Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / ethernet / cadence / macb.c
diff --git a/kernel/drivers/net/ethernet/cadence/macb.c b/kernel/drivers/net/ethernet/cadence/macb.c
new file mode 100644 (file)
index 0000000..fc646a4
--- /dev/null
@@ -0,0 +1,2935 @@
+/*
+ * Cadence MACB/GEM Ethernet Controller driver
+ *
+ * Copyright (C) 2004-2006 Atmel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+#include <linux/clk.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include <linux/circ_buf.h>
+#include <linux/slab.h>
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/gpio.h>
+#include <linux/interrupt.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/dma-mapping.h>
+#include <linux/platform_data/macb.h>
+#include <linux/platform_device.h>
+#include <linux/phy.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/of_mdio.h>
+#include <linux/of_net.h>
+
+#include "macb.h"
+
+#define MACB_RX_BUFFER_SIZE    128
+#define RX_BUFFER_MULTIPLE     64  /* bytes */
+#define RX_RING_SIZE           512 /* must be power of 2 */
+#define RX_RING_BYTES          (sizeof(struct macb_dma_desc) * RX_RING_SIZE)
+
+#define TX_RING_SIZE           128 /* must be power of 2 */
+#define TX_RING_BYTES          (sizeof(struct macb_dma_desc) * TX_RING_SIZE)
+
+/* level of occupied TX descriptors under which we wake up TX process */
+#define MACB_TX_WAKEUP_THRESH  (3 * TX_RING_SIZE / 4)
+
+#define MACB_RX_INT_FLAGS      (MACB_BIT(RCOMP) | MACB_BIT(RXUBR)      \
+                                | MACB_BIT(ISR_ROVR))
+#define MACB_TX_ERR_FLAGS      (MACB_BIT(ISR_TUND)                     \
+                                       | MACB_BIT(ISR_RLE)             \
+                                       | MACB_BIT(TXERR))
+#define MACB_TX_INT_FLAGS      (MACB_TX_ERR_FLAGS | MACB_BIT(TCOMP))
+
+#define MACB_MAX_TX_LEN                ((unsigned int)((1 << MACB_TX_FRMLEN_SIZE) - 1))
+#define GEM_MAX_TX_LEN         ((unsigned int)((1 << GEM_TX_FRMLEN_SIZE) - 1))
+
+/*
+ * Graceful stop timeouts in us. We should allow up to
+ * 1 frame time (10 Mbits/s, full-duplex, ignoring collisions)
+ */
+#define MACB_HALT_TIMEOUT      1230
+
+/* Ring buffer accessors */
+static unsigned int macb_tx_ring_wrap(unsigned int index)
+{
+       return index & (TX_RING_SIZE - 1);
+}
+
+static struct macb_dma_desc *macb_tx_desc(struct macb_queue *queue,
+                                         unsigned int index)
+{
+       return &queue->tx_ring[macb_tx_ring_wrap(index)];
+}
+
+static struct macb_tx_skb *macb_tx_skb(struct macb_queue *queue,
+                                      unsigned int index)
+{
+       return &queue->tx_skb[macb_tx_ring_wrap(index)];
+}
+
+static dma_addr_t macb_tx_dma(struct macb_queue *queue, unsigned int index)
+{
+       dma_addr_t offset;
+
+       offset = macb_tx_ring_wrap(index) * sizeof(struct macb_dma_desc);
+
+       return queue->tx_ring_dma + offset;
+}
+
+static unsigned int macb_rx_ring_wrap(unsigned int index)
+{
+       return index & (RX_RING_SIZE - 1);
+}
+
+static struct macb_dma_desc *macb_rx_desc(struct macb *bp, unsigned int index)
+{
+       return &bp->rx_ring[macb_rx_ring_wrap(index)];
+}
+
+static void *macb_rx_buffer(struct macb *bp, unsigned int index)
+{
+       return bp->rx_buffers + bp->rx_buffer_size * macb_rx_ring_wrap(index);
+}
+
+static void macb_set_hwaddr(struct macb *bp)
+{
+       u32 bottom;
+       u16 top;
+
+       bottom = cpu_to_le32(*((u32 *)bp->dev->dev_addr));
+       macb_or_gem_writel(bp, SA1B, bottom);
+       top = cpu_to_le16(*((u16 *)(bp->dev->dev_addr + 4)));
+       macb_or_gem_writel(bp, SA1T, top);
+
+       /* Clear unused address register sets */
+       macb_or_gem_writel(bp, SA2B, 0);
+       macb_or_gem_writel(bp, SA2T, 0);
+       macb_or_gem_writel(bp, SA3B, 0);
+       macb_or_gem_writel(bp, SA3T, 0);
+       macb_or_gem_writel(bp, SA4B, 0);
+       macb_or_gem_writel(bp, SA4T, 0);
+}
+
+static void macb_get_hwaddr(struct macb *bp)
+{
+       struct macb_platform_data *pdata;
+       u32 bottom;
+       u16 top;
+       u8 addr[6];
+       int i;
+
+       pdata = dev_get_platdata(&bp->pdev->dev);
+
+       /* Check all 4 address register for vaild address */
+       for (i = 0; i < 4; i++) {
+               bottom = macb_or_gem_readl(bp, SA1B + i * 8);
+               top = macb_or_gem_readl(bp, SA1T + i * 8);
+
+               if (pdata && pdata->rev_eth_addr) {
+                       addr[5] = bottom & 0xff;
+                       addr[4] = (bottom >> 8) & 0xff;
+                       addr[3] = (bottom >> 16) & 0xff;
+                       addr[2] = (bottom >> 24) & 0xff;
+                       addr[1] = top & 0xff;
+                       addr[0] = (top & 0xff00) >> 8;
+               } else {
+                       addr[0] = bottom & 0xff;
+                       addr[1] = (bottom >> 8) & 0xff;
+                       addr[2] = (bottom >> 16) & 0xff;
+                       addr[3] = (bottom >> 24) & 0xff;
+                       addr[4] = top & 0xff;
+                       addr[5] = (top >> 8) & 0xff;
+               }
+
+               if (is_valid_ether_addr(addr)) {
+                       memcpy(bp->dev->dev_addr, addr, sizeof(addr));
+                       return;
+               }
+       }
+
+       netdev_info(bp->dev, "invalid hw address, using random\n");
+       eth_hw_addr_random(bp->dev);
+}
+
+static int macb_mdio_read(struct mii_bus *bus, int mii_id, int regnum)
+{
+       struct macb *bp = bus->priv;
+       int value;
+
+       macb_writel(bp, MAN, (MACB_BF(SOF, MACB_MAN_SOF)
+                             | MACB_BF(RW, MACB_MAN_READ)
+                             | MACB_BF(PHYA, mii_id)
+                             | MACB_BF(REGA, regnum)
+                             | MACB_BF(CODE, MACB_MAN_CODE)));
+
+       /* wait for end of transfer */
+       while (!MACB_BFEXT(IDLE, macb_readl(bp, NSR)))
+               cpu_relax();
+
+       value = MACB_BFEXT(DATA, macb_readl(bp, MAN));
+
+       return value;
+}
+
+static int macb_mdio_write(struct mii_bus *bus, int mii_id, int regnum,
+                          u16 value)
+{
+       struct macb *bp = bus->priv;
+
+       macb_writel(bp, MAN, (MACB_BF(SOF, MACB_MAN_SOF)
+                             | MACB_BF(RW, MACB_MAN_WRITE)
+                             | MACB_BF(PHYA, mii_id)
+                             | MACB_BF(REGA, regnum)
+                             | MACB_BF(CODE, MACB_MAN_CODE)
+                             | MACB_BF(DATA, value)));
+
+       /* wait for end of transfer */
+       while (!MACB_BFEXT(IDLE, macb_readl(bp, NSR)))
+               cpu_relax();
+
+       return 0;
+}
+
+/**
+ * macb_set_tx_clk() - Set a clock to a new frequency
+ * @clk                Pointer to the clock to change
+ * @rate       New frequency in Hz
+ * @dev                Pointer to the struct net_device
+ */
+static void macb_set_tx_clk(struct clk *clk, int speed, struct net_device *dev)
+{
+       long ferr, rate, rate_rounded;
+
+       if (!clk)
+               return;
+
+       switch (speed) {
+       case SPEED_10:
+               rate = 2500000;
+               break;
+       case SPEED_100:
+               rate = 25000000;
+               break;
+       case SPEED_1000:
+               rate = 125000000;
+               break;
+       default:
+               return;
+       }
+
+       rate_rounded = clk_round_rate(clk, rate);
+       if (rate_rounded < 0)
+               return;
+
+       /* RGMII allows 50 ppm frequency error. Test and warn if this limit
+        * is not satisfied.
+        */
+       ferr = abs(rate_rounded - rate);
+       ferr = DIV_ROUND_UP(ferr, rate / 100000);
+       if (ferr > 5)
+               netdev_warn(dev, "unable to generate target frequency: %ld Hz\n",
+                               rate);
+
+       if (clk_set_rate(clk, rate_rounded))
+               netdev_err(dev, "adjusting tx_clk failed.\n");
+}
+
+static void macb_handle_link_change(struct net_device *dev)
+{
+       struct macb *bp = netdev_priv(dev);
+       struct phy_device *phydev = bp->phy_dev;
+       unsigned long flags;
+
+       int status_change = 0;
+
+       spin_lock_irqsave(&bp->lock, flags);
+
+       if (phydev->link) {
+               if ((bp->speed != phydev->speed) ||
+                   (bp->duplex != phydev->duplex)) {
+                       u32 reg;
+
+                       reg = macb_readl(bp, NCFGR);
+                       reg &= ~(MACB_BIT(SPD) | MACB_BIT(FD));
+                       if (macb_is_gem(bp))
+                               reg &= ~GEM_BIT(GBE);
+
+                       if (phydev->duplex)
+                               reg |= MACB_BIT(FD);
+                       if (phydev->speed == SPEED_100)
+                               reg |= MACB_BIT(SPD);
+                       if (phydev->speed == SPEED_1000 &&
+                           bp->caps & MACB_CAPS_GIGABIT_MODE_AVAILABLE)
+                               reg |= GEM_BIT(GBE);
+
+                       macb_or_gem_writel(bp, NCFGR, reg);
+
+                       bp->speed = phydev->speed;
+                       bp->duplex = phydev->duplex;
+                       status_change = 1;
+               }
+       }
+
+       if (phydev->link != bp->link) {
+               if (!phydev->link) {
+                       bp->speed = 0;
+                       bp->duplex = -1;
+               }
+               bp->link = phydev->link;
+
+               status_change = 1;
+       }
+
+       spin_unlock_irqrestore(&bp->lock, flags);
+
+       if (status_change) {
+               if (phydev->link) {
+                       /* Update the TX clock rate if and only if the link is
+                        * up and there has been a link change.
+                        */
+                       macb_set_tx_clk(bp->tx_clk, phydev->speed, dev);
+
+                       netif_carrier_on(dev);
+                       netdev_info(dev, "link up (%d/%s)\n",
+                                   phydev->speed,
+                                   phydev->duplex == DUPLEX_FULL ?
+                                   "Full" : "Half");
+               } else {
+                       netif_carrier_off(dev);
+                       netdev_info(dev, "link down\n");
+               }
+       }
+}
+
+/* based on au1000_eth. c*/
+static int macb_mii_probe(struct net_device *dev)
+{
+       struct macb *bp = netdev_priv(dev);
+       struct macb_platform_data *pdata;
+       struct phy_device *phydev;
+       int phy_irq;
+       int ret;
+
+       phydev = phy_find_first(bp->mii_bus);
+       if (!phydev) {
+               netdev_err(dev, "no PHY found\n");
+               return -ENXIO;
+       }
+
+       pdata = dev_get_platdata(&bp->pdev->dev);
+       if (pdata && gpio_is_valid(pdata->phy_irq_pin)) {
+               ret = devm_gpio_request(&bp->pdev->dev, pdata->phy_irq_pin, "phy int");
+               if (!ret) {
+                       phy_irq = gpio_to_irq(pdata->phy_irq_pin);
+                       phydev->irq = (phy_irq < 0) ? PHY_POLL : phy_irq;
+               }
+       }
+
+       /* attach the mac to the phy */
+       ret = phy_connect_direct(dev, phydev, &macb_handle_link_change,
+                                bp->phy_interface);
+       if (ret) {
+               netdev_err(dev, "Could not attach to PHY\n");
+               return ret;
+       }
+
+       /* mask with MAC supported features */
+       if (macb_is_gem(bp) && bp->caps & MACB_CAPS_GIGABIT_MODE_AVAILABLE)
+               phydev->supported &= PHY_GBIT_FEATURES;
+       else
+               phydev->supported &= PHY_BASIC_FEATURES;
+
+       if (bp->caps & MACB_CAPS_NO_GIGABIT_HALF)
+               phydev->supported &= ~SUPPORTED_1000baseT_Half;
+
+       phydev->advertising = phydev->supported;
+
+       bp->link = 0;
+       bp->speed = 0;
+       bp->duplex = -1;
+       bp->phy_dev = phydev;
+
+       return 0;
+}
+
+static int macb_mii_init(struct macb *bp)
+{
+       struct macb_platform_data *pdata;
+       struct device_node *np;
+       int err = -ENXIO, i;
+
+       /* Enable management port */
+       macb_writel(bp, NCR, MACB_BIT(MPE));
+
+       bp->mii_bus = mdiobus_alloc();
+       if (bp->mii_bus == NULL) {
+               err = -ENOMEM;
+               goto err_out;
+       }
+
+       bp->mii_bus->name = "MACB_mii_bus";
+       bp->mii_bus->read = &macb_mdio_read;
+       bp->mii_bus->write = &macb_mdio_write;
+       snprintf(bp->mii_bus->id, MII_BUS_ID_SIZE, "%s-%x",
+               bp->pdev->name, bp->pdev->id);
+       bp->mii_bus->priv = bp;
+       bp->mii_bus->parent = &bp->dev->dev;
+       pdata = dev_get_platdata(&bp->pdev->dev);
+
+       bp->mii_bus->irq = kmalloc(sizeof(int)*PHY_MAX_ADDR, GFP_KERNEL);
+       if (!bp->mii_bus->irq) {
+               err = -ENOMEM;
+               goto err_out_free_mdiobus;
+       }
+
+       dev_set_drvdata(&bp->dev->dev, bp->mii_bus);
+
+       np = bp->pdev->dev.of_node;
+       if (np) {
+               /* try dt phy registration */
+               err = of_mdiobus_register(bp->mii_bus, np);
+
+               /* fallback to standard phy registration if no phy were
+                  found during dt phy registration */
+               if (!err && !phy_find_first(bp->mii_bus)) {
+                       for (i = 0; i < PHY_MAX_ADDR; i++) {
+                               struct phy_device *phydev;
+
+                               phydev = mdiobus_scan(bp->mii_bus, i);
+                               if (IS_ERR(phydev)) {
+                                       err = PTR_ERR(phydev);
+                                       break;
+                               }
+                       }
+
+                       if (err)
+                               goto err_out_unregister_bus;
+               }
+       } else {
+               for (i = 0; i < PHY_MAX_ADDR; i++)
+                       bp->mii_bus->irq[i] = PHY_POLL;
+
+               if (pdata)
+                       bp->mii_bus->phy_mask = pdata->phy_mask;
+
+               err = mdiobus_register(bp->mii_bus);
+       }
+
+       if (err)
+               goto err_out_free_mdio_irq;
+
+       err = macb_mii_probe(bp->dev);
+       if (err)
+               goto err_out_unregister_bus;
+
+       return 0;
+
+err_out_unregister_bus:
+       mdiobus_unregister(bp->mii_bus);
+err_out_free_mdio_irq:
+       kfree(bp->mii_bus->irq);
+err_out_free_mdiobus:
+       mdiobus_free(bp->mii_bus);
+err_out:
+       return err;
+}
+
+static void macb_update_stats(struct macb *bp)
+{
+       u32 __iomem *reg = bp->regs + MACB_PFR;
+       u32 *p = &bp->hw_stats.macb.rx_pause_frames;
+       u32 *end = &bp->hw_stats.macb.tx_pause_frames + 1;
+
+       WARN_ON((unsigned long)(end - p - 1) != (MACB_TPF - MACB_PFR) / 4);
+
+       for(; p < end; p++, reg++)
+               *p += readl_relaxed(reg);
+}
+
+static int macb_halt_tx(struct macb *bp)
+{
+       unsigned long   halt_time, timeout;
+       u32             status;
+
+       macb_writel(bp, NCR, macb_readl(bp, NCR) | MACB_BIT(THALT));
+
+       timeout = jiffies + usecs_to_jiffies(MACB_HALT_TIMEOUT);
+       do {
+               halt_time = jiffies;
+               status = macb_readl(bp, TSR);
+               if (!(status & MACB_BIT(TGO)))
+                       return 0;
+
+               usleep_range(10, 250);
+       } while (time_before(halt_time, timeout));
+
+       return -ETIMEDOUT;
+}
+
+static void macb_tx_unmap(struct macb *bp, struct macb_tx_skb *tx_skb)
+{
+       if (tx_skb->mapping) {
+               if (tx_skb->mapped_as_page)
+                       dma_unmap_page(&bp->pdev->dev, tx_skb->mapping,
+                                      tx_skb->size, DMA_TO_DEVICE);
+               else
+                       dma_unmap_single(&bp->pdev->dev, tx_skb->mapping,
+                                        tx_skb->size, DMA_TO_DEVICE);
+               tx_skb->mapping = 0;
+       }
+
+       if (tx_skb->skb) {
+               dev_kfree_skb_any(tx_skb->skb);
+               tx_skb->skb = NULL;
+       }
+}
+
+static void macb_tx_error_task(struct work_struct *work)
+{
+       struct macb_queue       *queue = container_of(work, struct macb_queue,
+                                                     tx_error_task);
+       struct macb             *bp = queue->bp;
+       struct macb_tx_skb      *tx_skb;
+       struct macb_dma_desc    *desc;
+       struct sk_buff          *skb;
+       unsigned int            tail;
+       unsigned long           flags;
+
+       netdev_vdbg(bp->dev, "macb_tx_error_task: q = %u, t = %u, h = %u\n",
+                   (unsigned int)(queue - bp->queues),
+                   queue->tx_tail, queue->tx_head);
+
+       /* Prevent the queue IRQ handlers from running: each of them may call
+        * macb_tx_interrupt(), which in turn may call netif_wake_subqueue().
+        * As explained below, we have to halt the transmission before updating
+        * TBQP registers so we call netif_tx_stop_all_queues() to notify the
+        * network engine about the macb/gem being halted.
+        */
+       spin_lock_irqsave(&bp->lock, flags);
+
+       /* Make sure nobody is trying to queue up new packets */
+       netif_tx_stop_all_queues(bp->dev);
+
+       /*
+        * Stop transmission now
+        * (in case we have just queued new packets)
+        * macb/gem must be halted to write TBQP register
+        */
+       if (macb_halt_tx(bp))
+               /* Just complain for now, reinitializing TX path can be good */
+               netdev_err(bp->dev, "BUG: halt tx timed out\n");
+
+       /*
+        * Treat frames in TX queue including the ones that caused the error.
+        * Free transmit buffers in upper layer.
+        */
+       for (tail = queue->tx_tail; tail != queue->tx_head; tail++) {
+               u32     ctrl;
+
+               desc = macb_tx_desc(queue, tail);
+               ctrl = desc->ctrl;
+               tx_skb = macb_tx_skb(queue, tail);
+               skb = tx_skb->skb;
+
+               if (ctrl & MACB_BIT(TX_USED)) {
+                       /* skb is set for the last buffer of the frame */
+                       while (!skb) {
+                               macb_tx_unmap(bp, tx_skb);
+                               tail++;
+                               tx_skb = macb_tx_skb(queue, tail);
+                               skb = tx_skb->skb;
+                       }
+
+                       /* ctrl still refers to the first buffer descriptor
+                        * since it's the only one written back by the hardware
+                        */
+                       if (!(ctrl & MACB_BIT(TX_BUF_EXHAUSTED))) {
+                               netdev_vdbg(bp->dev, "txerr skb %u (data %p) TX complete\n",
+                                           macb_tx_ring_wrap(tail), skb->data);
+                               bp->stats.tx_packets++;
+                               bp->stats.tx_bytes += skb->len;
+                       }
+               } else {
+                       /*
+                        * "Buffers exhausted mid-frame" errors may only happen
+                        * if the driver is buggy, so complain loudly about those.
+                        * Statistics are updated by hardware.
+                        */
+                       if (ctrl & MACB_BIT(TX_BUF_EXHAUSTED))
+                               netdev_err(bp->dev,
+                                          "BUG: TX buffers exhausted mid-frame\n");
+
+                       desc->ctrl = ctrl | MACB_BIT(TX_USED);
+               }
+
+               macb_tx_unmap(bp, tx_skb);
+       }
+
+       /* Set end of TX queue */
+       desc = macb_tx_desc(queue, 0);
+       desc->addr = 0;
+       desc->ctrl = MACB_BIT(TX_USED);
+
+       /* Make descriptor updates visible to hardware */
+       wmb();
+
+       /* Reinitialize the TX desc queue */
+       queue_writel(queue, TBQP, queue->tx_ring_dma);
+       /* Make TX ring reflect state of hardware */
+       queue->tx_head = 0;
+       queue->tx_tail = 0;
+
+       /* Housework before enabling TX IRQ */
+       macb_writel(bp, TSR, macb_readl(bp, TSR));
+       queue_writel(queue, IER, MACB_TX_INT_FLAGS);
+
+       /* Now we are ready to start transmission again */
+       netif_tx_start_all_queues(bp->dev);
+       macb_writel(bp, NCR, macb_readl(bp, NCR) | MACB_BIT(TSTART));
+
+       spin_unlock_irqrestore(&bp->lock, flags);
+}
+
+static void macb_tx_interrupt(struct macb_queue *queue)
+{
+       unsigned int tail;
+       unsigned int head;
+       u32 status;
+       struct macb *bp = queue->bp;
+       u16 queue_index = queue - bp->queues;
+
+       status = macb_readl(bp, TSR);
+       macb_writel(bp, TSR, status);
+
+       if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
+               queue_writel(queue, ISR, MACB_BIT(TCOMP));
+
+       netdev_vdbg(bp->dev, "macb_tx_interrupt status = 0x%03lx\n",
+               (unsigned long)status);
+
+       head = queue->tx_head;
+       for (tail = queue->tx_tail; tail != head; tail++) {
+               struct macb_tx_skb      *tx_skb;
+               struct sk_buff          *skb;
+               struct macb_dma_desc    *desc;
+               u32                     ctrl;
+
+               desc = macb_tx_desc(queue, tail);
+
+               /* Make hw descriptor updates visible to CPU */
+               rmb();
+
+               ctrl = desc->ctrl;
+
+               /* TX_USED bit is only set by hardware on the very first buffer
+                * descriptor of the transmitted frame.
+                */
+               if (!(ctrl & MACB_BIT(TX_USED)))
+                       break;
+
+               /* Process all buffers of the current transmitted frame */
+               for (;; tail++) {
+                       tx_skb = macb_tx_skb(queue, tail);
+                       skb = tx_skb->skb;
+
+                       /* First, update TX stats if needed */
+                       if (skb) {
+                               netdev_vdbg(bp->dev, "skb %u (data %p) TX complete\n",
+                                           macb_tx_ring_wrap(tail), skb->data);
+                               bp->stats.tx_packets++;
+                               bp->stats.tx_bytes += skb->len;
+                       }
+
+                       /* Now we can safely release resources */
+                       macb_tx_unmap(bp, tx_skb);
+
+                       /* skb is set only for the last buffer of the frame.
+                        * WARNING: at this point skb has been freed by
+                        * macb_tx_unmap().
+                        */
+                       if (skb)
+                               break;
+               }
+       }
+
+       queue->tx_tail = tail;
+       if (__netif_subqueue_stopped(bp->dev, queue_index) &&
+           CIRC_CNT(queue->tx_head, queue->tx_tail,
+                    TX_RING_SIZE) <= MACB_TX_WAKEUP_THRESH)
+               netif_wake_subqueue(bp->dev, queue_index);
+}
+
+static void gem_rx_refill(struct macb *bp)
+{
+       unsigned int            entry;
+       struct sk_buff          *skb;
+       dma_addr_t              paddr;
+
+       while (CIRC_SPACE(bp->rx_prepared_head, bp->rx_tail, RX_RING_SIZE) > 0) {
+               entry = macb_rx_ring_wrap(bp->rx_prepared_head);
+
+               /* Make hw descriptor updates visible to CPU */
+               rmb();
+
+               bp->rx_prepared_head++;
+
+               if (bp->rx_skbuff[entry] == NULL) {
+                       /* allocate sk_buff for this free entry in ring */
+                       skb = netdev_alloc_skb(bp->dev, bp->rx_buffer_size);
+                       if (unlikely(skb == NULL)) {
+                               netdev_err(bp->dev,
+                                          "Unable to allocate sk_buff\n");
+                               break;
+                       }
+
+                       /* now fill corresponding descriptor entry */
+                       paddr = dma_map_single(&bp->pdev->dev, skb->data,
+                                              bp->rx_buffer_size, DMA_FROM_DEVICE);
+                       if (dma_mapping_error(&bp->pdev->dev, paddr)) {
+                               dev_kfree_skb(skb);
+                               break;
+                       }
+
+                       bp->rx_skbuff[entry] = skb;
+
+                       if (entry == RX_RING_SIZE - 1)
+                               paddr |= MACB_BIT(RX_WRAP);
+                       bp->rx_ring[entry].addr = paddr;
+                       bp->rx_ring[entry].ctrl = 0;
+
+                       /* properly align Ethernet header */
+                       skb_reserve(skb, NET_IP_ALIGN);
+               } else {
+                       bp->rx_ring[entry].addr &= ~MACB_BIT(RX_USED);
+                       bp->rx_ring[entry].ctrl = 0;
+               }
+       }
+
+       /* Make descriptor updates visible to hardware */
+       wmb();
+
+       netdev_vdbg(bp->dev, "rx ring: prepared head %d, tail %d\n",
+                  bp->rx_prepared_head, bp->rx_tail);
+}
+
+/* Mark DMA descriptors from begin up to and not including end as unused */
+static void discard_partial_frame(struct macb *bp, unsigned int begin,
+                                 unsigned int end)
+{
+       unsigned int frag;
+
+       for (frag = begin; frag != end; frag++) {
+               struct macb_dma_desc *desc = macb_rx_desc(bp, frag);
+               desc->addr &= ~MACB_BIT(RX_USED);
+       }
+
+       /* Make descriptor updates visible to hardware */
+       wmb();
+
+       /*
+        * When this happens, the hardware stats registers for
+        * whatever caused this is updated, so we don't have to record
+        * anything.
+        */
+}
+
+static int gem_rx(struct macb *bp, int budget)
+{
+       unsigned int            len;
+       unsigned int            entry;
+       struct sk_buff          *skb;
+       struct macb_dma_desc    *desc;
+       int                     count = 0;
+
+       while (count < budget) {
+               u32 addr, ctrl;
+
+               entry = macb_rx_ring_wrap(bp->rx_tail);
+               desc = &bp->rx_ring[entry];
+
+               /* Make hw descriptor updates visible to CPU */
+               rmb();
+
+               addr = desc->addr;
+               ctrl = desc->ctrl;
+
+               if (!(addr & MACB_BIT(RX_USED)))
+                       break;
+
+               bp->rx_tail++;
+               count++;
+
+               if (!(ctrl & MACB_BIT(RX_SOF) && ctrl & MACB_BIT(RX_EOF))) {
+                       netdev_err(bp->dev,
+                                  "not whole frame pointed by descriptor\n");
+                       bp->stats.rx_dropped++;
+                       break;
+               }
+               skb = bp->rx_skbuff[entry];
+               if (unlikely(!skb)) {
+                       netdev_err(bp->dev,
+                                  "inconsistent Rx descriptor chain\n");
+                       bp->stats.rx_dropped++;
+                       break;
+               }
+               /* now everything is ready for receiving packet */
+               bp->rx_skbuff[entry] = NULL;
+               len = MACB_BFEXT(RX_FRMLEN, ctrl);
+
+               netdev_vdbg(bp->dev, "gem_rx %u (len %u)\n", entry, len);
+
+               skb_put(skb, len);
+               addr = MACB_BF(RX_WADDR, MACB_BFEXT(RX_WADDR, addr));
+               dma_unmap_single(&bp->pdev->dev, addr,
+                                bp->rx_buffer_size, DMA_FROM_DEVICE);
+
+               skb->protocol = eth_type_trans(skb, bp->dev);
+               skb_checksum_none_assert(skb);
+               if (bp->dev->features & NETIF_F_RXCSUM &&
+                   !(bp->dev->flags & IFF_PROMISC) &&
+                   GEM_BFEXT(RX_CSUM, ctrl) & GEM_RX_CSUM_CHECKED_MASK)
+                       skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+               bp->stats.rx_packets++;
+               bp->stats.rx_bytes += skb->len;
+
+#if defined(DEBUG) && defined(VERBOSE_DEBUG)
+               netdev_vdbg(bp->dev, "received skb of length %u, csum: %08x\n",
+                           skb->len, skb->csum);
+               print_hex_dump(KERN_DEBUG, " mac: ", DUMP_PREFIX_ADDRESS, 16, 1,
+                              skb_mac_header(skb), 16, true);
+               print_hex_dump(KERN_DEBUG, "data: ", DUMP_PREFIX_ADDRESS, 16, 1,
+                              skb->data, 32, true);
+#endif
+
+               netif_receive_skb(skb);
+       }
+
+       gem_rx_refill(bp);
+
+       return count;
+}
+
+static int macb_rx_frame(struct macb *bp, unsigned int first_frag,
+                        unsigned int last_frag)
+{
+       unsigned int len;
+       unsigned int frag;
+       unsigned int offset;
+       struct sk_buff *skb;
+       struct macb_dma_desc *desc;
+
+       desc = macb_rx_desc(bp, last_frag);
+       len = MACB_BFEXT(RX_FRMLEN, desc->ctrl);
+
+       netdev_vdbg(bp->dev, "macb_rx_frame frags %u - %u (len %u)\n",
+               macb_rx_ring_wrap(first_frag),
+               macb_rx_ring_wrap(last_frag), len);
+
+       /*
+        * The ethernet header starts NET_IP_ALIGN bytes into the
+        * first buffer. Since the header is 14 bytes, this makes the
+        * payload word-aligned.
+        *
+        * Instead of calling skb_reserve(NET_IP_ALIGN), we just copy
+        * the two padding bytes into the skb so that we avoid hitting
+        * the slowpath in memcpy(), and pull them off afterwards.
+        */
+       skb = netdev_alloc_skb(bp->dev, len + NET_IP_ALIGN);
+       if (!skb) {
+               bp->stats.rx_dropped++;
+               for (frag = first_frag; ; frag++) {
+                       desc = macb_rx_desc(bp, frag);
+                       desc->addr &= ~MACB_BIT(RX_USED);
+                       if (frag == last_frag)
+                               break;
+               }
+
+               /* Make descriptor updates visible to hardware */
+               wmb();
+
+               return 1;
+       }
+
+       offset = 0;
+       len += NET_IP_ALIGN;
+       skb_checksum_none_assert(skb);
+       skb_put(skb, len);
+
+       for (frag = first_frag; ; frag++) {
+               unsigned int frag_len = bp->rx_buffer_size;
+
+               if (offset + frag_len > len) {
+                       BUG_ON(frag != last_frag);
+                       frag_len = len - offset;
+               }
+               skb_copy_to_linear_data_offset(skb, offset,
+                               macb_rx_buffer(bp, frag), frag_len);
+               offset += bp->rx_buffer_size;
+               desc = macb_rx_desc(bp, frag);
+               desc->addr &= ~MACB_BIT(RX_USED);
+
+               if (frag == last_frag)
+                       break;
+       }
+
+       /* Make descriptor updates visible to hardware */
+       wmb();
+
+       __skb_pull(skb, NET_IP_ALIGN);
+       skb->protocol = eth_type_trans(skb, bp->dev);
+
+       bp->stats.rx_packets++;
+       bp->stats.rx_bytes += skb->len;
+       netdev_vdbg(bp->dev, "received skb of length %u, csum: %08x\n",
+                  skb->len, skb->csum);
+       netif_receive_skb(skb);
+
+       return 0;
+}
+
+static int macb_rx(struct macb *bp, int budget)
+{
+       int received = 0;
+       unsigned int tail;
+       int first_frag = -1;
+
+       for (tail = bp->rx_tail; budget > 0; tail++) {
+               struct macb_dma_desc *desc = macb_rx_desc(bp, tail);
+               u32 addr, ctrl;
+
+               /* Make hw descriptor updates visible to CPU */
+               rmb();
+
+               addr = desc->addr;
+               ctrl = desc->ctrl;
+
+               if (!(addr & MACB_BIT(RX_USED)))
+                       break;
+
+               if (ctrl & MACB_BIT(RX_SOF)) {
+                       if (first_frag != -1)
+                               discard_partial_frame(bp, first_frag, tail);
+                       first_frag = tail;
+               }
+
+               if (ctrl & MACB_BIT(RX_EOF)) {
+                       int dropped;
+                       BUG_ON(first_frag == -1);
+
+                       dropped = macb_rx_frame(bp, first_frag, tail);
+                       first_frag = -1;
+                       if (!dropped) {
+                               received++;
+                               budget--;
+                       }
+               }
+       }
+
+       if (first_frag != -1)
+               bp->rx_tail = first_frag;
+       else
+               bp->rx_tail = tail;
+
+       return received;
+}
+
+static int macb_poll(struct napi_struct *napi, int budget)
+{
+       struct macb *bp = container_of(napi, struct macb, napi);
+       int work_done;
+       u32 status;
+
+       status = macb_readl(bp, RSR);
+       macb_writel(bp, RSR, status);
+
+       work_done = 0;
+
+       netdev_vdbg(bp->dev, "poll: status = %08lx, budget = %d\n",
+                  (unsigned long)status, budget);
+
+       work_done = bp->macbgem_ops.mog_rx(bp, budget);
+       if (work_done < budget) {
+               napi_complete(napi);
+
+               /* Packets received while interrupts were disabled */
+               status = macb_readl(bp, RSR);
+               if (status) {
+                       if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
+                               macb_writel(bp, ISR, MACB_BIT(RCOMP));
+                       napi_reschedule(napi);
+               } else {
+                       macb_writel(bp, IER, MACB_RX_INT_FLAGS);
+               }
+       }
+
+       /* TODO: Handle errors */
+
+       return work_done;
+}
+
+static irqreturn_t macb_interrupt(int irq, void *dev_id)
+{
+       struct macb_queue *queue = dev_id;
+       struct macb *bp = queue->bp;
+       struct net_device *dev = bp->dev;
+       u32 status, ctrl;
+
+       status = queue_readl(queue, ISR);
+
+       if (unlikely(!status))
+               return IRQ_NONE;
+
+       spin_lock(&bp->lock);
+
+       while (status) {
+               /* close possible race with dev_close */
+               if (unlikely(!netif_running(dev))) {
+                       queue_writel(queue, IDR, -1);
+                       break;
+               }
+
+               netdev_vdbg(bp->dev, "queue = %u, isr = 0x%08lx\n",
+                           (unsigned int)(queue - bp->queues),
+                           (unsigned long)status);
+
+               if (status & MACB_RX_INT_FLAGS) {
+                       /*
+                        * There's no point taking any more interrupts
+                        * until we have processed the buffers. The
+                        * scheduling call may fail if the poll routine
+                        * is already scheduled, so disable interrupts
+                        * now.
+                        */
+                       queue_writel(queue, IDR, MACB_RX_INT_FLAGS);
+                       if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
+                               queue_writel(queue, ISR, MACB_BIT(RCOMP));
+
+                       if (napi_schedule_prep(&bp->napi)) {
+                               netdev_vdbg(bp->dev, "scheduling RX softirq\n");
+                               __napi_schedule(&bp->napi);
+                       }
+               }
+
+               if (unlikely(status & (MACB_TX_ERR_FLAGS))) {
+                       queue_writel(queue, IDR, MACB_TX_INT_FLAGS);
+                       schedule_work(&queue->tx_error_task);
+
+                       if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
+                               queue_writel(queue, ISR, MACB_TX_ERR_FLAGS);
+
+                       break;
+               }
+
+               if (status & MACB_BIT(TCOMP))
+                       macb_tx_interrupt(queue);
+
+               /*
+                * Link change detection isn't possible with RMII, so we'll
+                * add that if/when we get our hands on a full-blown MII PHY.
+                */
+
+               /* There is a hardware issue under heavy load where DMA can
+                * stop, this causes endless "used buffer descriptor read"
+                * interrupts but it can be cleared by re-enabling RX. See
+                * the at91 manual, section 41.3.1 or the Zynq manual
+                * section 16.7.4 for details.
+                */
+               if (status & MACB_BIT(RXUBR)) {
+                       ctrl = macb_readl(bp, NCR);
+                       macb_writel(bp, NCR, ctrl & ~MACB_BIT(RE));
+                       macb_writel(bp, NCR, ctrl | MACB_BIT(RE));
+
+                       if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
+                               macb_writel(bp, ISR, MACB_BIT(RXUBR));
+               }
+
+               if (status & MACB_BIT(ISR_ROVR)) {
+                       /* We missed at least one packet */
+                       if (macb_is_gem(bp))
+                               bp->hw_stats.gem.rx_overruns++;
+                       else
+                               bp->hw_stats.macb.rx_overruns++;
+
+                       if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
+                               queue_writel(queue, ISR, MACB_BIT(ISR_ROVR));
+               }
+
+               if (status & MACB_BIT(HRESP)) {
+                       /*
+                        * TODO: Reset the hardware, and maybe move the
+                        * netdev_err to a lower-priority context as well
+                        * (work queue?)
+                        */
+                       netdev_err(dev, "DMA bus error: HRESP not OK\n");
+
+                       if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
+                               queue_writel(queue, ISR, MACB_BIT(HRESP));
+               }
+
+               status = queue_readl(queue, ISR);
+       }
+
+       spin_unlock(&bp->lock);
+
+       return IRQ_HANDLED;
+}
+
+#ifdef CONFIG_NET_POLL_CONTROLLER
+/*
+ * Polling receive - used by netconsole and other diagnostic tools
+ * to allow network i/o with interrupts disabled.
+ */
+static void macb_poll_controller(struct net_device *dev)
+{
+       struct macb *bp = netdev_priv(dev);
+       struct macb_queue *queue;
+       unsigned long flags;
+       unsigned int q;
+
+       local_irq_save(flags);
+       for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue)
+               macb_interrupt(dev->irq, queue);
+       local_irq_restore(flags);
+}
+#endif
+
+static inline unsigned int macb_count_tx_descriptors(struct macb *bp,
+                                                    unsigned int len)
+{
+       return (len + bp->max_tx_length - 1) / bp->max_tx_length;
+}
+
+static unsigned int macb_tx_map(struct macb *bp,
+                               struct macb_queue *queue,
+                               struct sk_buff *skb)
+{
+       dma_addr_t mapping;
+       unsigned int len, entry, i, tx_head = queue->tx_head;
+       struct macb_tx_skb *tx_skb = NULL;
+       struct macb_dma_desc *desc;
+       unsigned int offset, size, count = 0;
+       unsigned int f, nr_frags = skb_shinfo(skb)->nr_frags;
+       unsigned int eof = 1;
+       u32 ctrl;
+
+       /* First, map non-paged data */
+       len = skb_headlen(skb);
+       offset = 0;
+       while (len) {
+               size = min(len, bp->max_tx_length);
+               entry = macb_tx_ring_wrap(tx_head);
+               tx_skb = &queue->tx_skb[entry];
+
+               mapping = dma_map_single(&bp->pdev->dev,
+                                        skb->data + offset,
+                                        size, DMA_TO_DEVICE);
+               if (dma_mapping_error(&bp->pdev->dev, mapping))
+                       goto dma_error;
+
+               /* Save info to properly release resources */
+               tx_skb->skb = NULL;
+               tx_skb->mapping = mapping;
+               tx_skb->size = size;
+               tx_skb->mapped_as_page = false;
+
+               len -= size;
+               offset += size;
+               count++;
+               tx_head++;
+       }
+
+       /* Then, map paged data from fragments */
+       for (f = 0; f < nr_frags; f++) {
+               const skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
+
+               len = skb_frag_size(frag);
+               offset = 0;
+               while (len) {
+                       size = min(len, bp->max_tx_length);
+                       entry = macb_tx_ring_wrap(tx_head);
+                       tx_skb = &queue->tx_skb[entry];
+
+                       mapping = skb_frag_dma_map(&bp->pdev->dev, frag,
+                                                  offset, size, DMA_TO_DEVICE);
+                       if (dma_mapping_error(&bp->pdev->dev, mapping))
+                               goto dma_error;
+
+                       /* Save info to properly release resources */
+                       tx_skb->skb = NULL;
+                       tx_skb->mapping = mapping;
+                       tx_skb->size = size;
+                       tx_skb->mapped_as_page = true;
+
+                       len -= size;
+                       offset += size;
+                       count++;
+                       tx_head++;
+               }
+       }
+
+       /* Should never happen */
+       if (unlikely(tx_skb == NULL)) {
+               netdev_err(bp->dev, "BUG! empty skb!\n");
+               return 0;
+       }
+
+       /* This is the last buffer of the frame: save socket buffer */
+       tx_skb->skb = skb;
+
+       /* Update TX ring: update buffer descriptors in reverse order
+        * to avoid race condition
+        */
+
+       /* Set 'TX_USED' bit in buffer descriptor at tx_head position
+        * to set the end of TX queue
+        */
+       i = tx_head;
+       entry = macb_tx_ring_wrap(i);
+       ctrl = MACB_BIT(TX_USED);
+       desc = &queue->tx_ring[entry];
+       desc->ctrl = ctrl;
+
+       do {
+               i--;
+               entry = macb_tx_ring_wrap(i);
+               tx_skb = &queue->tx_skb[entry];
+               desc = &queue->tx_ring[entry];
+
+               ctrl = (u32)tx_skb->size;
+               if (eof) {
+                       ctrl |= MACB_BIT(TX_LAST);
+                       eof = 0;
+               }
+               if (unlikely(entry == (TX_RING_SIZE - 1)))
+                       ctrl |= MACB_BIT(TX_WRAP);
+
+               /* Set TX buffer descriptor */
+               desc->addr = tx_skb->mapping;
+               /* desc->addr must be visible to hardware before clearing
+                * 'TX_USED' bit in desc->ctrl.
+                */
+               wmb();
+               desc->ctrl = ctrl;
+       } while (i != queue->tx_head);
+
+       queue->tx_head = tx_head;
+
+       return count;
+
+dma_error:
+       netdev_err(bp->dev, "TX DMA map failed\n");
+
+       for (i = queue->tx_head; i != tx_head; i++) {
+               tx_skb = macb_tx_skb(queue, i);
+
+               macb_tx_unmap(bp, tx_skb);
+       }
+
+       return 0;
+}
+
+static int macb_start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       u16 queue_index = skb_get_queue_mapping(skb);
+       struct macb *bp = netdev_priv(dev);
+       struct macb_queue *queue = &bp->queues[queue_index];
+       unsigned long flags;
+       unsigned int count, nr_frags, frag_size, f;
+
+#if defined(DEBUG) && defined(VERBOSE_DEBUG)
+       netdev_vdbg(bp->dev,
+                  "start_xmit: queue %hu len %u head %p data %p tail %p end %p\n",
+                  queue_index, skb->len, skb->head, skb->data,
+                  skb_tail_pointer(skb), skb_end_pointer(skb));
+       print_hex_dump(KERN_DEBUG, "data: ", DUMP_PREFIX_OFFSET, 16, 1,
+                      skb->data, 16, true);
+#endif
+
+       /* Count how many TX buffer descriptors are needed to send this
+        * socket buffer: skb fragments of jumbo frames may need to be
+        * splitted into many buffer descriptors.
+        */
+       count = macb_count_tx_descriptors(bp, skb_headlen(skb));
+       nr_frags = skb_shinfo(skb)->nr_frags;
+       for (f = 0; f < nr_frags; f++) {
+               frag_size = skb_frag_size(&skb_shinfo(skb)->frags[f]);
+               count += macb_count_tx_descriptors(bp, frag_size);
+       }
+
+       spin_lock_irqsave(&bp->lock, flags);
+
+       /* This is a hard error, log it. */
+       if (CIRC_SPACE(queue->tx_head, queue->tx_tail, TX_RING_SIZE) < count) {
+               netif_stop_subqueue(dev, queue_index);
+               spin_unlock_irqrestore(&bp->lock, flags);
+               netdev_dbg(bp->dev, "tx_head = %u, tx_tail = %u\n",
+                          queue->tx_head, queue->tx_tail);
+               return NETDEV_TX_BUSY;
+       }
+
+       /* Map socket buffer for DMA transfer */
+       if (!macb_tx_map(bp, queue, skb)) {
+               dev_kfree_skb_any(skb);
+               goto unlock;
+       }
+
+       /* Make newly initialized descriptor visible to hardware */
+       wmb();
+
+       skb_tx_timestamp(skb);
+
+       macb_writel(bp, NCR, macb_readl(bp, NCR) | MACB_BIT(TSTART));
+
+       if (CIRC_SPACE(queue->tx_head, queue->tx_tail, TX_RING_SIZE) < 1)
+               netif_stop_subqueue(dev, queue_index);
+
+unlock:
+       spin_unlock_irqrestore(&bp->lock, flags);
+
+       return NETDEV_TX_OK;
+}
+
+static void macb_init_rx_buffer_size(struct macb *bp, size_t size)
+{
+       if (!macb_is_gem(bp)) {
+               bp->rx_buffer_size = MACB_RX_BUFFER_SIZE;
+       } else {
+               bp->rx_buffer_size = size;
+
+               if (bp->rx_buffer_size % RX_BUFFER_MULTIPLE) {
+                       netdev_dbg(bp->dev,
+                                   "RX buffer must be multiple of %d bytes, expanding\n",
+                                   RX_BUFFER_MULTIPLE);
+                       bp->rx_buffer_size =
+                               roundup(bp->rx_buffer_size, RX_BUFFER_MULTIPLE);
+               }
+       }
+
+       netdev_dbg(bp->dev, "mtu [%u] rx_buffer_size [%Zu]\n",
+                  bp->dev->mtu, bp->rx_buffer_size);
+}
+
+static void gem_free_rx_buffers(struct macb *bp)
+{
+       struct sk_buff          *skb;
+       struct macb_dma_desc    *desc;
+       dma_addr_t              addr;
+       int i;
+
+       if (!bp->rx_skbuff)
+               return;
+
+       for (i = 0; i < RX_RING_SIZE; i++) {
+               skb = bp->rx_skbuff[i];
+
+               if (skb == NULL)
+                       continue;
+
+               desc = &bp->rx_ring[i];
+               addr = MACB_BF(RX_WADDR, MACB_BFEXT(RX_WADDR, desc->addr));
+               dma_unmap_single(&bp->pdev->dev, addr, bp->rx_buffer_size,
+                                DMA_FROM_DEVICE);
+               dev_kfree_skb_any(skb);
+               skb = NULL;
+       }
+
+       kfree(bp->rx_skbuff);
+       bp->rx_skbuff = NULL;
+}
+
+static void macb_free_rx_buffers(struct macb *bp)
+{
+       if (bp->rx_buffers) {
+               dma_free_coherent(&bp->pdev->dev,
+                                 RX_RING_SIZE * bp->rx_buffer_size,
+                                 bp->rx_buffers, bp->rx_buffers_dma);
+               bp->rx_buffers = NULL;
+       }
+}
+
+static void macb_free_consistent(struct macb *bp)
+{
+       struct macb_queue *queue;
+       unsigned int q;
+
+       bp->macbgem_ops.mog_free_rx_buffers(bp);
+       if (bp->rx_ring) {
+               dma_free_coherent(&bp->pdev->dev, RX_RING_BYTES,
+                                 bp->rx_ring, bp->rx_ring_dma);
+               bp->rx_ring = NULL;
+       }
+
+       for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
+               kfree(queue->tx_skb);
+               queue->tx_skb = NULL;
+               if (queue->tx_ring) {
+                       dma_free_coherent(&bp->pdev->dev, TX_RING_BYTES,
+                                         queue->tx_ring, queue->tx_ring_dma);
+                       queue->tx_ring = NULL;
+               }
+       }
+}
+
+static int gem_alloc_rx_buffers(struct macb *bp)
+{
+       int size;
+
+       size = RX_RING_SIZE * sizeof(struct sk_buff *);
+       bp->rx_skbuff = kzalloc(size, GFP_KERNEL);
+       if (!bp->rx_skbuff)
+               return -ENOMEM;
+       else
+               netdev_dbg(bp->dev,
+                          "Allocated %d RX struct sk_buff entries at %p\n",
+                          RX_RING_SIZE, bp->rx_skbuff);
+       return 0;
+}
+
+static int macb_alloc_rx_buffers(struct macb *bp)
+{
+       int size;
+
+       size = RX_RING_SIZE * bp->rx_buffer_size;
+       bp->rx_buffers = dma_alloc_coherent(&bp->pdev->dev, size,
+                                           &bp->rx_buffers_dma, GFP_KERNEL);
+       if (!bp->rx_buffers)
+               return -ENOMEM;
+       else
+               netdev_dbg(bp->dev,
+                          "Allocated RX buffers of %d bytes at %08lx (mapped %p)\n",
+                          size, (unsigned long)bp->rx_buffers_dma, bp->rx_buffers);
+       return 0;
+}
+
+static int macb_alloc_consistent(struct macb *bp)
+{
+       struct macb_queue *queue;
+       unsigned int q;
+       int size;
+
+       for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
+               size = TX_RING_BYTES;
+               queue->tx_ring = dma_alloc_coherent(&bp->pdev->dev, size,
+                                                   &queue->tx_ring_dma,
+                                                   GFP_KERNEL);
+               if (!queue->tx_ring)
+                       goto out_err;
+               netdev_dbg(bp->dev,
+                          "Allocated TX ring for queue %u of %d bytes at %08lx (mapped %p)\n",
+                          q, size, (unsigned long)queue->tx_ring_dma,
+                          queue->tx_ring);
+
+               size = TX_RING_SIZE * sizeof(struct macb_tx_skb);
+               queue->tx_skb = kmalloc(size, GFP_KERNEL);
+               if (!queue->tx_skb)
+                       goto out_err;
+       }
+
+       size = RX_RING_BYTES;
+       bp->rx_ring = dma_alloc_coherent(&bp->pdev->dev, size,
+                                        &bp->rx_ring_dma, GFP_KERNEL);
+       if (!bp->rx_ring)
+               goto out_err;
+       netdev_dbg(bp->dev,
+                  "Allocated RX ring of %d bytes at %08lx (mapped %p)\n",
+                  size, (unsigned long)bp->rx_ring_dma, bp->rx_ring);
+
+       if (bp->macbgem_ops.mog_alloc_rx_buffers(bp))
+               goto out_err;
+
+       return 0;
+
+out_err:
+       macb_free_consistent(bp);
+       return -ENOMEM;
+}
+
+static void gem_init_rings(struct macb *bp)
+{
+       struct macb_queue *queue;
+       unsigned int q;
+       int i;
+
+       for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
+               for (i = 0; i < TX_RING_SIZE; i++) {
+                       queue->tx_ring[i].addr = 0;
+                       queue->tx_ring[i].ctrl = MACB_BIT(TX_USED);
+               }
+               queue->tx_ring[TX_RING_SIZE - 1].ctrl |= MACB_BIT(TX_WRAP);
+               queue->tx_head = 0;
+               queue->tx_tail = 0;
+       }
+
+       bp->rx_tail = 0;
+       bp->rx_prepared_head = 0;
+
+       gem_rx_refill(bp);
+}
+
+static void macb_init_rings(struct macb *bp)
+{
+       int i;
+       dma_addr_t addr;
+
+       addr = bp->rx_buffers_dma;
+       for (i = 0; i < RX_RING_SIZE; i++) {
+               bp->rx_ring[i].addr = addr;
+               bp->rx_ring[i].ctrl = 0;
+               addr += bp->rx_buffer_size;
+       }
+       bp->rx_ring[RX_RING_SIZE - 1].addr |= MACB_BIT(RX_WRAP);
+
+       for (i = 0; i < TX_RING_SIZE; i++) {
+               bp->queues[0].tx_ring[i].addr = 0;
+               bp->queues[0].tx_ring[i].ctrl = MACB_BIT(TX_USED);
+       }
+       bp->queues[0].tx_head = 0;
+       bp->queues[0].tx_tail = 0;
+       bp->queues[0].tx_ring[TX_RING_SIZE - 1].ctrl |= MACB_BIT(TX_WRAP);
+
+       bp->rx_tail = 0;
+}
+
+static void macb_reset_hw(struct macb *bp)
+{
+       struct macb_queue *queue;
+       unsigned int q;
+
+       /*
+        * Disable RX and TX (XXX: Should we halt the transmission
+        * more gracefully?)
+        */
+       macb_writel(bp, NCR, 0);
+
+       /* Clear the stats registers (XXX: Update stats first?) */
+       macb_writel(bp, NCR, MACB_BIT(CLRSTAT));
+
+       /* Clear all status flags */
+       macb_writel(bp, TSR, -1);
+       macb_writel(bp, RSR, -1);
+
+       /* Disable all interrupts */
+       for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
+               queue_writel(queue, IDR, -1);
+               queue_readl(queue, ISR);
+       }
+}
+
+static u32 gem_mdc_clk_div(struct macb *bp)
+{
+       u32 config;
+       unsigned long pclk_hz = clk_get_rate(bp->pclk);
+
+       if (pclk_hz <= 20000000)
+               config = GEM_BF(CLK, GEM_CLK_DIV8);
+       else if (pclk_hz <= 40000000)
+               config = GEM_BF(CLK, GEM_CLK_DIV16);
+       else if (pclk_hz <= 80000000)
+               config = GEM_BF(CLK, GEM_CLK_DIV32);
+       else if (pclk_hz <= 120000000)
+               config = GEM_BF(CLK, GEM_CLK_DIV48);
+       else if (pclk_hz <= 160000000)
+               config = GEM_BF(CLK, GEM_CLK_DIV64);
+       else
+               config = GEM_BF(CLK, GEM_CLK_DIV96);
+
+       return config;
+}
+
+static u32 macb_mdc_clk_div(struct macb *bp)
+{
+       u32 config;
+       unsigned long pclk_hz;
+
+       if (macb_is_gem(bp))
+               return gem_mdc_clk_div(bp);
+
+       pclk_hz = clk_get_rate(bp->pclk);
+       if (pclk_hz <= 20000000)
+               config = MACB_BF(CLK, MACB_CLK_DIV8);
+       else if (pclk_hz <= 40000000)
+               config = MACB_BF(CLK, MACB_CLK_DIV16);
+       else if (pclk_hz <= 80000000)
+               config = MACB_BF(CLK, MACB_CLK_DIV32);
+       else
+               config = MACB_BF(CLK, MACB_CLK_DIV64);
+
+       return config;
+}
+
+/*
+ * Get the DMA bus width field of the network configuration register that we
+ * should program.  We find the width from decoding the design configuration
+ * register to find the maximum supported data bus width.
+ */
+static u32 macb_dbw(struct macb *bp)
+{
+       if (!macb_is_gem(bp))
+               return 0;
+
+       switch (GEM_BFEXT(DBWDEF, gem_readl(bp, DCFG1))) {
+       case 4:
+               return GEM_BF(DBW, GEM_DBW128);
+       case 2:
+               return GEM_BF(DBW, GEM_DBW64);
+       case 1:
+       default:
+               return GEM_BF(DBW, GEM_DBW32);
+       }
+}
+
+/*
+ * Configure the receive DMA engine
+ * - use the correct receive buffer size
+ * - set best burst length for DMA operations
+ *   (if not supported by FIFO, it will fallback to default)
+ * - set both rx/tx packet buffers to full memory size
+ * These are configurable parameters for GEM.
+ */
+static void macb_configure_dma(struct macb *bp)
+{
+       u32 dmacfg;
+       u32 tmp, ncr;
+
+       if (macb_is_gem(bp)) {
+               dmacfg = gem_readl(bp, DMACFG) & ~GEM_BF(RXBS, -1L);
+               dmacfg |= GEM_BF(RXBS, bp->rx_buffer_size / RX_BUFFER_MULTIPLE);
+               if (bp->dma_burst_length)
+                       dmacfg = GEM_BFINS(FBLDO, bp->dma_burst_length, dmacfg);
+               dmacfg |= GEM_BIT(TXPBMS) | GEM_BF(RXBMS, -1L);
+               dmacfg &= ~GEM_BIT(ENDIA_PKT);
+
+               /* Find the CPU endianness by using the loopback bit of net_ctrl
+                * register. save it first. When the CPU is in big endian we
+                * need to program swaped mode for management descriptor access.
+                */
+               ncr = macb_readl(bp, NCR);
+               __raw_writel(MACB_BIT(LLB), bp->regs + MACB_NCR);
+               tmp =  __raw_readl(bp->regs + MACB_NCR);
+
+               if (tmp == MACB_BIT(LLB))
+                       dmacfg &= ~GEM_BIT(ENDIA_DESC);
+               else
+                       dmacfg |= GEM_BIT(ENDIA_DESC); /* CPU in big endian */
+
+               /* Restore net_ctrl */
+               macb_writel(bp, NCR, ncr);
+
+               if (bp->dev->features & NETIF_F_HW_CSUM)
+                       dmacfg |= GEM_BIT(TXCOEN);
+               else
+                       dmacfg &= ~GEM_BIT(TXCOEN);
+               netdev_dbg(bp->dev, "Cadence configure DMA with 0x%08x\n",
+                          dmacfg);
+               gem_writel(bp, DMACFG, dmacfg);
+       }
+}
+
+static void macb_init_hw(struct macb *bp)
+{
+       struct macb_queue *queue;
+       unsigned int q;
+
+       u32 config;
+
+       macb_reset_hw(bp);
+       macb_set_hwaddr(bp);
+
+       config = macb_mdc_clk_div(bp);
+       config |= MACB_BF(RBOF, NET_IP_ALIGN);  /* Make eth data aligned */
+       config |= MACB_BIT(PAE);                /* PAuse Enable */
+       config |= MACB_BIT(DRFCS);              /* Discard Rx FCS */
+       config |= MACB_BIT(BIG);                /* Receive oversized frames */
+       if (bp->dev->flags & IFF_PROMISC)
+               config |= MACB_BIT(CAF);        /* Copy All Frames */
+       else if (macb_is_gem(bp) && bp->dev->features & NETIF_F_RXCSUM)
+               config |= GEM_BIT(RXCOEN);
+       if (!(bp->dev->flags & IFF_BROADCAST))
+               config |= MACB_BIT(NBC);        /* No BroadCast */
+       config |= macb_dbw(bp);
+       macb_writel(bp, NCFGR, config);
+       bp->speed = SPEED_10;
+       bp->duplex = DUPLEX_HALF;
+
+       macb_configure_dma(bp);
+
+       /* Initialize TX and RX buffers */
+       macb_writel(bp, RBQP, bp->rx_ring_dma);
+       for (q = 0, queue = bp->queues; q < bp->num_queues; ++q, ++queue) {
+               queue_writel(queue, TBQP, queue->tx_ring_dma);
+
+               /* Enable interrupts */
+               queue_writel(queue, IER,
+                            MACB_RX_INT_FLAGS |
+                            MACB_TX_INT_FLAGS |
+                            MACB_BIT(HRESP));
+       }
+
+       /* Enable TX and RX */
+       macb_writel(bp, NCR, MACB_BIT(RE) | MACB_BIT(TE) | MACB_BIT(MPE));
+}
+
+/*
+ * The hash address register is 64 bits long and takes up two
+ * locations in the memory map.  The least significant bits are stored
+ * in EMAC_HSL and the most significant bits in EMAC_HSH.
+ *
+ * The unicast hash enable and the multicast hash enable bits in the
+ * network configuration register enable the reception of hash matched
+ * frames. The destination address is reduced to a 6 bit index into
+ * the 64 bit hash register using the following hash function.  The
+ * hash function is an exclusive or of every sixth bit of the
+ * destination address.
+ *
+ * hi[5] = da[5] ^ da[11] ^ da[17] ^ da[23] ^ da[29] ^ da[35] ^ da[41] ^ da[47]
+ * hi[4] = da[4] ^ da[10] ^ da[16] ^ da[22] ^ da[28] ^ da[34] ^ da[40] ^ da[46]
+ * hi[3] = da[3] ^ da[09] ^ da[15] ^ da[21] ^ da[27] ^ da[33] ^ da[39] ^ da[45]
+ * hi[2] = da[2] ^ da[08] ^ da[14] ^ da[20] ^ da[26] ^ da[32] ^ da[38] ^ da[44]
+ * hi[1] = da[1] ^ da[07] ^ da[13] ^ da[19] ^ da[25] ^ da[31] ^ da[37] ^ da[43]
+ * hi[0] = da[0] ^ da[06] ^ da[12] ^ da[18] ^ da[24] ^ da[30] ^ da[36] ^ da[42]
+ *
+ * da[0] represents the least significant bit of the first byte
+ * received, that is, the multicast/unicast indicator, and da[47]
+ * represents the most significant bit of the last byte received.  If
+ * the hash index, hi[n], points to a bit that is set in the hash
+ * register then the frame will be matched according to whether the
+ * frame is multicast or unicast.  A multicast match will be signalled
+ * if the multicast hash enable bit is set, da[0] is 1 and the hash
+ * index points to a bit set in the hash register.  A unicast match
+ * will be signalled if the unicast hash enable bit is set, da[0] is 0
+ * and the hash index points to a bit set in the hash register.  To
+ * receive all multicast frames, the hash register should be set with
+ * all ones and the multicast hash enable bit should be set in the
+ * network configuration register.
+ */
+
+static inline int hash_bit_value(int bitnr, __u8 *addr)
+{
+       if (addr[bitnr / 8] & (1 << (bitnr % 8)))
+               return 1;
+       return 0;
+}
+
+/*
+ * Return the hash index value for the specified address.
+ */
+static int hash_get_index(__u8 *addr)
+{
+       int i, j, bitval;
+       int hash_index = 0;
+
+       for (j = 0; j < 6; j++) {
+               for (i = 0, bitval = 0; i < 8; i++)
+                       bitval ^= hash_bit_value(i * 6 + j, addr);
+
+               hash_index |= (bitval << j);
+       }
+
+       return hash_index;
+}
+
+/*
+ * Add multicast addresses to the internal multicast-hash table.
+ */
+static void macb_sethashtable(struct net_device *dev)
+{
+       struct netdev_hw_addr *ha;
+       unsigned long mc_filter[2];
+       unsigned int bitnr;
+       struct macb *bp = netdev_priv(dev);
+
+       mc_filter[0] = mc_filter[1] = 0;
+
+       netdev_for_each_mc_addr(ha, dev) {
+               bitnr = hash_get_index(ha->addr);
+               mc_filter[bitnr >> 5] |= 1 << (bitnr & 31);
+       }
+
+       macb_or_gem_writel(bp, HRB, mc_filter[0]);
+       macb_or_gem_writel(bp, HRT, mc_filter[1]);
+}
+
+/*
+ * Enable/Disable promiscuous and multicast modes.
+ */
+static void macb_set_rx_mode(struct net_device *dev)
+{
+       unsigned long cfg;
+       struct macb *bp = netdev_priv(dev);
+
+       cfg = macb_readl(bp, NCFGR);
+
+       if (dev->flags & IFF_PROMISC) {
+               /* Enable promiscuous mode */
+               cfg |= MACB_BIT(CAF);
+
+               /* Disable RX checksum offload */
+               if (macb_is_gem(bp))
+                       cfg &= ~GEM_BIT(RXCOEN);
+       } else {
+               /* Disable promiscuous mode */
+               cfg &= ~MACB_BIT(CAF);
+
+               /* Enable RX checksum offload only if requested */
+               if (macb_is_gem(bp) && dev->features & NETIF_F_RXCSUM)
+                       cfg |= GEM_BIT(RXCOEN);
+       }
+
+       if (dev->flags & IFF_ALLMULTI) {
+               /* Enable all multicast mode */
+               macb_or_gem_writel(bp, HRB, -1);
+               macb_or_gem_writel(bp, HRT, -1);
+               cfg |= MACB_BIT(NCFGR_MTI);
+       } else if (!netdev_mc_empty(dev)) {
+               /* Enable specific multicasts */
+               macb_sethashtable(dev);
+               cfg |= MACB_BIT(NCFGR_MTI);
+       } else if (dev->flags & (~IFF_ALLMULTI)) {
+               /* Disable all multicast mode */
+               macb_or_gem_writel(bp, HRB, 0);
+               macb_or_gem_writel(bp, HRT, 0);
+               cfg &= ~MACB_BIT(NCFGR_MTI);
+       }
+
+       macb_writel(bp, NCFGR, cfg);
+}
+
+static int macb_open(struct net_device *dev)
+{
+       struct macb *bp = netdev_priv(dev);
+       size_t bufsz = dev->mtu + ETH_HLEN + ETH_FCS_LEN + NET_IP_ALIGN;
+       int err;
+
+       netdev_dbg(bp->dev, "open\n");
+
+       /* carrier starts down */
+       netif_carrier_off(dev);
+
+       /* if the phy is not yet register, retry later*/
+       if (!bp->phy_dev)
+               return -EAGAIN;
+
+       /* RX buffers initialization */
+       macb_init_rx_buffer_size(bp, bufsz);
+
+       err = macb_alloc_consistent(bp);
+       if (err) {
+               netdev_err(dev, "Unable to allocate DMA memory (error %d)\n",
+                          err);
+               return err;
+       }
+
+       napi_enable(&bp->napi);
+
+       bp->macbgem_ops.mog_init_rings(bp);
+       macb_init_hw(bp);
+
+       /* schedule a link state check */
+       phy_start(bp->phy_dev);
+
+       netif_tx_start_all_queues(dev);
+
+       return 0;
+}
+
+static int macb_close(struct net_device *dev)
+{
+       struct macb *bp = netdev_priv(dev);
+       unsigned long flags;
+
+       netif_tx_stop_all_queues(dev);
+       napi_disable(&bp->napi);
+
+       if (bp->phy_dev)
+               phy_stop(bp->phy_dev);
+
+       spin_lock_irqsave(&bp->lock, flags);
+       macb_reset_hw(bp);
+       netif_carrier_off(dev);
+       spin_unlock_irqrestore(&bp->lock, flags);
+
+       macb_free_consistent(bp);
+
+       return 0;
+}
+
+static void gem_update_stats(struct macb *bp)
+{
+       int i;
+       u32 *p = &bp->hw_stats.gem.tx_octets_31_0;
+
+       for (i = 0; i < GEM_STATS_LEN; ++i, ++p) {
+               u32 offset = gem_statistics[i].offset;
+               u64 val = readl_relaxed(bp->regs + offset);
+
+               bp->ethtool_stats[i] += val;
+               *p += val;
+
+               if (offset == GEM_OCTTXL || offset == GEM_OCTRXL) {
+                       /* Add GEM_OCTTXH, GEM_OCTRXH */
+                       val = readl_relaxed(bp->regs + offset + 4);
+                       bp->ethtool_stats[i] += ((u64)val) << 32;
+                       *(++p) += val;
+               }
+       }
+}
+
+static struct net_device_stats *gem_get_stats(struct macb *bp)
+{
+       struct gem_stats *hwstat = &bp->hw_stats.gem;
+       struct net_device_stats *nstat = &bp->stats;
+
+       gem_update_stats(bp);
+
+       nstat->rx_errors = (hwstat->rx_frame_check_sequence_errors +
+                           hwstat->rx_alignment_errors +
+                           hwstat->rx_resource_errors +
+                           hwstat->rx_overruns +
+                           hwstat->rx_oversize_frames +
+                           hwstat->rx_jabbers +
+                           hwstat->rx_undersized_frames +
+                           hwstat->rx_length_field_frame_errors);
+       nstat->tx_errors = (hwstat->tx_late_collisions +
+                           hwstat->tx_excessive_collisions +
+                           hwstat->tx_underrun +
+                           hwstat->tx_carrier_sense_errors);
+       nstat->multicast = hwstat->rx_multicast_frames;
+       nstat->collisions = (hwstat->tx_single_collision_frames +
+                            hwstat->tx_multiple_collision_frames +
+                            hwstat->tx_excessive_collisions);
+       nstat->rx_length_errors = (hwstat->rx_oversize_frames +
+                                  hwstat->rx_jabbers +
+                                  hwstat->rx_undersized_frames +
+                                  hwstat->rx_length_field_frame_errors);
+       nstat->rx_over_errors = hwstat->rx_resource_errors;
+       nstat->rx_crc_errors = hwstat->rx_frame_check_sequence_errors;
+       nstat->rx_frame_errors = hwstat->rx_alignment_errors;
+       nstat->rx_fifo_errors = hwstat->rx_overruns;
+       nstat->tx_aborted_errors = hwstat->tx_excessive_collisions;
+       nstat->tx_carrier_errors = hwstat->tx_carrier_sense_errors;
+       nstat->tx_fifo_errors = hwstat->tx_underrun;
+
+       return nstat;
+}
+
+static void gem_get_ethtool_stats(struct net_device *dev,
+                                 struct ethtool_stats *stats, u64 *data)
+{
+       struct macb *bp;
+
+       bp = netdev_priv(dev);
+       gem_update_stats(bp);
+       memcpy(data, &bp->ethtool_stats, sizeof(u64) * GEM_STATS_LEN);
+}
+
+static int gem_get_sset_count(struct net_device *dev, int sset)
+{
+       switch (sset) {
+       case ETH_SS_STATS:
+               return GEM_STATS_LEN;
+       default:
+               return -EOPNOTSUPP;
+       }
+}
+
+static void gem_get_ethtool_strings(struct net_device *dev, u32 sset, u8 *p)
+{
+       int i;
+
+       switch (sset) {
+       case ETH_SS_STATS:
+               for (i = 0; i < GEM_STATS_LEN; i++, p += ETH_GSTRING_LEN)
+                       memcpy(p, gem_statistics[i].stat_string,
+                              ETH_GSTRING_LEN);
+               break;
+       }
+}
+
+static struct net_device_stats *macb_get_stats(struct net_device *dev)
+{
+       struct macb *bp = netdev_priv(dev);
+       struct net_device_stats *nstat = &bp->stats;
+       struct macb_stats *hwstat = &bp->hw_stats.macb;
+
+       if (macb_is_gem(bp))
+               return gem_get_stats(bp);
+
+       /* read stats from hardware */
+       macb_update_stats(bp);
+
+       /* Convert HW stats into netdevice stats */
+       nstat->rx_errors = (hwstat->rx_fcs_errors +
+                           hwstat->rx_align_errors +
+                           hwstat->rx_resource_errors +
+                           hwstat->rx_overruns +
+                           hwstat->rx_oversize_pkts +
+                           hwstat->rx_jabbers +
+                           hwstat->rx_undersize_pkts +
+                           hwstat->rx_length_mismatch);
+       nstat->tx_errors = (hwstat->tx_late_cols +
+                           hwstat->tx_excessive_cols +
+                           hwstat->tx_underruns +
+                           hwstat->tx_carrier_errors +
+                           hwstat->sqe_test_errors);
+       nstat->collisions = (hwstat->tx_single_cols +
+                            hwstat->tx_multiple_cols +
+                            hwstat->tx_excessive_cols);
+       nstat->rx_length_errors = (hwstat->rx_oversize_pkts +
+                                  hwstat->rx_jabbers +
+                                  hwstat->rx_undersize_pkts +
+                                  hwstat->rx_length_mismatch);
+       nstat->rx_over_errors = hwstat->rx_resource_errors +
+                                  hwstat->rx_overruns;
+       nstat->rx_crc_errors = hwstat->rx_fcs_errors;
+       nstat->rx_frame_errors = hwstat->rx_align_errors;
+       nstat->rx_fifo_errors = hwstat->rx_overruns;
+       /* XXX: What does "missed" mean? */
+       nstat->tx_aborted_errors = hwstat->tx_excessive_cols;
+       nstat->tx_carrier_errors = hwstat->tx_carrier_errors;
+       nstat->tx_fifo_errors = hwstat->tx_underruns;
+       /* Don't know about heartbeat or window errors... */
+
+       return nstat;
+}
+
+static int macb_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct macb *bp = netdev_priv(dev);
+       struct phy_device *phydev = bp->phy_dev;
+
+       if (!phydev)
+               return -ENODEV;
+
+       return phy_ethtool_gset(phydev, cmd);
+}
+
+static int macb_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct macb *bp = netdev_priv(dev);
+       struct phy_device *phydev = bp->phy_dev;
+
+       if (!phydev)
+               return -ENODEV;
+
+       return phy_ethtool_sset(phydev, cmd);
+}
+
+static int macb_get_regs_len(struct net_device *netdev)
+{
+       return MACB_GREGS_NBR * sizeof(u32);
+}
+
+static void macb_get_regs(struct net_device *dev, struct ethtool_regs *regs,
+                         void *p)
+{
+       struct macb *bp = netdev_priv(dev);
+       unsigned int tail, head;
+       u32 *regs_buff = p;
+
+       regs->version = (macb_readl(bp, MID) & ((1 << MACB_REV_SIZE) - 1))
+                       | MACB_GREGS_VERSION;
+
+       tail = macb_tx_ring_wrap(bp->queues[0].tx_tail);
+       head = macb_tx_ring_wrap(bp->queues[0].tx_head);
+
+       regs_buff[0]  = macb_readl(bp, NCR);
+       regs_buff[1]  = macb_or_gem_readl(bp, NCFGR);
+       regs_buff[2]  = macb_readl(bp, NSR);
+       regs_buff[3]  = macb_readl(bp, TSR);
+       regs_buff[4]  = macb_readl(bp, RBQP);
+       regs_buff[5]  = macb_readl(bp, TBQP);
+       regs_buff[6]  = macb_readl(bp, RSR);
+       regs_buff[7]  = macb_readl(bp, IMR);
+
+       regs_buff[8]  = tail;
+       regs_buff[9]  = head;
+       regs_buff[10] = macb_tx_dma(&bp->queues[0], tail);
+       regs_buff[11] = macb_tx_dma(&bp->queues[0], head);
+
+       regs_buff[12] = macb_or_gem_readl(bp, USRIO);
+       if (macb_is_gem(bp)) {
+               regs_buff[13] = gem_readl(bp, DMACFG);
+       }
+}
+
+static const struct ethtool_ops macb_ethtool_ops = {
+       .get_settings           = macb_get_settings,
+       .set_settings           = macb_set_settings,
+       .get_regs_len           = macb_get_regs_len,
+       .get_regs               = macb_get_regs,
+       .get_link               = ethtool_op_get_link,
+       .get_ts_info            = ethtool_op_get_ts_info,
+};
+
+static const struct ethtool_ops gem_ethtool_ops = {
+       .get_settings           = macb_get_settings,
+       .set_settings           = macb_set_settings,
+       .get_regs_len           = macb_get_regs_len,
+       .get_regs               = macb_get_regs,
+       .get_link               = ethtool_op_get_link,
+       .get_ts_info            = ethtool_op_get_ts_info,
+       .get_ethtool_stats      = gem_get_ethtool_stats,
+       .get_strings            = gem_get_ethtool_strings,
+       .get_sset_count         = gem_get_sset_count,
+};
+
+static int macb_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
+{
+       struct macb *bp = netdev_priv(dev);
+       struct phy_device *phydev = bp->phy_dev;
+
+       if (!netif_running(dev))
+               return -EINVAL;
+
+       if (!phydev)
+               return -ENODEV;
+
+       return phy_mii_ioctl(phydev, rq, cmd);
+}
+
+static int macb_set_features(struct net_device *netdev,
+                            netdev_features_t features)
+{
+       struct macb *bp = netdev_priv(netdev);
+       netdev_features_t changed = features ^ netdev->features;
+
+       /* TX checksum offload */
+       if ((changed & NETIF_F_HW_CSUM) && macb_is_gem(bp)) {
+               u32 dmacfg;
+
+               dmacfg = gem_readl(bp, DMACFG);
+               if (features & NETIF_F_HW_CSUM)
+                       dmacfg |= GEM_BIT(TXCOEN);
+               else
+                       dmacfg &= ~GEM_BIT(TXCOEN);
+               gem_writel(bp, DMACFG, dmacfg);
+       }
+
+       /* RX checksum offload */
+       if ((changed & NETIF_F_RXCSUM) && macb_is_gem(bp)) {
+               u32 netcfg;
+
+               netcfg = gem_readl(bp, NCFGR);
+               if (features & NETIF_F_RXCSUM &&
+                   !(netdev->flags & IFF_PROMISC))
+                       netcfg |= GEM_BIT(RXCOEN);
+               else
+                       netcfg &= ~GEM_BIT(RXCOEN);
+               gem_writel(bp, NCFGR, netcfg);
+       }
+
+       return 0;
+}
+
+static const struct net_device_ops macb_netdev_ops = {
+       .ndo_open               = macb_open,
+       .ndo_stop               = macb_close,
+       .ndo_start_xmit         = macb_start_xmit,
+       .ndo_set_rx_mode        = macb_set_rx_mode,
+       .ndo_get_stats          = macb_get_stats,
+       .ndo_do_ioctl           = macb_ioctl,
+       .ndo_validate_addr      = eth_validate_addr,
+       .ndo_change_mtu         = eth_change_mtu,
+       .ndo_set_mac_address    = eth_mac_addr,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = macb_poll_controller,
+#endif
+       .ndo_set_features       = macb_set_features,
+};
+
+/*
+ * Configure peripheral capabilities according to device tree
+ * and integration options used
+ */
+static void macb_configure_caps(struct macb *bp, const struct macb_config *dt_conf)
+{
+       u32 dcfg;
+
+       if (dt_conf)
+               bp->caps = dt_conf->caps;
+
+       if (macb_is_gem_hw(bp->regs)) {
+               bp->caps |= MACB_CAPS_MACB_IS_GEM;
+
+               dcfg = gem_readl(bp, DCFG1);
+               if (GEM_BFEXT(IRQCOR, dcfg) == 0)
+                       bp->caps |= MACB_CAPS_ISR_CLEAR_ON_WRITE;
+               dcfg = gem_readl(bp, DCFG2);
+               if ((dcfg & (GEM_BIT(RX_PKT_BUFF) | GEM_BIT(TX_PKT_BUFF))) == 0)
+                       bp->caps |= MACB_CAPS_FIFO_MODE;
+       }
+
+       netdev_dbg(bp->dev, "Cadence caps 0x%08x\n", bp->caps);
+}
+
+static void macb_probe_queues(void __iomem *mem,
+                             unsigned int *queue_mask,
+                             unsigned int *num_queues)
+{
+       unsigned int hw_q;
+
+       *queue_mask = 0x1;
+       *num_queues = 1;
+
+       /* is it macb or gem ?
+        *
+        * We need to read directly from the hardware here because
+        * we are early in the probe process and don't have the
+        * MACB_CAPS_MACB_IS_GEM flag positioned
+        */
+       if (!macb_is_gem_hw(mem))
+               return;
+
+       /* bit 0 is never set but queue 0 always exists */
+       *queue_mask = readl_relaxed(mem + GEM_DCFG6) & 0xff;
+
+       *queue_mask |= 0x1;
+
+       for (hw_q = 1; hw_q < MACB_MAX_QUEUES; ++hw_q)
+               if (*queue_mask & (1 << hw_q))
+                       (*num_queues)++;
+}
+
+static int macb_clk_init(struct platform_device *pdev, struct clk **pclk,
+                        struct clk **hclk, struct clk **tx_clk)
+{
+       int err;
+
+       *pclk = devm_clk_get(&pdev->dev, "pclk");
+       if (IS_ERR(*pclk)) {
+               err = PTR_ERR(*pclk);
+               dev_err(&pdev->dev, "failed to get macb_clk (%u)\n", err);
+               return err;
+       }
+
+       *hclk = devm_clk_get(&pdev->dev, "hclk");
+       if (IS_ERR(*hclk)) {
+               err = PTR_ERR(*hclk);
+               dev_err(&pdev->dev, "failed to get hclk (%u)\n", err);
+               return err;
+       }
+
+       *tx_clk = devm_clk_get(&pdev->dev, "tx_clk");
+       if (IS_ERR(*tx_clk))
+               *tx_clk = NULL;
+
+       err = clk_prepare_enable(*pclk);
+       if (err) {
+               dev_err(&pdev->dev, "failed to enable pclk (%u)\n", err);
+               return err;
+       }
+
+       err = clk_prepare_enable(*hclk);
+       if (err) {
+               dev_err(&pdev->dev, "failed to enable hclk (%u)\n", err);
+               goto err_disable_pclk;
+       }
+
+       err = clk_prepare_enable(*tx_clk);
+       if (err) {
+               dev_err(&pdev->dev, "failed to enable tx_clk (%u)\n", err);
+               goto err_disable_hclk;
+       }
+
+       return 0;
+
+err_disable_hclk:
+       clk_disable_unprepare(*hclk);
+
+err_disable_pclk:
+       clk_disable_unprepare(*pclk);
+
+       return err;
+}
+
+static int macb_init(struct platform_device *pdev)
+{
+       struct net_device *dev = platform_get_drvdata(pdev);
+       unsigned int hw_q, q;
+       struct macb *bp = netdev_priv(dev);
+       struct macb_queue *queue;
+       int err;
+       u32 val;
+
+       /* set the queue register mapping once for all: queue0 has a special
+        * register mapping but we don't want to test the queue index then
+        * compute the corresponding register offset at run time.
+        */
+       for (hw_q = 0, q = 0; hw_q < MACB_MAX_QUEUES; ++hw_q) {
+               if (!(bp->queue_mask & (1 << hw_q)))
+                       continue;
+
+               queue = &bp->queues[q];
+               queue->bp = bp;
+               if (hw_q) {
+                       queue->ISR  = GEM_ISR(hw_q - 1);
+                       queue->IER  = GEM_IER(hw_q - 1);
+                       queue->IDR  = GEM_IDR(hw_q - 1);
+                       queue->IMR  = GEM_IMR(hw_q - 1);
+                       queue->TBQP = GEM_TBQP(hw_q - 1);
+               } else {
+                       /* queue0 uses legacy registers */
+                       queue->ISR  = MACB_ISR;
+                       queue->IER  = MACB_IER;
+                       queue->IDR  = MACB_IDR;
+                       queue->IMR  = MACB_IMR;
+                       queue->TBQP = MACB_TBQP;
+               }
+
+               /* get irq: here we use the linux queue index, not the hardware
+                * queue index. the queue irq definitions in the device tree
+                * must remove the optional gaps that could exist in the
+                * hardware queue mask.
+                */
+               queue->irq = platform_get_irq(pdev, q);
+               err = devm_request_irq(&pdev->dev, queue->irq, macb_interrupt,
+                                      IRQF_SHARED, dev->name, queue);
+               if (err) {
+                       dev_err(&pdev->dev,
+                               "Unable to request IRQ %d (error %d)\n",
+                               queue->irq, err);
+                       return err;
+               }
+
+               INIT_WORK(&queue->tx_error_task, macb_tx_error_task);
+               q++;
+       }
+
+       dev->netdev_ops = &macb_netdev_ops;
+       netif_napi_add(dev, &bp->napi, macb_poll, 64);
+
+       /* setup appropriated routines according to adapter type */
+       if (macb_is_gem(bp)) {
+               bp->max_tx_length = GEM_MAX_TX_LEN;
+               bp->macbgem_ops.mog_alloc_rx_buffers = gem_alloc_rx_buffers;
+               bp->macbgem_ops.mog_free_rx_buffers = gem_free_rx_buffers;
+               bp->macbgem_ops.mog_init_rings = gem_init_rings;
+               bp->macbgem_ops.mog_rx = gem_rx;
+               dev->ethtool_ops = &gem_ethtool_ops;
+       } else {
+               bp->max_tx_length = MACB_MAX_TX_LEN;
+               bp->macbgem_ops.mog_alloc_rx_buffers = macb_alloc_rx_buffers;
+               bp->macbgem_ops.mog_free_rx_buffers = macb_free_rx_buffers;
+               bp->macbgem_ops.mog_init_rings = macb_init_rings;
+               bp->macbgem_ops.mog_rx = macb_rx;
+               dev->ethtool_ops = &macb_ethtool_ops;
+       }
+
+       /* Set features */
+       dev->hw_features = NETIF_F_SG;
+       /* Checksum offload is only available on gem with packet buffer */
+       if (macb_is_gem(bp) && !(bp->caps & MACB_CAPS_FIFO_MODE))
+               dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
+       if (bp->caps & MACB_CAPS_SG_DISABLED)
+               dev->hw_features &= ~NETIF_F_SG;
+       dev->features = dev->hw_features;
+
+       val = 0;
+       if (bp->phy_interface == PHY_INTERFACE_MODE_RGMII)
+               val = GEM_BIT(RGMII);
+       else if (bp->phy_interface == PHY_INTERFACE_MODE_RMII &&
+                (bp->caps & MACB_CAPS_USRIO_DEFAULT_IS_MII))
+               val = MACB_BIT(RMII);
+       else if (!(bp->caps & MACB_CAPS_USRIO_DEFAULT_IS_MII))
+               val = MACB_BIT(MII);
+
+       if (bp->caps & MACB_CAPS_USRIO_HAS_CLKEN)
+               val |= MACB_BIT(CLKEN);
+
+       macb_or_gem_writel(bp, USRIO, val);
+
+       /* Set MII management clock divider */
+       val = macb_mdc_clk_div(bp);
+       val |= macb_dbw(bp);
+       macb_writel(bp, NCFGR, val);
+
+       return 0;
+}
+
+#if defined(CONFIG_OF)
+/* 1518 rounded up */
+#define AT91ETHER_MAX_RBUFF_SZ 0x600
+/* max number of receive buffers */
+#define AT91ETHER_MAX_RX_DESCR 9
+
+/* Initialize and start the Receiver and Transmit subsystems */
+static int at91ether_start(struct net_device *dev)
+{
+       struct macb *lp = netdev_priv(dev);
+       dma_addr_t addr;
+       u32 ctl;
+       int i;
+
+       lp->rx_ring = dma_alloc_coherent(&lp->pdev->dev,
+                                        (AT91ETHER_MAX_RX_DESCR *
+                                         sizeof(struct macb_dma_desc)),
+                                        &lp->rx_ring_dma, GFP_KERNEL);
+       if (!lp->rx_ring)
+               return -ENOMEM;
+
+       lp->rx_buffers = dma_alloc_coherent(&lp->pdev->dev,
+                                           AT91ETHER_MAX_RX_DESCR *
+                                           AT91ETHER_MAX_RBUFF_SZ,
+                                           &lp->rx_buffers_dma, GFP_KERNEL);
+       if (!lp->rx_buffers) {
+               dma_free_coherent(&lp->pdev->dev,
+                                 AT91ETHER_MAX_RX_DESCR *
+                                 sizeof(struct macb_dma_desc),
+                                 lp->rx_ring, lp->rx_ring_dma);
+               lp->rx_ring = NULL;
+               return -ENOMEM;
+       }
+
+       addr = lp->rx_buffers_dma;
+       for (i = 0; i < AT91ETHER_MAX_RX_DESCR; i++) {
+               lp->rx_ring[i].addr = addr;
+               lp->rx_ring[i].ctrl = 0;
+               addr += AT91ETHER_MAX_RBUFF_SZ;
+       }
+
+       /* Set the Wrap bit on the last descriptor */
+       lp->rx_ring[AT91ETHER_MAX_RX_DESCR - 1].addr |= MACB_BIT(RX_WRAP);
+
+       /* Reset buffer index */
+       lp->rx_tail = 0;
+
+       /* Program address of descriptor list in Rx Buffer Queue register */
+       macb_writel(lp, RBQP, lp->rx_ring_dma);
+
+       /* Enable Receive and Transmit */
+       ctl = macb_readl(lp, NCR);
+       macb_writel(lp, NCR, ctl | MACB_BIT(RE) | MACB_BIT(TE));
+
+       return 0;
+}
+
+/* Open the ethernet interface */
+static int at91ether_open(struct net_device *dev)
+{
+       struct macb *lp = netdev_priv(dev);
+       u32 ctl;
+       int ret;
+
+       /* Clear internal statistics */
+       ctl = macb_readl(lp, NCR);
+       macb_writel(lp, NCR, ctl | MACB_BIT(CLRSTAT));
+
+       macb_set_hwaddr(lp);
+
+       ret = at91ether_start(dev);
+       if (ret)
+               return ret;
+
+       /* Enable MAC interrupts */
+       macb_writel(lp, IER, MACB_BIT(RCOMP)    |
+                            MACB_BIT(RXUBR)    |
+                            MACB_BIT(ISR_TUND) |
+                            MACB_BIT(ISR_RLE)  |
+                            MACB_BIT(TCOMP)    |
+                            MACB_BIT(ISR_ROVR) |
+                            MACB_BIT(HRESP));
+
+       /* schedule a link state check */
+       phy_start(lp->phy_dev);
+
+       netif_start_queue(dev);
+
+       return 0;
+}
+
+/* Close the interface */
+static int at91ether_close(struct net_device *dev)
+{
+       struct macb *lp = netdev_priv(dev);
+       u32 ctl;
+
+       /* Disable Receiver and Transmitter */
+       ctl = macb_readl(lp, NCR);
+       macb_writel(lp, NCR, ctl & ~(MACB_BIT(TE) | MACB_BIT(RE)));
+
+       /* Disable MAC interrupts */
+       macb_writel(lp, IDR, MACB_BIT(RCOMP)    |
+                            MACB_BIT(RXUBR)    |
+                            MACB_BIT(ISR_TUND) |
+                            MACB_BIT(ISR_RLE)  |
+                            MACB_BIT(TCOMP)    |
+                            MACB_BIT(ISR_ROVR) |
+                            MACB_BIT(HRESP));
+
+       netif_stop_queue(dev);
+
+       dma_free_coherent(&lp->pdev->dev,
+                         AT91ETHER_MAX_RX_DESCR *
+                         sizeof(struct macb_dma_desc),
+                         lp->rx_ring, lp->rx_ring_dma);
+       lp->rx_ring = NULL;
+
+       dma_free_coherent(&lp->pdev->dev,
+                         AT91ETHER_MAX_RX_DESCR * AT91ETHER_MAX_RBUFF_SZ,
+                         lp->rx_buffers, lp->rx_buffers_dma);
+       lp->rx_buffers = NULL;
+
+       return 0;
+}
+
+/* Transmit packet */
+static int at91ether_start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct macb *lp = netdev_priv(dev);
+
+       if (macb_readl(lp, TSR) & MACB_BIT(RM9200_BNQ)) {
+               netif_stop_queue(dev);
+
+               /* Store packet information (to free when Tx completed) */
+               lp->skb = skb;
+               lp->skb_length = skb->len;
+               lp->skb_physaddr = dma_map_single(NULL, skb->data, skb->len,
+                                                       DMA_TO_DEVICE);
+
+               /* Set address of the data in the Transmit Address register */
+               macb_writel(lp, TAR, lp->skb_physaddr);
+               /* Set length of the packet in the Transmit Control register */
+               macb_writel(lp, TCR, skb->len);
+
+       } else {
+               netdev_err(dev, "%s called, but device is busy!\n", __func__);
+               return NETDEV_TX_BUSY;
+       }
+
+       return NETDEV_TX_OK;
+}
+
+/* Extract received frame from buffer descriptors and sent to upper layers.
+ * (Called from interrupt context)
+ */
+static void at91ether_rx(struct net_device *dev)
+{
+       struct macb *lp = netdev_priv(dev);
+       unsigned char *p_recv;
+       struct sk_buff *skb;
+       unsigned int pktlen;
+
+       while (lp->rx_ring[lp->rx_tail].addr & MACB_BIT(RX_USED)) {
+               p_recv = lp->rx_buffers + lp->rx_tail * AT91ETHER_MAX_RBUFF_SZ;
+               pktlen = MACB_BF(RX_FRMLEN, lp->rx_ring[lp->rx_tail].ctrl);
+               skb = netdev_alloc_skb(dev, pktlen + 2);
+               if (skb) {
+                       skb_reserve(skb, 2);
+                       memcpy(skb_put(skb, pktlen), p_recv, pktlen);
+
+                       skb->protocol = eth_type_trans(skb, dev);
+                       lp->stats.rx_packets++;
+                       lp->stats.rx_bytes += pktlen;
+                       netif_rx(skb);
+               } else {
+                       lp->stats.rx_dropped++;
+               }
+
+               if (lp->rx_ring[lp->rx_tail].ctrl & MACB_BIT(RX_MHASH_MATCH))
+                       lp->stats.multicast++;
+
+               /* reset ownership bit */
+               lp->rx_ring[lp->rx_tail].addr &= ~MACB_BIT(RX_USED);
+
+               /* wrap after last buffer */
+               if (lp->rx_tail == AT91ETHER_MAX_RX_DESCR - 1)
+                       lp->rx_tail = 0;
+               else
+                       lp->rx_tail++;
+       }
+}
+
+/* MAC interrupt handler */
+static irqreturn_t at91ether_interrupt(int irq, void *dev_id)
+{
+       struct net_device *dev = dev_id;
+       struct macb *lp = netdev_priv(dev);
+       u32 intstatus, ctl;
+
+       /* MAC Interrupt Status register indicates what interrupts are pending.
+        * It is automatically cleared once read.
+        */
+       intstatus = macb_readl(lp, ISR);
+
+       /* Receive complete */
+       if (intstatus & MACB_BIT(RCOMP))
+               at91ether_rx(dev);
+
+       /* Transmit complete */
+       if (intstatus & MACB_BIT(TCOMP)) {
+               /* The TCOM bit is set even if the transmission failed */
+               if (intstatus & (MACB_BIT(ISR_TUND) | MACB_BIT(ISR_RLE)))
+                       lp->stats.tx_errors++;
+
+               if (lp->skb) {
+                       dev_kfree_skb_irq(lp->skb);
+                       lp->skb = NULL;
+                       dma_unmap_single(NULL, lp->skb_physaddr,
+                                        lp->skb_length, DMA_TO_DEVICE);
+                       lp->stats.tx_packets++;
+                       lp->stats.tx_bytes += lp->skb_length;
+               }
+               netif_wake_queue(dev);
+       }
+
+       /* Work-around for EMAC Errata section 41.3.1 */
+       if (intstatus & MACB_BIT(RXUBR)) {
+               ctl = macb_readl(lp, NCR);
+               macb_writel(lp, NCR, ctl & ~MACB_BIT(RE));
+               macb_writel(lp, NCR, ctl | MACB_BIT(RE));
+       }
+
+       if (intstatus & MACB_BIT(ISR_ROVR))
+               netdev_err(dev, "ROVR error\n");
+
+       return IRQ_HANDLED;
+}
+
+#ifdef CONFIG_NET_POLL_CONTROLLER
+static void at91ether_poll_controller(struct net_device *dev)
+{
+       unsigned long flags;
+
+       local_irq_save(flags);
+       at91ether_interrupt(dev->irq, dev);
+       local_irq_restore(flags);
+}
+#endif
+
+static const struct net_device_ops at91ether_netdev_ops = {
+       .ndo_open               = at91ether_open,
+       .ndo_stop               = at91ether_close,
+       .ndo_start_xmit         = at91ether_start_xmit,
+       .ndo_get_stats          = macb_get_stats,
+       .ndo_set_rx_mode        = macb_set_rx_mode,
+       .ndo_set_mac_address    = eth_mac_addr,
+       .ndo_do_ioctl           = macb_ioctl,
+       .ndo_validate_addr      = eth_validate_addr,
+       .ndo_change_mtu         = eth_change_mtu,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = at91ether_poll_controller,
+#endif
+};
+
+static int at91ether_clk_init(struct platform_device *pdev, struct clk **pclk,
+                             struct clk **hclk, struct clk **tx_clk)
+{
+       int err;
+
+       *hclk = NULL;
+       *tx_clk = NULL;
+
+       *pclk = devm_clk_get(&pdev->dev, "ether_clk");
+       if (IS_ERR(*pclk))
+               return PTR_ERR(*pclk);
+
+       err = clk_prepare_enable(*pclk);
+       if (err) {
+               dev_err(&pdev->dev, "failed to enable pclk (%u)\n", err);
+               return err;
+       }
+
+       return 0;
+}
+
+static int at91ether_init(struct platform_device *pdev)
+{
+       struct net_device *dev = platform_get_drvdata(pdev);
+       struct macb *bp = netdev_priv(dev);
+       int err;
+       u32 reg;
+
+       dev->netdev_ops = &at91ether_netdev_ops;
+       dev->ethtool_ops = &macb_ethtool_ops;
+
+       err = devm_request_irq(&pdev->dev, dev->irq, at91ether_interrupt,
+                              0, dev->name, dev);
+       if (err)
+               return err;
+
+       macb_writel(bp, NCR, 0);
+
+       reg = MACB_BF(CLK, MACB_CLK_DIV32) | MACB_BIT(BIG);
+       if (bp->phy_interface == PHY_INTERFACE_MODE_RMII)
+               reg |= MACB_BIT(RM9200_RMII);
+
+       macb_writel(bp, NCFGR, reg);
+
+       return 0;
+}
+
+static const struct macb_config at91sam9260_config = {
+       .caps = MACB_CAPS_USRIO_HAS_CLKEN | MACB_CAPS_USRIO_DEFAULT_IS_MII,
+       .clk_init = macb_clk_init,
+       .init = macb_init,
+};
+
+static const struct macb_config pc302gem_config = {
+       .caps = MACB_CAPS_SG_DISABLED | MACB_CAPS_GIGABIT_MODE_AVAILABLE,
+       .dma_burst_length = 16,
+       .clk_init = macb_clk_init,
+       .init = macb_init,
+};
+
+static const struct macb_config sama5d3_config = {
+       .caps = MACB_CAPS_SG_DISABLED | MACB_CAPS_GIGABIT_MODE_AVAILABLE,
+       .dma_burst_length = 16,
+       .clk_init = macb_clk_init,
+       .init = macb_init,
+};
+
+static const struct macb_config sama5d4_config = {
+       .caps = 0,
+       .dma_burst_length = 4,
+       .clk_init = macb_clk_init,
+       .init = macb_init,
+};
+
+static const struct macb_config emac_config = {
+       .clk_init = at91ether_clk_init,
+       .init = at91ether_init,
+};
+
+static const struct macb_config zynq_config = {
+       .caps = MACB_CAPS_SG_DISABLED | MACB_CAPS_GIGABIT_MODE_AVAILABLE |
+               MACB_CAPS_NO_GIGABIT_HALF,
+       .dma_burst_length = 16,
+       .clk_init = macb_clk_init,
+       .init = macb_init,
+};
+
+static const struct of_device_id macb_dt_ids[] = {
+       { .compatible = "cdns,at32ap7000-macb" },
+       { .compatible = "cdns,at91sam9260-macb", .data = &at91sam9260_config },
+       { .compatible = "cdns,macb" },
+       { .compatible = "cdns,pc302-gem", .data = &pc302gem_config },
+       { .compatible = "cdns,gem", .data = &pc302gem_config },
+       { .compatible = "atmel,sama5d3-gem", .data = &sama5d3_config },
+       { .compatible = "atmel,sama5d4-gem", .data = &sama5d4_config },
+       { .compatible = "cdns,at91rm9200-emac", .data = &emac_config },
+       { .compatible = "cdns,emac", .data = &emac_config },
+       { .compatible = "cdns,zynq-gem", .data = &zynq_config },
+       { /* sentinel */ }
+};
+MODULE_DEVICE_TABLE(of, macb_dt_ids);
+#endif /* CONFIG_OF */
+
+static int macb_probe(struct platform_device *pdev)
+{
+       int (*clk_init)(struct platform_device *, struct clk **,
+                       struct clk **, struct clk **)
+                                             = macb_clk_init;
+       int (*init)(struct platform_device *) = macb_init;
+       struct device_node *np = pdev->dev.of_node;
+       const struct macb_config *macb_config = NULL;
+       struct clk *pclk, *hclk, *tx_clk;
+       unsigned int queue_mask, num_queues;
+       struct macb_platform_data *pdata;
+       struct phy_device *phydev;
+       struct net_device *dev;
+       struct resource *regs;
+       void __iomem *mem;
+       const char *mac;
+       struct macb *bp;
+       int err;
+
+       if (np) {
+               const struct of_device_id *match;
+
+               match = of_match_node(macb_dt_ids, np);
+               if (match && match->data) {
+                       macb_config = match->data;
+                       clk_init = macb_config->clk_init;
+                       init = macb_config->init;
+               }
+       }
+
+       err = clk_init(pdev, &pclk, &hclk, &tx_clk);
+       if (err)
+               return err;
+
+       regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       mem = devm_ioremap_resource(&pdev->dev, regs);
+       if (IS_ERR(mem)) {
+               err = PTR_ERR(mem);
+               goto err_disable_clocks;
+       }
+
+       macb_probe_queues(mem, &queue_mask, &num_queues);
+       dev = alloc_etherdev_mq(sizeof(*bp), num_queues);
+       if (!dev) {
+               err = -ENOMEM;
+               goto err_disable_clocks;
+       }
+
+       dev->base_addr = regs->start;
+
+       SET_NETDEV_DEV(dev, &pdev->dev);
+
+       bp = netdev_priv(dev);
+       bp->pdev = pdev;
+       bp->dev = dev;
+       bp->regs = mem;
+       bp->num_queues = num_queues;
+       bp->queue_mask = queue_mask;
+       if (macb_config)
+               bp->dma_burst_length = macb_config->dma_burst_length;
+       bp->pclk = pclk;
+       bp->hclk = hclk;
+       bp->tx_clk = tx_clk;
+       spin_lock_init(&bp->lock);
+
+       /* setup capabilities */
+       macb_configure_caps(bp, macb_config);
+
+       platform_set_drvdata(pdev, dev);
+
+       dev->irq = platform_get_irq(pdev, 0);
+       if (dev->irq < 0) {
+               err = dev->irq;
+               goto err_disable_clocks;
+       }
+
+       mac = of_get_mac_address(np);
+       if (mac)
+               memcpy(bp->dev->dev_addr, mac, ETH_ALEN);
+       else
+               macb_get_hwaddr(bp);
+
+       err = of_get_phy_mode(np);
+       if (err < 0) {
+               pdata = dev_get_platdata(&pdev->dev);
+               if (pdata && pdata->is_rmii)
+                       bp->phy_interface = PHY_INTERFACE_MODE_RMII;
+               else
+                       bp->phy_interface = PHY_INTERFACE_MODE_MII;
+       } else {
+               bp->phy_interface = err;
+       }
+
+       /* IP specific init */
+       err = init(pdev);
+       if (err)
+               goto err_out_free_netdev;
+
+       err = register_netdev(dev);
+       if (err) {
+               dev_err(&pdev->dev, "Cannot register net device, aborting.\n");
+               goto err_out_unregister_netdev;
+       }
+
+       err = macb_mii_init(bp);
+       if (err)
+               goto err_out_unregister_netdev;
+
+       netif_carrier_off(dev);
+
+       netdev_info(dev, "Cadence %s rev 0x%08x at 0x%08lx irq %d (%pM)\n",
+                   macb_is_gem(bp) ? "GEM" : "MACB", macb_readl(bp, MID),
+                   dev->base_addr, dev->irq, dev->dev_addr);
+
+       phydev = bp->phy_dev;
+       netdev_info(dev, "attached PHY driver [%s] (mii_bus:phy_addr=%s, irq=%d)\n",
+                   phydev->drv->name, dev_name(&phydev->dev), phydev->irq);
+
+       return 0;
+
+err_out_unregister_netdev:
+       unregister_netdev(dev);
+
+err_out_free_netdev:
+       free_netdev(dev);
+
+err_disable_clocks:
+       clk_disable_unprepare(tx_clk);
+       clk_disable_unprepare(hclk);
+       clk_disable_unprepare(pclk);
+
+       return err;
+}
+
+static int macb_remove(struct platform_device *pdev)
+{
+       struct net_device *dev;
+       struct macb *bp;
+
+       dev = platform_get_drvdata(pdev);
+
+       if (dev) {
+               bp = netdev_priv(dev);
+               if (bp->phy_dev)
+                       phy_disconnect(bp->phy_dev);
+               mdiobus_unregister(bp->mii_bus);
+               kfree(bp->mii_bus->irq);
+               mdiobus_free(bp->mii_bus);
+               unregister_netdev(dev);
+               clk_disable_unprepare(bp->tx_clk);
+               clk_disable_unprepare(bp->hclk);
+               clk_disable_unprepare(bp->pclk);
+               free_netdev(dev);
+       }
+
+       return 0;
+}
+
+static int __maybe_unused macb_suspend(struct device *dev)
+{
+       struct platform_device *pdev = to_platform_device(dev);
+       struct net_device *netdev = platform_get_drvdata(pdev);
+       struct macb *bp = netdev_priv(netdev);
+
+       netif_carrier_off(netdev);
+       netif_device_detach(netdev);
+
+       clk_disable_unprepare(bp->tx_clk);
+       clk_disable_unprepare(bp->hclk);
+       clk_disable_unprepare(bp->pclk);
+
+       return 0;
+}
+
+static int __maybe_unused macb_resume(struct device *dev)
+{
+       struct platform_device *pdev = to_platform_device(dev);
+       struct net_device *netdev = platform_get_drvdata(pdev);
+       struct macb *bp = netdev_priv(netdev);
+
+       clk_prepare_enable(bp->pclk);
+       clk_prepare_enable(bp->hclk);
+       clk_prepare_enable(bp->tx_clk);
+
+       netif_device_attach(netdev);
+
+       return 0;
+}
+
+static SIMPLE_DEV_PM_OPS(macb_pm_ops, macb_suspend, macb_resume);
+
+static struct platform_driver macb_driver = {
+       .probe          = macb_probe,
+       .remove         = macb_remove,
+       .driver         = {
+               .name           = "macb",
+               .of_match_table = of_match_ptr(macb_dt_ids),
+               .pm     = &macb_pm_ops,
+       },
+};
+
+module_platform_driver(macb_driver);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Cadence MACB/GEM Ethernet driver");
+MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
+MODULE_ALIAS("platform:macb");