bcm63xx_enet.c

/*
 * Driver for BCM963xx builtin Ethernet mac
 *
 * Copyright (C) 2008 Maxime Bizon <mbizon@freebox.fr>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
#include <linux/init.h>
#include <linux/module.h>
#include <linux/clk.h>
#include <linux/etherdevice.h>
#include <linux/delay.h>
#include <linux/ethtool.h>
#include <linux/crc32.h>
#include <linux/err.h>
#include <linux/dma-mapping.h>
#include <linux/platform_device.h>
#include <linux/if_vlan.h>

#include <bcm63xx_dev_enet.h>
#include "bcm63xx_enet.h"

static char bcm_enet_driver_name[] = "bcm63xx_enet";
static char bcm_enet_driver_version[] = "1.0";

static int copybreak __read_mostly = 128;
module_param(copybreak, int, 0);
MODULE_PARM_DESC(copybreak, "Receive copy threshold");

/* io memory shared between all devices */
static void __iomem *bcm_enet_shared_base;

/*
 * io helpers to access mac registers
 */
static inline u32 enet_readl(struct bcm_enet_priv *priv, u32 off)
{
	return bcm_readl(priv->base + off);
}

static inline void enet_writel(struct bcm_enet_priv *priv,
			       u32 val, u32 off)
{
	bcm_writel(val, priv->base + off);
}

/*
 * io helpers to access shared registers
 */
static inline u32 enet_dma_readl(struct bcm_enet_priv *priv, u32 off)
{
	return bcm_readl(bcm_enet_shared_base + off);
}

static inline void enet_dma_writel(struct bcm_enet_priv *priv,
				       u32 val, u32 off)
{
	bcm_writel(val, bcm_enet_shared_base + off);
}

/*
 * write given data into mii register and wait for transfer to end
 * with timeout (average measured transfer time is 25us)
 */
static int do_mdio_op(struct bcm_enet_priv *priv, unsigned int data)
{
	int limit;

	/* make sure mii interrupt status is cleared */
	enet_writel(priv, ENET_IR_MII, ENET_IR_REG);

	enet_writel(priv, data, ENET_MIIDATA_REG);
	wmb();

	/* busy wait on mii interrupt bit, with timeout */
	limit = 1000;
	do {
		if (enet_readl(priv, ENET_IR_REG) & ENET_IR_MII)
			break;
		udelay(1);
	} while (limit-- >= 0);

	return (limit < 0) ? 1 : 0;
}

/*
 * MII internal read callback
 */
static int bcm_enet_mdio_read(struct bcm_enet_priv *priv, int mii_id,
			      int regnum)
{
	u32 tmp, val;

	tmp = regnum << ENET_MIIDATA_REG_SHIFT;
	tmp |= 0x2 << ENET_MIIDATA_TA_SHIFT;
	tmp |= mii_id << ENET_MIIDATA_PHYID_SHIFT;
	tmp |= ENET_MIIDATA_OP_READ_MASK;

	if (do_mdio_op(priv, tmp))
		return -1;

	val = enet_readl(priv, ENET_MIIDATA_REG);
	val &= 0xffff;
	return val;
}

/*
 * MII internal write callback
 */
static int bcm_enet_mdio_write(struct bcm_enet_priv *priv, int mii_id,
			       int regnum, u16 value)
{
	u32 tmp;

	tmp = (value & 0xffff) << ENET_MIIDATA_DATA_SHIFT;
	tmp |= 0x2 << ENET_MIIDATA_TA_SHIFT;
	tmp |= regnum << ENET_MIIDATA_REG_SHIFT;
	tmp |= mii_id << ENET_MIIDATA_PHYID_SHIFT;
	tmp |= ENET_MIIDATA_OP_WRITE_MASK;

	(void)do_mdio_op(priv, tmp);
	return 0;
}

/*
 * MII read callback from phylib
 */
static int bcm_enet_mdio_read_phylib(struct mii_bus *bus, int mii_id,
				     int regnum)
{
	return bcm_enet_mdio_read(bus->priv, mii_id, regnum);
}

/*
 * MII write callback from phylib
 */
static int bcm_enet_mdio_write_phylib(struct mii_bus *bus, int mii_id,
				      int regnum, u16 value)
{
	return bcm_enet_mdio_write(bus->priv, mii_id, regnum, value);
}

/*
 * MII read callback from mii core
 */
static int bcm_enet_mdio_read_mii(struct net_device *dev, int mii_id,
				  int regnum)
{
	return bcm_enet_mdio_read(netdev_priv(dev), mii_id, regnum);
}

/*
 * MII write callback from mii core
 */
static void bcm_enet_mdio_write_mii(struct net_device *dev, int mii_id,
				    int regnum, int value)
{
	bcm_enet_mdio_write(netdev_priv(dev), mii_id, regnum, value);
}

/*
 * refill rx queue
 */
static int bcm_enet_refill_rx(struct net_device *dev)
{
	struct bcm_enet_priv *priv;

	priv = netdev_priv(dev);

	while (priv->rx_desc_count < priv->rx_ring_size) {
		struct bcm_enet_desc *desc;
		struct sk_buff *skb;
		dma_addr_t p;
		int desc_idx;
		u32 len_stat;

		desc_idx = priv->rx_dirty_desc;
		desc = &priv->rx_desc_cpu[desc_idx];

		if (!priv->rx_skb[desc_idx]) {
			skb = netdev_alloc_skb(dev, priv->rx_skb_size);
			if (!skb)
				break;
			priv->rx_skb[desc_idx] = skb;

			p = dma_map_single(&priv->pdev->dev, skb->data,
					   priv->rx_skb_size,
					   DMA_FROM_DEVICE);
			desc->address = p;
		}

		len_stat = priv->rx_skb_size << DMADESC_LENGTH_SHIFT;
		len_stat |= DMADESC_OWNER_MASK;
		if (priv->rx_dirty_desc == priv->rx_ring_size - 1) {
			len_stat |= DMADESC_WRAP_MASK;
			priv->rx_dirty_desc = 0;
		} else {
			priv->rx_dirty_desc++;
		}
		wmb();
		desc->len_stat = len_stat;

		priv->rx_desc_count++;

		/* tell dma engine we allocated one buffer */
		enet_dma_writel(priv, 1, ENETDMA_BUFALLOC_REG(priv->rx_chan));
	}

	/* If rx ring is still empty, set a timer to try allocating
	 * again at a later time. */
	if (priv->rx_desc_count == 0 && netif_running(dev)) {
		dev_warn(&priv->pdev->dev, "unable to refill rx ring\n");
		priv->rx_timeout.expires = jiffies + HZ;
		add_timer(&priv->rx_timeout);
	}

	return 0;
}

/*
 * timer callback to defer refill rx queue in case we're OOM
 */
static void bcm_enet_refill_rx_timer(unsigned long data)
{
	struct net_device *dev;
	struct bcm_enet_priv *priv;

	dev = (struct net_device *)data;
	priv = netdev_priv(dev);

	spin_lock(&priv->rx_lock);
	bcm_enet_refill_rx((struct net_device *)data);
	spin_unlock(&priv->rx_lock);
}

/*
 * extract packet from rx queue
 */
static int bcm_enet_receive_queue(struct net_device *dev, int budget)
{
	struct bcm_enet_priv *priv;
	struct device *kdev;
	int processed;

	priv = netdev_priv(dev);
	kdev = &priv->pdev->dev;
	processed = 0;

	/* don't scan ring further than number of refilled
	 * descriptor */
	if (budget > priv->rx_desc_count)
		budget = priv->rx_desc_count;

	do {
		struct bcm_enet_desc *desc;
		struct sk_buff *skb;
		int desc_idx;
		u32 len_stat;
		unsigned int len;

		desc_idx = priv->rx_curr_desc;
		desc = &priv->rx_desc_cpu[desc_idx];

		/* make sure we actually read the descriptor status at
		 * each loop */
		rmb();

		len_stat = desc->len_stat;

		/* break if dma ownership belongs to hw */
		if (len_stat & DMADESC_OWNER_MASK)
			break;

		processed++;
		priv->rx_curr_desc++;
		if (priv->rx_curr_desc == priv->rx_ring_size)
			priv->rx_curr_desc = 0;
		priv->rx_desc_count--;

		/* if the packet does not have start of packet _and_
		 * end of packet flag set, then just recycle it */
		if ((len_stat & DMADESC_ESOP_MASK) != DMADESC_ESOP_MASK) {
			priv->stats.rx_dropped++;
			continue;
		}

		/* recycle packet if it's marked as bad */
		if (unlikely(len_stat & DMADESC_ERR_MASK)) {
			priv->stats.rx_errors++;

			if (len_stat & DMADESC_OVSIZE_MASK)
				priv->stats.rx_length_errors++;
			if (len_stat & DMADESC_CRC_MASK)
				priv->stats.rx_crc_errors++;
			if (len_stat & DMADESC_UNDER_MASK)
				priv->stats.rx_frame_errors++;
			if (len_stat & DMADESC_OV_MASK)
				priv->stats.rx_fifo_errors++;
			continue;
		}

		/* valid packet */
		skb = priv->rx_skb[desc_idx];
		len = (len_stat & DMADESC_LENGTH_MASK) >> DMADESC_LENGTH_SHIFT;
		/* don't include FCS */
		len -= 4;

		if (len < copybreak) {
			struct sk_buff *nskb;

			nskb = netdev_alloc_skb(dev, len + NET_IP_ALIGN);
			if (!nskb) {
				/* forget packet, just rearm desc */
				priv->stats.rx_dropped++;
				continue;
			}

			/* since we're copying the data, we can align
			 * them properly */
			skb_reserve(nskb, NET_IP_ALIGN);
			dma_sync_single_for_cpu(kdev, desc->address,
						len, DMA_FROM_DEVICE);
			memcpy(nskb->data, skb->data, len);
			dma_sync_single_for_device(kdev, desc->address,
						   len, DMA_FROM_DEVICE);
			skb = nskb;
		} else {
			dma_unmap_single(&priv->pdev->dev, desc->address,
					 priv->rx_skb_size, DMA_FROM_DEVICE);
			priv->rx_skb[desc_idx] = NULL;
		}

		skb_put(skb, len);
		skb->dev = dev;
		skb->protocol = eth_type_trans(skb, dev);
		priv->stats.rx_packets++;
		priv->stats.rx_bytes += len;
		dev->last_rx = jiffies;
		netif_receive_skb(skb);

	} while (--budget > 0);

	if (processed || !priv->rx_desc_count) {
		bcm_enet_refill_rx(dev);

		/* kick rx dma */
		enet_dma_writel(priv, ENETDMA_CHANCFG_EN_MASK,
				ENETDMA_CHANCFG_REG(priv->rx_chan));
	}

	return processed;
}


/*
 * try to or force reclaim of transmitted buffers
 */
static int bcm_enet_tx_reclaim(struct net_device *dev, int force)
{
	struct bcm_enet_priv *priv;
	int released;

	priv = netdev_priv(dev);
	released = 0;

	while (priv->tx_desc_count < priv->tx_ring_size) {
		struct bcm_enet_desc *desc;
		struct sk_buff *skb;

		/* We run in a bh and fight against start_xmit, which
		 * is called with bh disabled  */
		spin_lock(&priv->tx_lock);

		desc = &priv->tx_desc_cpu[priv->tx_dirty_desc];

		if (!force && (desc->len_stat & DMADESC_OWNER_MASK)) {
			spin_unlock(&priv->tx_lock);
			break;
		}

		/* ensure other field of the descriptor were not read
		 * before we checked ownership */
		rmb();

		skb = priv->tx_skb[priv->tx_dirty_desc];
		priv->tx_skb[priv->tx_dirty_desc] = NULL;
		dma_unmap_single(&priv->pdev->dev, desc->address, skb->len,
				 DMA_TO_DEVICE);

		priv->tx_dirty_desc++;
		if (priv->tx_dirty_desc == priv->tx_ring_size)
			priv->tx_dirty_desc = 0;
		priv->tx_desc_count++;

		spin_unlock(&priv->tx_lock);

		if (desc->len_stat & DMADESC_UNDER_MASK)
			priv->stats.tx_errors++;

		dev_kfree_skb(skb);
		released++;
	}

	if (netif_queue_stopped(dev) && released)
		netif_wake_queue(dev);

	return released;
}

/*
 * poll func, called by network core
 */
static int bcm_enet_poll(struct napi_struct *napi, int budget)
{
	struct bcm_enet_priv *priv;
	struct net_device *dev;
	int tx_work_done, rx_work_done;

	priv = container_of(napi, struct bcm_enet_priv, napi);
	dev = priv->net_dev;

	/* ack interrupts */
	enet_dma_writel(priv, ENETDMA_IR_PKTDONE_MASK,
			ENETDMA_IR_REG(priv->rx_chan));
	enet_dma_writel(priv, ENETDMA_IR_PKTDONE_MASK,
			ENETDMA_IR_REG(priv->tx_chan));

	/* reclaim sent skb */
	tx_work_done = bcm_enet_tx_reclaim(dev, 0);

	spin_lock(&priv->rx_lock);
	rx_work_done = bcm_enet_receive_queue(dev, budget);
	spin_unlock(&priv->rx_lock);

	if (rx_work_done >= budget || tx_work_done > 0) {
		/* rx/tx queue is not yet empty/clean */
		return rx_work_done;
	}

	/* no more packet in rx/tx queue, remove device from poll
	 * queue */
	napi_complete(napi);

	/* restore rx/tx interrupt */
	enet_dma_writel(priv, ENETDMA_IR_PKTDONE_MASK,
			ENETDMA_IRMASK_REG(priv->rx_chan));
	enet_dma_writel(priv, ENETDMA_IR_PKTDONE_MASK,
			ENETDMA_IRMASK_REG(priv->tx_chan));

	return rx_work_done;
}

/*
 * mac interrupt handler
 */
static irqreturn_t bcm_enet_isr_mac(int irq, void *dev_id)
{
	struct net_device *dev;
	struct bcm_enet_priv *priv;
	u32 stat;

	dev = dev_id;
	priv = netdev_priv(dev);

	stat = enet_readl(priv, ENET_IR_REG);
	if (!(stat & ENET_IR_MIB))
		return IRQ_NONE;

	/* clear & mask interrupt */
	enet_writel(priv, ENET_IR_MIB, ENET_IR_REG);
	enet_writel(priv, 0, ENET_IRMASK_REG);

	/* read mib registers in workqueue */
	schedule_work(&priv->mib_update_task);

	return IRQ_HANDLED;
}

/*
 * rx/tx dma interrupt handler
 */
static irqreturn_t bcm_enet_isr_dma(int irq, void *dev_id)
{
	struct net_device *dev;
	struct bcm_enet_priv *priv;

	dev = dev_id;
	priv = netdev_priv(dev);

	/* mask rx/tx interrupts */
	enet_dma_writel(priv, 0, ENETDMA_IRMASK_REG(priv->rx_chan));
	enet_dma_writel(priv, 0, ENETDMA_IRMASK_REG(priv->tx_chan));

	napi_schedule(&priv->napi);

	return IRQ_HANDLED;
}

/*
 * tx request callback
 */
static int bcm_enet_start_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct bcm_enet_priv *priv;
	struct bcm_enet_desc *desc;
	u32 len_stat;
	int ret;

	priv = netdev_priv(dev);

	/* lock against tx reclaim */
	spin_lock(&priv->tx_lock);

	/* make sure  the tx hw queue  is not full,  should not happen
	 * since we stop queue before it's the case */
	if (unlikely(!priv->tx_desc_count)) {
		netif_stop_queue(dev);
		dev_err(&priv->pdev->dev, "xmit called with no tx desc "
			"available?\n");
		ret = NETDEV_TX_BUSY;
		goto out_unlock;
	}

	/* point to the next available desc */
	desc = &priv->tx_desc_cpu[priv->tx_curr_desc];
	priv->tx_skb[priv->tx_curr_desc] = skb;

	/* fill descriptor */
	desc->address = dma_map_single(&priv->pdev->dev, skb->data, skb->len,
				       DMA_TO_DEVICE);

	len_stat = (skb->len << DMADESC_LENGTH_SHIFT) & DMADESC_LENGTH_MASK;
	len_stat |= DMADESC_ESOP_MASK |
		DMADESC_APPEND_CRC |
		DMADESC_OWNER_MASK;

	priv->tx_curr_desc++;
	if (priv->tx_curr_desc == priv->tx_ring_size) {
		priv->tx_curr_desc = 0;
		len_stat |= DMADESC_WRAP_MASK;
	}
	priv->tx_desc_count--;

	/* dma might be already polling, make sure we update desc
	 * fields in correct order */
	wmb();
	desc->len_stat = len_stat;
	wmb();

	/* kick tx dma */
	enet_dma_writel(priv, ENETDMA_CHANCFG_EN_MASK,
			ENETDMA_CHANCFG_REG(priv->tx_chan));

	/* stop queue if no more desc available */
	if (!priv->tx_desc_count)
		netif_stop_queue(dev);

	priv->stats.tx_bytes += skb->len;
	priv->stats.tx_packets++;
	dev->trans_start = jiffies;
	ret = NETDEV_TX_OK;

out_unlock:
	spin_unlock(&priv->tx_lock);
	return ret;
}

/*
 * Change the interface's mac address.
 */
static int bcm_enet_set_mac_address(struct net_device *dev, void *p)
{
	struct bcm_enet_priv *priv;
	struct sockaddr *addr = p;
	u32 val;

	priv = netdev_priv(dev);
	memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);

	/* use perfect match register 0 to store my mac address */
	val = (dev->dev_addr[2] << 24) | (dev->dev_addr[3] << 16) |
		(dev->dev_addr[4] << 8) | dev->dev_addr[5];
	enet_writel(priv, val, ENET_PML_REG(0));

	val = (dev->dev_addr[0] << 8 | dev->dev_addr[1]);
	val |= ENET_PMH_DATAVALID_MASK;
	enet_writel(priv, val, ENET_PMH_REG(0));

	return 0;
}

/*
 * Change rx mode (promiscous/allmulti) and update multicast list
 */
static void bcm_enet_set_multicast_list(struct net_device *dev)
{
	struct bcm_enet_priv *priv;
	struct dev_mc_list *mc_list;
	u32 val;
	int i;

	priv = netdev_priv(dev);

	val = enet_readl(priv, ENET_RXCFG_REG);

	if (dev->flags & IFF_PROMISC)
		val |= ENET_RXCFG_PROMISC_MASK;
	else
		val &= ~ENET_RXCFG_PROMISC_MASK;

	/* only 3 perfect match registers left, first one is used for
	 * own mac address */
	if ((dev->flags & IFF_ALLMULTI) || dev->mc_count > 3)
		val |= ENET_RXCFG_ALLMCAST_MASK;
	else
		val &= ~ENET_RXCFG_ALLMCAST_MASK;

	/* no need to set perfect match registers if we catch all
	 * multicast */
	if (val & ENET_RXCFG_ALLMCAST_MASK) {
		enet_writel(priv, val, ENET_RXCFG_REG);
		return;
	}

	for (i = 0, mc_list = dev->mc_list;
	     (mc_list != NULL) && (i < dev->mc_count) && (i < 3);
	     i++, mc_list = mc_list->next) {
		u8 *dmi_addr;
		u32 tmp;

		/* filter non ethernet address */
		if (mc_list->dmi_addrlen != 6)
			continue;

		/* update perfect match registers */
		dmi_addr = mc_list->dmi_addr;
		tmp = (dmi_addr[2] << 24) | (dmi_addr[3] << 16) |
			(dmi_addr[4] << 8) | dmi_addr[5];
		enet_writel(priv, tmp, ENET_PML_REG(i + 1));

		tmp = (dmi_addr[0] << 8 | dmi_addr[1]);
		tmp |= ENET_PMH_DATAVALID_MASK;
		enet_writel(priv, tmp, ENET_PMH_REG(i + 1));
	}

	for (; i < 3; i++) {
		enet_writel(priv, 0, ENET_PML_REG(i + 1));
		enet_writel(priv, 0, ENET_PMH_REG(i + 1));
	}

	enet_writel(priv, val, ENET_RXCFG_REG);
}

/*
 * set mac duplex parameters
 */
static void bcm_enet_set_duplex(struct bcm_enet_priv *priv, int fullduplex)
{
	u32 val;

	val = enet_readl(priv, ENET_TXCTL_REG);
	if (fullduplex)
		val |= ENET_TXCTL_FD_MASK;
	else
		val &= ~ENET_TXCTL_FD_MASK;
	enet_writel(priv, val, ENET_TXCTL_REG);
}

/*
 * set mac flow control parameters
 */
static void bcm_enet_set_flow(struct bcm_enet_priv *priv, int rx_en, int tx_en)
{
	u32 val;

	/* rx flow control (pause frame handling) */
	val = enet_readl(priv, ENET_RXCFG_REG);
	if (rx_en)
		val |= ENET_RXCFG_ENFLOW_MASK;
	else
		val &= ~ENET_RXCFG_ENFLOW_MASK;
	enet_writel(priv, val, ENET_RXCFG_REG);

	/* tx flow control (pause frame generation) */
	val = enet_dma_readl(priv, ENETDMA_CFG_REG);
	if (tx_en)
		val |= ENETDMA_CFG_FLOWCH_MASK(priv->rx_chan);
	else
		val &= ~ENETDMA_CFG_FLOWCH_MASK(priv->rx_chan);
	enet_dma_writel(priv, val, ENETDMA_CFG_REG);
}

/*
 * link changed callback (from phylib)
 */
static void bcm_enet_adjust_phy_link(struct net_device *dev)
{
	struct bcm_enet_priv *priv;
	struct phy_device *phydev;
	int status_changed;

	priv = netdev_priv(dev);
	phydev = priv->phydev;
	status_changed = 0;

	if (priv->old_link != phydev->link) {
		status_changed = 1;
		priv->old_link = phydev->link;
	}

	/* reflect duplex change in mac configuration */
	if (phydev->link && phydev->duplex != priv->old_duplex) {
		bcm_enet_set_duplex(priv,
				    (phydev->duplex == DUPLEX_FULL) ? 1 : 0);
		status_changed = 1;
		priv->old_duplex = phydev->duplex;
	}

	/* enable flow control if remote advertise it (trust phylib to
	 * check that duplex is full */
	if (phydev->link && phydev->pause != priv->old_pause) {
		int rx_pause_en, tx_pause_en;

		if (phydev->pause) {
			/* pause was advertised by lpa and us */
			rx_pause_en = 1;
			tx_pause_en = 1;
		} else if (!priv->pause_auto) {
			/* pause setting overrided by user */
			rx_pause_en = priv->pause_rx;
			tx_pause_en = priv->pause_tx;
		} else {
			rx_pause_en = 0;
			tx_pause_en = 0;
		}

		bcm_enet_set_flow(priv, rx_pause_en, tx_pause_en);
		status_changed = 1;
		priv->old_pause = phydev->pause;
	}

	if (status_changed) {
		pr_info("%s: link %s", dev->name, phydev->link ?
			"UP" : "DOWN");
		if (phydev->link)
			pr_cont(" - %d/%s - flow control %s", phydev->speed,
			       DUPLEX_FULL == phydev->duplex ? "full" : "half",
			       phydev->pause == 1 ? "rx&tx" : "off");

		pr_cont("\n");
	}
}

/*
 * link changed callback (if phylib is not used)
 */
static void bcm_enet_adjust_link(struct net_device *dev)
{
	struct bcm_enet_priv *priv;

	priv = netdev_priv(dev);
	bcm_enet_set_duplex(priv, priv->force_duplex_full);
	bcm_enet_set_flow(priv, priv->pause_rx, priv->pause_tx);
	netif_carrier_on(dev);

	pr_info("%s: link forced UP - %d/%s - flow control %s/%s\n",
		dev->name,
		priv->force_speed_100 ? 100 : 10,
		priv->force_duplex_full ? "full" : "half",
		priv->pause_rx ? "rx" : "off",
		priv->pause_tx ? "tx" : "off");
}

/*
 * open callback, allocate dma rings & buffers and start rx operation
 */
static int bcm_enet_open(struct net_device *dev)
{
	struct bcm_enet_priv *priv;
	struct sockaddr addr;
	struct device *kdev;
	struct phy_device *phydev;
	int i, ret;
	unsigned int size;
	char phy_id[MII_BUS_ID_SIZE + 3];
	void *p;
	u32 val;

	priv = netdev_priv(dev);
	kdev = &priv->pdev->dev;

	if (priv->has_phy) {
		/* connect to PHY */
		snprintf(phy_id, sizeof(phy_id), PHY_ID_FMT,
			 priv->mac_id ? "1" : "0", priv->phy_id);

		phydev = phy_connect(dev, phy_id, &bcm_enet_adjust_phy_link, 0,
				     PHY_INTERFACE_MODE_MII);

		if (IS_ERR(phydev)) {
			dev_err(kdev, "could not attach to PHY\n");
			return PTR_ERR(phydev);
		}

		/* mask with MAC supported features */
		phydev->supported &= (SUPPORTED_10baseT_Half |
				      SUPPORTED_10baseT_Full |
				      SUPPORTED_100baseT_Half |
				      SUPPORTED_100baseT_Full |
				      SUPPORTED_Autoneg |
				      SUPPORTED_Pause |
				      SUPPORTED_MII);
		phydev->advertising = phydev->supported;

		if (priv->pause_auto && priv->pause_rx && priv->pause_tx)
			phydev->advertising |= SUPPORTED_Pause;
		else
			phydev->advertising &= ~SUPPORTED_Pause;

		dev_info(kdev, "attached PHY at address %d [%s]\n",
			 phydev->addr, phydev->drv->name);

		priv->old_link = 0;
		priv->old_duplex = -1;
		priv->old_pause = -1;
		priv->phydev = phydev;
	}

	/* mask all interrupts and request them */
	enet_writel(priv, 0, ENET_IRMASK_REG);
	enet_dma_writel(priv, 0, ENETDMA_IRMASK_REG(priv->rx_chan));
	enet_dma_writel(priv, 0, ENETDMA_IRMASK_REG(priv->tx_chan));

	ret = request_irq(dev->irq, bcm_enet_isr_mac, 0, dev->name, dev);
	if (ret)
		goto out_phy_disconnect;

	ret = request_irq(priv->irq_rx, bcm_enet_isr_dma,
			  IRQF_SAMPLE_RANDOM | IRQF_DISABLED, dev->name, dev);
	if (ret)
		goto out_freeirq;

	ret = request_irq(priv->irq_tx, bcm_enet_isr_dma,
			  IRQF_DISABLED, dev->name, dev);
	if (ret)
		goto out_freeirq_rx;

	/* initialize perfect match registers */
	for (i = 0; i < 4; i++) {
		enet_writel(priv, 0, ENET_PML_REG(i));
		enet_writel(priv, 0, ENET_PMH_REG(i));
	}

	/* write device mac address */
	memcpy(addr.sa_data, dev->dev_addr, ETH_ALEN);
	bcm_enet_set_mac_address(dev, &addr);

	/* allocate rx dma ring */
	size = priv->rx_ring_size * sizeof(struct bcm_enet_desc);
	p = dma_alloc_coherent(kdev, size, &priv->rx_desc_dma, GFP_KERNEL);
	if (!p) {
		dev_err(kdev, "cannot allocate rx ring %u\n", size);
		ret = -ENOMEM;
		goto out_freeirq_tx;
	}

	memset(p, 0, size);
	priv->rx_desc_alloc_size = size;
	priv->rx_desc_cpu = p;

	/* allocate tx dma ring */
	size = priv->tx_ring_size * sizeof(struct bcm_enet_desc);
	p = dma_alloc_coherent(kdev, size, &priv->tx_desc_dma, GFP_KERNEL);
	if (!p) {
		dev_err(kdev, "cannot allocate tx ring\n");
		ret = -ENOMEM;
		goto out_free_rx_ring;
	}

	memset(p, 0, size);
	priv->tx_desc_alloc_size = size;
	priv->tx_desc_cpu = p;

	priv->tx_skb = kzalloc(sizeof(struct sk_buff *) * priv->tx_ring_size,
			       GFP_KERNEL);
	if (!priv->tx_skb) {
		dev_err(kdev, "cannot allocate rx skb queue\n");
		ret = -ENOMEM;
		goto out_free_tx_ring;
	}

	priv->tx_desc_count = priv->tx_ring_size;
	priv->tx_dirty_desc = 0;
	priv->tx_curr_desc = 0;
	spin_lock_init(&priv->tx_lock);

	/* init & fill rx ring with skbs */
	priv->rx_skb = kzalloc(sizeof(struct sk_buff *) * priv->rx_ring_size,
			       GFP_KERNEL);
	if (!priv->rx_skb) {
		dev_err(kdev, "cannot allocate rx skb queue\n");
		ret = -ENOMEM;
		goto out_free_tx_skb;
	}

	priv->rx_desc_count = 0;
	priv->rx_dirty_desc = 0;
	priv->rx_curr_desc = 0;

	/* initialize flow control buffer allocation */
	enet_dma_writel(priv, ENETDMA_BUFALLOC_FORCE_MASK | 0,
			ENETDMA_BUFALLOC_REG(priv->rx_chan));

	if (bcm_enet_refill_rx(dev)) {
		dev_err(kdev, "cannot allocate rx skb queue\n");
		ret = -ENOMEM;
		goto out;
	}

	/* write rx & tx ring addresses */
	enet_dma_writel(priv, priv->rx_desc_dma,
			ENETDMA_RSTART_REG(priv->rx_chan));
	enet_dma_writel(priv, priv->tx_desc_dma,
			ENETDMA_RSTART_REG(priv->tx_chan));

	/* clear remaining state ram for rx & tx channel */
	enet_dma_writel(priv, 0, ENETDMA_SRAM2_REG(priv->rx_chan));
	enet_dma_writel(priv, 0, ENETDMA_SRAM2_REG(priv->tx_chan));
	enet_dma_writel(priv, 0, ENETDMA_SRAM3_REG(priv->rx_chan));
	enet_dma_writel(priv, 0, ENETDMA_SRAM3_REG(priv->tx_chan));
	enet_dma_writel(priv, 0, ENETDMA_SRAM4_REG(priv->rx_chan));
	enet_dma_writel(priv, 0, ENETDMA_SRAM4_REG(priv->tx_chan));

	/* set max rx/tx length */
	enet_writel(priv, priv->hw_mtu, ENET_RXMAXLEN_REG);
	enet_writel(priv, priv->hw_mtu, ENET_TXMAXLEN_REG);

	/* set dma maximum burst len */
	enet_dma_writel(priv, BCMENET_DMA_MAXBURST,
			ENETDMA_MAXBURST_REG(priv->rx_chan));
	enet_dma_writel(priv, BCMENET_DMA_MAXBURST,
			ENETDMA_MAXBURST_REG(priv->tx_chan));

	/* set correct transmit fifo watermark */
	enet_writel(priv, BCMENET_TX_FIFO_TRESH, ENET_TXWMARK_REG);

	/* set flow control low/high threshold to 1/3 / 2/3 */
	val = priv->rx_ring_size / 3;
	enet_dma_writel(priv, val, ENETDMA_FLOWCL_REG(priv->rx_chan));
	val = (priv->rx_ring_size * 2) / 3;
	enet_dma_writel(priv, val, ENETDMA_FLOWCH_REG(priv->rx_chan));

	/* all set, enable mac and interrupts, start dma engine and
	 * kick rx dma channel */
	wmb();
	enet_writel(priv, ENET_CTL_ENABLE_MASK, ENET_CTL_REG);
	enet_dma_writel(priv, ENETDMA_CFG_EN_MASK, ENETDMA_CFG_REG);
	enet_dma_writel(priv, ENETDMA_CHANCFG_EN_MASK,
			ENETDMA_CHANCFG_REG(priv->rx_chan));

	/* watch "mib counters about to overflow" interrupt */
	enet_writel(priv, ENET_IR_MIB, ENET_IR_REG);
	enet_writel(priv, ENET_IR_MIB, ENET_IRMASK_REG);

	/* watch "packet transferred" interrupt in rx and tx */
	enet_dma_writel(priv, ENETDMA_IR_PKTDONE_MASK,
			ENETDMA_IR_REG(priv->rx_chan));
	enet_dma_writel(priv, ENETDMA_IR_PKTDONE_MASK,
			ENETDMA_IR_REG(priv->tx_chan));

	/* make sure we enable napi before rx interrupt  */
	napi_enable(&priv->napi);

	enet_dma_writel(priv, ENETDMA_IR_PKTDONE_MASK,
			ENETDMA_IRMASK_REG(priv->rx_chan));
	enet_dma_writel(priv, ENETDMA_IR_PKTDONE_MASK,
			ENETDMA_IRMASK_REG(priv->tx_chan));

	if (priv->has_phy)
		phy_start(priv->phydev);
	else
		bcm_enet_adjust_link(dev);

	netif_start_queue(dev);
	return 0;

out:
	for (i = 0; i < priv->rx_ring_size; i++) {