dev->linkstate != newlinkstate) {
		netif_start_queue(ndev);
		netif_wake_queue(ndev);
		printk(KERN_INFO "%s: link now %s mbps, %s duplex and up.\n",
			ndev->name,
			speeds[speed],
			fullduplex ? "full" : "half");
	} else if (newlinkstate & LINK_DOWN
		   && dev->linkstate != newlinkstate) {
		netif_stop_queue(ndev);
		printk(KERN_INFO "%s: link now down.\n", ndev->name);
	}

	dev->linkstate = newlinkstate;
}

static int ns83820_setup_rx(struct net_device *ndev)
{
	struct ns83820 *dev = PRIV(ndev);
	unsigned i;
	int ret;

	dprintk("ns83820_setup_rx(%p)\n", ndev);

	dev->rx_info.idle = 1;
	dev->rx_info.next_rx = 0;
	dev->rx_info.next_rx_desc = dev->rx_info.descs;
	dev->rx_info.next_empty = 0;

	for (i=0; i<NR_RX_DESC; i++)
		clear_rx_desc(dev, i);

	writel(0, dev->base + RXDP_HI);
	writel(dev->rx_info.phy_descs, dev->base + RXDP);

	ret = rx_refill(ndev, GFP_KERNEL);
	if (!ret) {
		dprintk("starting receiver\n");
		/* prevent the interrupt handler from stomping on us */
		spin_lock_irq(&dev->rx_info.lock);

		writel(0x0001, dev->base + CCSR);
		writel(0, dev->base + RFCR);
		writel(0x7fc00000, dev->base + RFCR);
		writel(0xffc00000, dev->base + RFCR);

		dev->rx_info.up = 1;

		phy_intr(ndev);

		/* Okay, let it rip */
		spin_lock_irq(&dev->misc_lock);
		dev->IMR_cache |= ISR_PHY;
		dev->IMR_cache |= ISR_RXRCMP;
		//dev->IMR_cache |= ISR_RXERR;
		//dev->IMR_cache |= ISR_RXOK;
		dev->IMR_cache |= ISR_RXORN;
		dev->IMR_cache |= ISR_RXSOVR;
		dev->IMR_cache |= ISR_RXDESC;
		dev->IMR_cache |= ISR_RXIDLE;
		dev->IMR_cache |= ISR_TXDESC;
		dev->IMR_cache |= ISR_TXIDLE;

		writel(dev->IMR_cache, dev->base + IMR);
		writel(1, dev->base + IER);
		spin_unlock_irq(&dev->misc_lock);

		kick_rx(ndev);

		spin_unlock_irq(&dev->rx_info.lock);
	}
	return ret;
}

static void ns83820_cleanup_rx(struct ns83820 *dev)
{
	unsigned i;
	unsigned long flags;

	dprintk("ns83820_cleanup_rx(%p)\n", dev);

	/* disable receive interrupts */
	spin_lock_irqsave(&dev->misc_lock, flags);
	dev->IMR_cache &= ~(ISR_RXOK | ISR_RXDESC | ISR_RXERR | ISR_RXEARLY | ISR_RXIDLE);
	writel(dev->IMR_cache, dev->base + IMR);
	spin_unlock_irqrestore(&dev->misc_lock, flags);

	/* synchronize with the interrupt handler and kill it */
	dev->rx_info.up = 0;
	synchronize_irq(dev->pci_dev->irq);

	/* touch the pci bus... */
	readl(dev->base + IMR);

	/* assumes the transmitter is already disabled and reset */
	writel(0, dev->base + RXDP_HI);
	writel(0, dev->base + RXDP);

	for (i=0; i<NR_RX_DESC; i++) {
		struct sk_buff *skb = dev->rx_info.skbs[i];
		dev->rx_info.skbs[i] = NULL;
		clear_rx_desc(dev, i);
		if (skb)
			kfree_skb(skb);
	}
}

static void FASTCALL(ns83820_rx_kick(struct net_device *ndev));
static void fastcall ns83820_rx_kick(struct net_device *ndev)
{
	struct ns83820 *dev = PRIV(ndev);
	/*if (nr_rx_empty(dev) >= NR_RX_DESC/4)*/ {
		if (dev->rx_info.up) {
			rx_refill_atomic(ndev);
			kick_rx(ndev);
		}
	}

	if (dev->rx_info.up && nr_rx_empty(dev) > NR_RX_DESC*3/4)
		schedule_work(&dev->tq_refill);
	else
		kick_rx(ndev);
	if (dev->rx_info.idle)
		printk(KERN_DEBUG "%s: BAD\n", ndev->name);
}

/* rx_irq
 *	
 */
static void FASTCALL(rx_irq(struct net_device *ndev));
static void fastcall rx_irq(struct net_device *ndev)
{
	struct ns83820 *dev = PRIV(ndev);
	struct rx_info *info = &dev->rx_info;
	unsigned next_rx;
	u32 cmdsts, *desc;
	unsigned long flags;
	int nr = 0;

	dprintk("rx_irq(%p)\n", ndev);
	dprintk("rxdp: %08x, descs: %08lx next_rx[%d]: %p next_empty[%d]: %p\n",
		readl(dev->base + RXDP),
		(long)(dev->rx_info.phy_descs),
		(int)dev->rx_info.next_rx,
		(dev->rx_info.descs + (DESC_SIZE * dev->rx_info.next_rx)),
		(int)dev->rx_info.next_empty,
		(dev->rx_info.descs + (DESC_SIZE * dev->rx_info.next_empty))
		);

	spin_lock_irqsave(&info->lock, flags);
	if (!info->up)
		goto out;

	dprintk("walking descs\n");
	next_rx = info->next_rx;
	desc = info->next_rx_desc;
	while ((CMDSTS_OWN & (cmdsts = le32_to_cpu(desc[DESC_CMDSTS]))) &&
	       (cmdsts != CMDSTS_OWN)) {
		struct sk_buff *skb;
		u32 extsts = le32_to_cpu(desc[DESC_EXTSTS]);
		dma_addr_t bufptr = desc_addr_get(desc + DESC_BUFPTR);

		dprintk("cmdsts: %08x\n", cmdsts);
		dprintk("link: %08x\n", cpu_to_le32(desc[DESC_LINK]));
		dprintk("extsts: %08x\n", extsts);

		skb = info->skbs[next_rx];
		info->skbs[next_rx] = NULL;
		info->next_rx = (next_rx + 1) % NR_RX_DESC;

		mb();
		clear_rx_desc(dev, next_rx);

		pci_unmap_single(dev->pci_dev, bufptr,
				 RX_BUF_SIZE, PCI_DMA_FROMDEVICE);
		if (likely(CMDSTS_OK & cmdsts)) {
			int len = cmdsts & 0xffff;
			skb_put(skb, len);
			if (unlikely(!skb))
				goto netdev_mangle_me_harder_failed;
			if (cmdsts & CMDSTS_DEST_MULTI)
				dev->stats.multicast ++;
			dev->stats.rx_packets ++;
			dev->stats.rx_bytes += len;
			if ((extsts & 0x002a0000) && !(extsts & 0x00540000)) {
				skb->ip_summed = CHECKSUM_UNNECESSARY;
			} else {
				skb->ip_summed = CHECKSUM_NONE;
			}
			skb->protocol = eth_type_trans(skb, ndev);
			if (NET_RX_DROP == netif_rx(skb)) {
netdev_mangle_me_harder_failed:
				dev->stats.rx_dropped ++;
			}
		} else {
			kfree_skb(skb);
		}

		nr++;
		next_rx = info->next_rx;
		desc = info->descs + (DESC_SIZE * next_rx);
	}
	info->next_rx = next_rx;
	info->next_rx_desc = info->descs + (DESC_SIZE * next_rx);

out:
	if (0 && !nr) {
		Dprintk("dazed: cmdsts_f: %08x\n", cmdsts);
	}

	spin_unlock_irqrestore(&info->lock, flags);
}

static void rx_action(unsigned long _dev)
{
	struct net_device *ndev = (void *)_dev;
	struct ns83820 *dev = PRIV(ndev);
	rx_irq(ndev);
	writel(ihr, dev->base + IHR);

	spin_lock_irq(&dev->misc_lock);
	dev->IMR_cache |= ISR_RXDESC;
	writel(dev->IMR_cache, dev->base + IMR);
	spin_unlock_irq(&dev->misc_lock);

	rx_irq(ndev);
	ns83820_rx_kick(ndev);
}

/* Packet Transmit code
 */
static inline void kick_tx(struct ns83820 *dev)
{
	dprintk("kick_tx(%p): tx_idx=%d free_idx=%d\n",
		dev, dev->tx_idx, dev->tx_free_idx);
	writel(CR_TXE, dev->base + CR);
}

/* No spinlock needed on the transmit irq path as the interrupt handler is
 * serialized.
 */
static void do_tx_done(struct net_device *ndev)
{
	struct ns83820 *dev = PRIV(ndev);
	u32 cmdsts, tx_done_idx, *desc;

	spin_lock_irq(&dev->tx_lock);

	dprintk("do_tx_done(%p)\n", ndev);
	tx_done_idx = dev->tx_done_idx;
	desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);

	dprintk("tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
		tx_done_idx, dev->tx_free_idx, le32_to_cpu(desc[DESC_CMDSTS]));
	while ((tx_done_idx != dev->tx_free_idx) &&
	       !(CMDSTS_OWN & (cmdsts = le32_to_cpu(desc[DESC_CMDSTS]))) ) {
		struct sk_buff *skb;
		unsigned len;
		dma_addr_t addr;

		if (cmdsts & CMDSTS_ERR)
			dev->stats.tx_errors ++;
		if (cmdsts & CMDSTS_OK)
			dev->stats.tx_packets ++;
		if (cmdsts & CMDSTS_OK)
			dev->stats.tx_bytes += cmdsts & 0xffff;

		dprintk("tx_done_idx=%d free_idx=%d cmdsts=%08x\n",
			tx_done_idx, dev->tx_free_idx, cmdsts);
		skb = dev->tx_skbs[tx_done_idx];
		dev->tx_skbs[tx_done_idx] = NULL;
		dprintk("done(%p)\n", skb);

		len = cmdsts & CMDSTS_LEN_MASK;
		addr = desc_addr_get(desc + DESC_BUFPTR);
		if (skb) {
			pci_unmap_single(dev->pci_dev,
					addr,
					len,
					PCI_DMA_TODEVICE);
			dev_kfree_skb_irq(skb);
			atomic_dec(&dev->nr_tx_skbs);
		} else
			pci_unmap_page(dev->pci_dev, 
					addr,
					len,
					PCI_DMA_TODEVICE);

		tx_done_idx = (tx_done_idx + 1) % NR_TX_DESC;
		dev->tx_done_idx = tx_done_idx;
		desc[DESC_CMDSTS] = cpu_to_le32(0);
		mb();
		desc = dev->tx_descs + (tx_done_idx * DESC_SIZE);
	}

	/* Allow network stack to resume queueing packets after we've
	 * finished transmitting at least 1/4 of the packets in the queue.
	 */
	if (netif_queue_stopped(ndev) && start_tx_okay(dev)) {
		dprintk("start_queue(%p)\n", ndev);
		netif_start_queue(ndev);
		netif_wake_queue(ndev);
	}
	spin_unlock_irq(&dev->tx_lock);
}

static void ns83820_cleanup_tx(struct ns83820 *dev)
{
	unsigned i;

	for (i=0; i<NR_TX_DESC; i++) {
		struct sk_buff *skb = dev->tx_skbs[i];
		dev->tx_skbs[i] = NULL;
		if (skb) {
			u32 *desc = dev->tx_descs + (i * DESC_SIZE);
			pci_unmap_single(dev->pci_dev,
					desc_addr_get(desc + DESC_BUFPTR),
					le32_to_cpu(desc[DESC_CMDSTS]) & CMDSTS_LEN_MASK,
					PCI_DMA_TODEVICE);
			dev_kfree_skb_irq(skb);
			atomic_dec(&dev->nr_tx_skbs);
		}
	}

	memset(dev->tx_descs, 0, NR_TX_DESC * DESC_SIZE * 4);
}

/* transmit routine.  This code relies on the network layer serializing
 * its calls in, but will run happily in parallel with the interrupt
 * handler.  This code currently has provisions for fragmenting tx buffers
 * while trying to track down a bug in either the zero copy code or
 * the tx fifo (hence the MAX_FRAG_LEN).
 */
static int ns83820_hard_start_xmit(struct sk_buff *skb, struct net_device *ndev)
{
	struct ns83820 *dev = PRIV(ndev);
	u32 free_idx, cmdsts, extsts;
	int nr_free, nr_frags;
	unsigned tx_done_idx, last_idx;
	dma_addr_t buf;
	unsigned len;
	skb_frag_t *frag;
	int stopped = 0;
	int do_intr = 0;
	volatile u32 *first_desc;

	dprintk("ns83820_hard_start_xmit\n");

	nr_frags =  skb_shinfo(skb)->nr_frags;
again:
	if (unlikely(dev->CFG_cache & CFG_LNKSTS)) {
		netif_stop_queue(ndev);
		if (unlikely(dev->CFG_cache & CFG_LNKSTS))
			return 1;
		netif_start_queue(ndev);
	}

	last_idx = free_idx = dev->tx_free_idx;
	tx_done_idx = dev->tx_done_idx;
	nr_free = (tx_done_idx + NR_TX_DESC-2 - free_idx) % NR_TX_DESC;
	nr_free -= 1;
	if (nr_free <= nr_frags) {
		dprintk("stop_queue - not enough(%p)\n", ndev);
		netif_stop_queue(ndev);

		/* Check again: we may have raced with a tx done irq */
		if (dev->tx_done_idx != tx_done_idx) {
			dprintk("restart queue(%p)\n", ndev);
			netif_start_queue(ndev);
			goto again;
		}
		return 1;
	}

	if (free_idx == dev->tx_intr_idx) {