]> err.no Git - linux-2.6/commitdiff
Driver for IXP4xx built-in Ethernet ports
authorKrzysztof Halasa <khc@pm.waw.pl>
Sun, 20 Apr 2008 17:06:39 +0000 (19:06 +0200)
committerJeff Garzik <jgarzik@redhat.com>
Tue, 29 Apr 2008 05:56:50 +0000 (01:56 -0400)
Adds a driver for built-in IXP4xx Ethernet ports.

Signed-off-by: Krzysztof HaƂasa <khc@pm.waw.pl>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
arch/arm/mach-ixp4xx/ixp4xx_npe.c
arch/arm/mach-ixp4xx/ixp4xx_qmgr.c
drivers/net/arm/Kconfig
drivers/net/arm/Makefile
drivers/net/arm/ixp4xx_eth.c [new file with mode: 0644]

index 83c137ec582ce63441556ed4f2e78c1451fa600d..63a23fa4aab4c21b9d03d44ce08fa5023621badf 100644 (file)
@@ -448,7 +448,9 @@ int npe_send_message(struct npe *npe, const void *msg, const char *what)
                return -ETIMEDOUT;
        }
 
+#if DEBUG_MSG > 1
        debug_msg(npe, "Sending a message took %i cycles\n", cycles);
+#endif
        return 0;
 }
 
@@ -484,7 +486,9 @@ int npe_recv_message(struct npe *npe, void *msg, const char *what)
                return -ETIMEDOUT;
        }
 
+#if DEBUG_MSG > 1
        debug_msg(npe, "Receiving a message took %i cycles\n", cycles);
+#endif
        return 0;
 }
 
index e8330132530175fe43363983eecee4d90ede5dc9..fab94eaecee7abb797b1645a121229f41d7eaf91 100644 (file)
@@ -184,6 +184,8 @@ void qmgr_release_queue(unsigned int queue)
        case 3: mask[0] = 0xFF; break;
        }
 
+       mask[1] = mask[2] = mask[3] = 0;
+
        while (addr--)
                shift_mask(mask);
 
index f9cc2b621fe200d6a4c588f2e075f4641c9d1ad1..8eda6eeb43b70ffbbc39463d0c9161cb89585fd1 100644 (file)
@@ -47,3 +47,11 @@ config EP93XX_ETH
        help
          This is a driver for the ethernet hardware included in EP93xx CPUs.
          Say Y if you are building a kernel for EP93xx based devices.
+
+config IXP4XX_ETH
+       tristate "Intel IXP4xx Ethernet support"
+       depends on ARM && ARCH_IXP4XX && IXP4XX_NPE && IXP4XX_QMGR
+       select MII
+       help
+         Say Y here if you want to use built-in Ethernet ports
+         on IXP4xx processor.
index a4c868278e115c5d1c5ca1fbf263fdd682e6d513..7c812ac2b6a51c7305f8b5e69b51b219959f20df 100644 (file)
@@ -9,3 +9,4 @@ obj-$(CONFIG_ARM_ETHER3)        += ether3.o
 obj-$(CONFIG_ARM_ETHER1)       += ether1.o
 obj-$(CONFIG_ARM_AT91_ETHER)   += at91_ether.o
 obj-$(CONFIG_EP93XX_ETH)       += ep93xx_eth.o
+obj-$(CONFIG_IXP4XX_ETH)       += ixp4xx_eth.o
diff --git a/drivers/net/arm/ixp4xx_eth.c b/drivers/net/arm/ixp4xx_eth.c
new file mode 100644 (file)
index 0000000..c617b64
--- /dev/null
@@ -0,0 +1,1265 @@
+/*
+ * Intel IXP4xx Ethernet driver for Linux
+ *
+ * Copyright (C) 2007 Krzysztof Halasa <khc@pm.waw.pl>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of version 2 of the GNU General Public License
+ * as published by the Free Software Foundation.
+ *
+ * Ethernet port config (0x00 is not present on IXP42X):
+ *
+ * logical port                0x00            0x10            0x20
+ * NPE                 0 (NPE-A)       1 (NPE-B)       2 (NPE-C)
+ * physical PortId     2               0               1
+ * TX queue            23              24              25
+ * RX-free queue       26              27              28
+ * TX-done queue is always 31, per-port RX and TX-ready queues are configurable
+ *
+ *
+ * Queue entries:
+ * bits 0 -> 1 - NPE ID (RX and TX-done)
+ * bits 0 -> 2 - priority (TX, per 802.1D)
+ * bits 3 -> 4 - port ID (user-set?)
+ * bits 5 -> 31        - physical descriptor address
+ */
+
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/dmapool.h>
+#include <linux/etherdevice.h>
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/mii.h>
+#include <linux/platform_device.h>
+#include <asm/arch/npe.h>
+#include <asm/arch/qmgr.h>
+
+#define DEBUG_QUEUES           0
+#define DEBUG_DESC             0
+#define DEBUG_RX               0
+#define DEBUG_TX               0
+#define DEBUG_PKT_BYTES                0
+#define DEBUG_MDIO             0
+#define DEBUG_CLOSE            0
+
+#define DRV_NAME               "ixp4xx_eth"
+
+#define MAX_NPES               3
+
+#define RX_DESCS               64 /* also length of all RX queues */
+#define TX_DESCS               16 /* also length of all TX queues */
+#define TXDONE_QUEUE_LEN       64 /* dwords */
+
+#define POOL_ALLOC_SIZE                (sizeof(struct desc) * (RX_DESCS + TX_DESCS))
+#define REGS_SIZE              0x1000
+#define MAX_MRU                        1536 /* 0x600 */
+#define RX_BUFF_SIZE           ALIGN((NET_IP_ALIGN) + MAX_MRU, 4)
+
+#define NAPI_WEIGHT            16
+#define MDIO_INTERVAL          (3 * HZ)
+#define MAX_MDIO_RETRIES       100 /* microseconds, typically 30 cycles */
+#define MAX_MII_RESET_RETRIES  100 /* mdio_read() cycles, typically 4 */
+#define MAX_CLOSE_WAIT         1000 /* microseconds, typically 2-3 cycles */
+
+#define NPE_ID(port_id)                ((port_id) >> 4)
+#define PHYSICAL_ID(port_id)   ((NPE_ID(port_id) + 2) % 3)
+#define TX_QUEUE(port_id)      (NPE_ID(port_id) + 23)
+#define RXFREE_QUEUE(port_id)  (NPE_ID(port_id) + 26)
+#define TXDONE_QUEUE           31
+
+/* TX Control Registers */
+#define TX_CNTRL0_TX_EN                0x01
+#define TX_CNTRL0_HALFDUPLEX   0x02
+#define TX_CNTRL0_RETRY                0x04
+#define TX_CNTRL0_PAD_EN       0x08
+#define TX_CNTRL0_APPEND_FCS   0x10
+#define TX_CNTRL0_2DEFER       0x20
+#define TX_CNTRL0_RMII         0x40 /* reduced MII */
+#define TX_CNTRL1_RETRIES      0x0F /* 4 bits */
+
+/* RX Control Registers */
+#define RX_CNTRL0_RX_EN                0x01
+#define RX_CNTRL0_PADSTRIP_EN  0x02
+#define RX_CNTRL0_SEND_FCS     0x04
+#define RX_CNTRL0_PAUSE_EN     0x08
+#define RX_CNTRL0_LOOP_EN      0x10
+#define RX_CNTRL0_ADDR_FLTR_EN 0x20
+#define RX_CNTRL0_RX_RUNT_EN   0x40
+#define RX_CNTRL0_BCAST_DIS    0x80
+#define RX_CNTRL1_DEFER_EN     0x01
+
+/* Core Control Register */
+#define CORE_RESET             0x01
+#define CORE_RX_FIFO_FLUSH     0x02
+#define CORE_TX_FIFO_FLUSH     0x04
+#define CORE_SEND_JAM          0x08
+#define CORE_MDC_EN            0x10 /* MDIO using NPE-B ETH-0 only */
+
+#define DEFAULT_TX_CNTRL0      (TX_CNTRL0_TX_EN | TX_CNTRL0_RETRY |    \
+                                TX_CNTRL0_PAD_EN | TX_CNTRL0_APPEND_FCS | \
+                                TX_CNTRL0_2DEFER)
+#define DEFAULT_RX_CNTRL0      RX_CNTRL0_RX_EN
+#define DEFAULT_CORE_CNTRL     CORE_MDC_EN
+
+
+/* NPE message codes */
+#define NPE_GETSTATUS                  0x00
+#define NPE_EDB_SETPORTADDRESS         0x01
+#define NPE_EDB_GETMACADDRESSDATABASE  0x02
+#define NPE_EDB_SETMACADDRESSSDATABASE 0x03
+#define NPE_GETSTATS                   0x04
+#define NPE_RESETSTATS                 0x05
+#define NPE_SETMAXFRAMELENGTHS         0x06
+#define NPE_VLAN_SETRXTAGMODE          0x07
+#define NPE_VLAN_SETDEFAULTRXVID       0x08
+#define NPE_VLAN_SETPORTVLANTABLEENTRY 0x09
+#define NPE_VLAN_SETPORTVLANTABLERANGE 0x0A
+#define NPE_VLAN_SETRXQOSENTRY         0x0B
+#define NPE_VLAN_SETPORTIDEXTRACTIONMODE 0x0C
+#define NPE_STP_SETBLOCKINGSTATE       0x0D
+#define NPE_FW_SETFIREWALLMODE         0x0E
+#define NPE_PC_SETFRAMECONTROLDURATIONID 0x0F
+#define NPE_PC_SETAPMACTABLE           0x11
+#define NPE_SETLOOPBACK_MODE           0x12
+#define NPE_PC_SETBSSIDTABLE           0x13
+#define NPE_ADDRESS_FILTER_CONFIG      0x14
+#define NPE_APPENDFCSCONFIG            0x15
+#define NPE_NOTIFY_MAC_RECOVERY_DONE   0x16
+#define NPE_MAC_RECOVERY_START         0x17
+
+
+#ifdef __ARMEB__
+typedef struct sk_buff buffer_t;
+#define free_buffer dev_kfree_skb
+#define free_buffer_irq dev_kfree_skb_irq
+#else
+typedef void buffer_t;
+#define free_buffer kfree
+#define free_buffer_irq kfree
+#endif
+
+struct eth_regs {
+       u32 tx_control[2], __res1[2];           /* 000 */
+       u32 rx_control[2], __res2[2];           /* 010 */
+       u32 random_seed, __res3[3];             /* 020 */
+       u32 partial_empty_threshold, __res4;    /* 030 */
+       u32 partial_full_threshold, __res5;     /* 038 */
+       u32 tx_start_bytes, __res6[3];          /* 040 */
+       u32 tx_deferral, rx_deferral, __res7[2];/* 050 */
+       u32 tx_2part_deferral[2], __res8[2];    /* 060 */
+       u32 slot_time, __res9[3];               /* 070 */
+       u32 mdio_command[4];                    /* 080 */
+       u32 mdio_status[4];                     /* 090 */
+       u32 mcast_mask[6], __res10[2];          /* 0A0 */
+       u32 mcast_addr[6], __res11[2];          /* 0C0 */
+       u32 int_clock_threshold, __res12[3];    /* 0E0 */
+       u32 hw_addr[6], __res13[61];            /* 0F0 */
+       u32 core_control;                       /* 1FC */
+};
+
+struct port {
+       struct resource *mem_res;
+       struct eth_regs __iomem *regs;
+       struct npe *npe;
+       struct net_device *netdev;
+       struct napi_struct napi;
+       struct net_device_stats stat;
+       struct mii_if_info mii;
+       struct delayed_work mdio_thread;
+       struct eth_plat_info *plat;
+       buffer_t *rx_buff_tab[RX_DESCS], *tx_buff_tab[TX_DESCS];
+       struct desc *desc_tab;  /* coherent */
+       u32 desc_tab_phys;
+       int id;                 /* logical port ID */
+       u16 mii_bmcr;
+};
+
+/* NPE message structure */
+struct msg {
+#ifdef __ARMEB__
+       u8 cmd, eth_id, byte2, byte3;
+       u8 byte4, byte5, byte6, byte7;
+#else
+       u8 byte3, byte2, eth_id, cmd;
+       u8 byte7, byte6, byte5, byte4;
+#endif
+};
+
+/* Ethernet packet descriptor */
+struct desc {
+       u32 next;               /* pointer to next buffer, unused */
+
+#ifdef __ARMEB__
+       u16 buf_len;            /* buffer length */
+       u16 pkt_len;            /* packet length */
+       u32 data;               /* pointer to data buffer in RAM */
+       u8 dest_id;
+       u8 src_id;
+       u16 flags;
+       u8 qos;
+       u8 padlen;
+       u16 vlan_tci;
+#else
+       u16 pkt_len;            /* packet length */
+       u16 buf_len;            /* buffer length */
+       u32 data;               /* pointer to data buffer in RAM */
+       u16 flags;
+       u8 src_id;
+       u8 dest_id;
+       u16 vlan_tci;
+       u8 padlen;
+       u8 qos;
+#endif
+
+#ifdef __ARMEB__
+       u8 dst_mac_0, dst_mac_1, dst_mac_2, dst_mac_3;
+       u8 dst_mac_4, dst_mac_5, src_mac_0, src_mac_1;
+       u8 src_mac_2, src_mac_3, src_mac_4, src_mac_5;
+#else
+       u8 dst_mac_3, dst_mac_2, dst_mac_1, dst_mac_0;
+       u8 src_mac_1, src_mac_0, dst_mac_5, dst_mac_4;
+       u8 src_mac_5, src_mac_4, src_mac_3, src_mac_2;
+#endif
+};
+
+
+#define rx_desc_phys(port, n)  ((port)->desc_tab_phys +                \
+                                (n) * sizeof(struct desc))
+#define rx_desc_ptr(port, n)   (&(port)->desc_tab[n])
+
+#define tx_desc_phys(port, n)  ((port)->desc_tab_phys +                \
+                                ((n) + RX_DESCS) * sizeof(struct desc))
+#define tx_desc_ptr(port, n)   (&(port)->desc_tab[(n) + RX_DESCS])
+
+#ifndef __ARMEB__
+static inline void memcpy_swab32(u32 *dest, u32 *src, int cnt)
+{
+       int i;
+       for (i = 0; i < cnt; i++)
+               dest[i] = swab32(src[i]);
+}
+#endif
+
+static spinlock_t mdio_lock;
+static struct eth_regs __iomem *mdio_regs; /* mdio command and status only */
+static int ports_open;
+static struct port *npe_port_tab[MAX_NPES];
+static struct dma_pool *dma_pool;
+
+
+static u16 mdio_cmd(struct net_device *dev, int phy_id, int location,
+                   int write, u16 cmd)
+{
+       int cycles = 0;
+
+       if (__raw_readl(&mdio_regs->mdio_command[3]) & 0x80) {
+               printk(KERN_ERR "%s: MII not ready to transmit\n", dev->name);
+               return 0;
+       }
+
+       if (write) {
+               __raw_writel(cmd & 0xFF, &mdio_regs->mdio_command[0]);
+               __raw_writel(cmd >> 8, &mdio_regs->mdio_command[1]);
+       }
+       __raw_writel(((phy_id << 5) | location) & 0xFF,
+                    &mdio_regs->mdio_command[2]);
+       __raw_writel((phy_id >> 3) | (write << 2) | 0x80 /* GO */,
+                    &mdio_regs->mdio_command[3]);
+
+       while ((cycles < MAX_MDIO_RETRIES) &&
+              (__raw_readl(&mdio_regs->mdio_command[3]) & 0x80)) {
+               udelay(1);
+               cycles++;
+       }
+
+       if (cycles == MAX_MDIO_RETRIES) {
+               printk(KERN_ERR "%s: MII write failed\n", dev->name);
+               return 0;
+       }
+
+#if DEBUG_MDIO
+       printk(KERN_DEBUG "%s: mdio_cmd() took %i cycles\n", dev->name,
+              cycles);
+#endif
+
+       if (write)
+               return 0;
+
+       if (__raw_readl(&mdio_regs->mdio_status[3]) & 0x80) {
+               printk(KERN_ERR "%s: MII read failed\n", dev->name);
+               return 0;
+       }
+
+       return (__raw_readl(&mdio_regs->mdio_status[0]) & 0xFF) |
+               (__raw_readl(&mdio_regs->mdio_status[1]) << 8);
+}
+
+static int mdio_read(struct net_device *dev, int phy_id, int location)
+{
+       unsigned long flags;
+       u16 val;
+
+       spin_lock_irqsave(&mdio_lock, flags);
+       val = mdio_cmd(dev, phy_id, location, 0, 0);
+       spin_unlock_irqrestore(&mdio_lock, flags);
+       return val;
+}
+
+static void mdio_write(struct net_device *dev, int phy_id, int location,
+                      int val)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&mdio_lock, flags);
+       mdio_cmd(dev, phy_id, location, 1, val);
+       spin_unlock_irqrestore(&mdio_lock, flags);
+}
+
+static void phy_reset(struct net_device *dev, int phy_id)
+{
+       struct port *port = netdev_priv(dev);
+       int cycles = 0;
+
+       mdio_write(dev, phy_id, MII_BMCR, port->mii_bmcr | BMCR_RESET);
+
+       while (cycles < MAX_MII_RESET_RETRIES) {
+               if (!(mdio_read(dev, phy_id, MII_BMCR) & BMCR_RESET)) {
+#if DEBUG_MDIO
+                       printk(KERN_DEBUG "%s: phy_reset() took %i cycles\n",
+                              dev->name, cycles);
+#endif
+                       return;
+               }
+               udelay(1);
+               cycles++;
+       }
+
+       printk(KERN_ERR "%s: MII reset failed\n", dev->name);
+}
+
+static void eth_set_duplex(struct port *port)
+{
+       if (port->mii.full_duplex)
+               __raw_writel(DEFAULT_TX_CNTRL0 & ~TX_CNTRL0_HALFDUPLEX,
+                            &port->regs->tx_control[0]);
+       else
+               __raw_writel(DEFAULT_TX_CNTRL0 | TX_CNTRL0_HALFDUPLEX,
+                            &port->regs->tx_control[0]);
+}
+
+
+static void phy_check_media(struct port *port, int init)
+{
+       if (mii_check_media(&port->mii, 1, init))
+               eth_set_duplex(port);
+       if (port->mii.force_media) { /* mii_check_media() doesn't work */
+               struct net_device *dev = port->netdev;
+               int cur_link = mii_link_ok(&port->mii);
+               int prev_link = netif_carrier_ok(dev);
+
+               if (!prev_link && cur_link) {
+                       printk(KERN_INFO "%s: link up\n", dev->name);
+                       netif_carrier_on(dev);
+               } else if (prev_link && !cur_link) {
+                       printk(KERN_INFO "%s: link down\n", dev->name);
+                       netif_carrier_off(dev);
+               }
+       }
+}
+
+
+static void mdio_thread(struct work_struct *work)
+{
+       struct port *port = container_of(work, struct port, mdio_thread.work);
+
+       phy_check_media(port, 0);
+       schedule_delayed_work(&port->mdio_thread, MDIO_INTERVAL);
+}
+
+
+static inline void debug_pkt(struct net_device *dev, const char *func,
+                            u8 *data, int len)
+{
+#if DEBUG_PKT_BYTES
+       int i;
+
+       printk(KERN_DEBUG "%s: %s(%i) ", dev->name, func, len);
+       for (i = 0; i < len; i++) {
+               if (i >= DEBUG_PKT_BYTES)
+                       break;
+               printk("%s%02X",
+                      ((i == 6) || (i == 12) || (i >= 14)) ? " " : "",
+                      data[i]);
+       }
+       printk("\n");
+#endif
+}
+
+
+static inline void debug_desc(u32 phys, struct desc *desc)
+{
+#if DEBUG_DESC
+       printk(KERN_DEBUG "%X: %X %3X %3X %08X %2X < %2X %4X %X"
+              " %X %X %02X%02X%02X%02X%02X%02X < %02X%02X%02X%02X%02X%02X\n",
+              phys, desc->next, desc->buf_len, desc->pkt_len,
+              desc->data, desc->dest_id, desc->src_id, desc->flags,
+              desc->qos, desc->padlen, desc->vlan_tci,
+              desc->dst_mac_0, desc->dst_mac_1, desc->dst_mac_2,
+              desc->dst_mac_3, desc->dst_mac_4, desc->dst_mac_5,
+              desc->src_mac_0, desc->src_mac_1, desc->src_mac_2,
+              desc->src_mac_3, desc->src_mac_4, desc->src_mac_5);
+#endif
+}
+
+static inline void debug_queue(unsigned int queue, int is_get, u32 phys)
+{
+#if DEBUG_QUEUES
+       static struct {
+               int queue;
+               char *name;
+       } names[] = {
+               { TX_QUEUE(0x10), "TX#0 " },
+               { TX_QUEUE(0x20), "TX#1 " },
+               { TX_QUEUE(0x00), "TX#2 " },
+               { RXFREE_QUEUE(0x10), "RX-free#0 " },
+               { RXFREE_QUEUE(0x20), "RX-free#1 " },
+               { RXFREE_QUEUE(0x00), "RX-free#2 " },
+               { TXDONE_QUEUE, "TX-done " },
+       };
+       int i;
+
+       for (i = 0; i < ARRAY_SIZE(names); i++)
+               if (names[i].queue == queue)
+                       break;
+
+       printk(KERN_DEBUG "Queue %i %s%s %X\n", queue,
+              i < ARRAY_SIZE(names) ? names[i].name : "",
+              is_get ? "->" : "<-", phys);
+#endif
+}
+
+static inline u32 queue_get_entry(unsigned int queue)
+{
+       u32 phys = qmgr_get_entry(queue);
+       debug_queue(queue, 1, phys);
+       return phys;
+}
+
+static inline int queue_get_desc(unsigned int queue, struct port *port,
+                                int is_tx)
+{
+       u32 phys, tab_phys, n_desc;
+       struct desc *tab;
+
+       if (!(phys = queue_get_entry(queue)))
+               return -1;
+
+       phys &= ~0x1F; /* mask out non-address bits */
+       tab_phys = is_tx ? tx_desc_phys(port, 0) : rx_desc_phys(port, 0);
+       tab = is_tx ? tx_desc_ptr(port, 0) : rx_desc_ptr(port, 0);
+       n_desc = (phys - tab_phys) / sizeof(struct desc);
+       BUG_ON(n_desc >= (is_tx ? TX_DESCS : RX_DESCS));
+       debug_desc(phys, &tab[n_desc]);
+       BUG_ON(tab[n_desc].next);
+       return n_desc;
+}
+
+static inline void queue_put_desc(unsigned int queue, u32 phys,
+                                 struct desc *desc)
+{
+       debug_queue(queue, 0, phys);
+       debug_desc(phys, desc);
+       BUG_ON(phys & 0x1F);
+       qmgr_put_entry(queue, phys);
+       BUG_ON(qmgr_stat_overflow(queue));
+}
+
+
+static inline void dma_unmap_tx(struct port *port, struct desc *desc)
+{
+#ifdef __ARMEB__
+       dma_unmap_single(&port->netdev->dev, desc->data,
+                        desc->buf_len, DMA_TO_DEVICE);
+#else
+       dma_unmap_single(&port->netdev->dev, desc->data & ~3,
+                        ALIGN((desc->data & 3) + desc->buf_len, 4),
+                        DMA_TO_DEVICE);
+#endif
+}
+
+
+static void eth_rx_irq(void *pdev)
+{
+       struct net_device *dev = pdev;
+       struct port *port = netdev_priv(dev);
+
+#if DEBUG_RX
+       printk(KERN_DEBUG "%s: eth_rx_irq\n", dev->name);
+#endif
+       qmgr_disable_irq(port->plat->rxq);
+       netif_rx_schedule(dev, &port->napi);
+}
+
+static int eth_poll(struct napi_struct *napi, int budget)
+{
+       struct port *port = container_of(napi, struct port, napi);
+       struct net_device *dev = port->netdev;
+       unsigned int rxq = port->plat->rxq, rxfreeq = RXFREE_QUEUE(port->id);
+       int received = 0;
+
+#if DEBUG_RX
+       printk(KERN_DEBUG "%s: eth_poll\n", dev->name);
+#endif
+
+       while (received < budget) {
+               struct sk_buff *skb;
+               struct desc *desc;
+               int n;
+#ifdef __ARMEB__
+               struct sk_buff *temp;
+               u32 phys;
+#endif
+
+               if ((n = queue_get_desc(rxq, port, 0)) < 0) {
+                       received = 0; /* No packet received */
+#if DEBUG_RX
+                       printk(KERN_DEBUG "%s: eth_poll netif_rx_complete\n",
+                              dev->name);
+#endif
+                       netif_rx_complete(dev, napi);
+                       qmgr_enable_irq(rxq);
+                       if (!qmgr_stat_empty(rxq) &&
+                           netif_rx_reschedule(dev, napi)) {
+#if DEBUG_RX
+                               printk(KERN_DEBUG "%s: eth_poll"
+                                      " netif_rx_reschedule successed\n",
+                                      dev->name);
+#endif
+                               qmgr_disable_irq(rxq);
+                               continue;
+                       }
+#if DEBUG_RX
+                       printk(KERN_DEBUG "%s: eth_poll all done\n",
+                              dev->name);
+#endif
+                       return 0; /* all work done */
+               }
+
+               desc = rx_desc_ptr(port, n);
+
+#ifdef __ARMEB__
+               if ((skb = netdev_alloc_skb(dev, RX_BUFF_SIZE))) {
+                       phys = dma_map_single(&dev->dev, skb->data,
+                                             RX_BUFF_SIZE, DMA_FROM_DEVICE);
+                       if (dma_mapping_error(phys)) {
+                               dev_kfree_skb(skb);
+                               skb = NULL;
+                       }
+               }
+#else
+               skb = netdev_alloc_skb(dev,
+                                      ALIGN(NET_IP_ALIGN + desc->pkt_len, 4));
+#endif
+
+               if (!skb) {
+                       port->stat.rx_dropped++;
+                       /* put the desc back on RX-ready queue */
+                       desc->buf_len = MAX_MRU;
+                       desc->pkt_len = 0;
+                       queue_put_desc(rxfreeq, rx_desc_phys(port, n), desc);
+                       continue;
+               }
+
+               /* process received frame */
+#ifdef __ARMEB__
+               temp = skb;
+               skb = port->rx_buff_tab[n];
+               dma_unmap_single(&dev->dev, desc->data - NET_IP_ALIGN,
+                                RX_BUFF_SIZE, DMA_FROM_DEVICE);
+#else
+               dma_sync_single(&dev->dev, desc->data - NET_IP_ALIGN,
+                               RX_BUFF_SIZE, DMA_FROM_DEVICE);
+               memcpy_swab32((u32 *)skb->data, (u32 *)port->rx_buff_tab[n],
+                             ALIGN(NET_IP_ALIGN + desc->pkt_len, 4) / 4);
+#endif
+               skb_reserve(skb, NET_IP_ALIGN);
+               skb_put(skb, desc->pkt_len);
+
+               debug_pkt(dev, "eth_poll", skb->data, skb->len);
+
+               skb->protocol = eth_type_trans(skb, dev);
+               dev->last_rx = jiffies;
+               port->stat.rx_packets++;
+               port->stat.rx_bytes += skb->len;
+               netif_receive_skb(skb);
+
+               /* put the new buffer on RX-free queue */
+#ifdef __ARMEB__
+               port->rx_buff_tab[n] = temp;
+               desc->data = phys + NET_IP_ALIGN;
+#endif
+               desc->buf_len = MAX_MRU;
+               desc->pkt_len = 0;
+               queue_put_desc(rxfreeq, rx_desc_phys(port, n), desc);
+               received++;
+       }
+
+#if DEBUG_RX
+       printk(KERN_DEBUG "eth_poll(): end, not all work done\n");
+#endif
+       return received;                /* not all work done */
+}
+
+
+static void eth_txdone_irq(void *unused)
+{
+       u32 phys;
+
+#if DEBUG_TX
+       printk(KERN_DEBUG DRV_NAME ": eth_txdone_irq\n");
+#endif
+       while ((phys = queue_get_entry(TXDONE_QUEUE)) != 0) {
+               u32 npe_id, n_desc;
+               struct port *port;
+               struct desc *desc;
+               int start;
+
+               npe_id = phys & 3;
+               BUG_ON(npe_id >= MAX_NPES);
+               port = npe_port_tab[npe_id];
+               BUG_ON(!port);
+               phys &= ~0x1F; /* mask out non-address bits */
+               n_desc = (phys - tx_desc_phys(port, 0)) / sizeof(struct desc);
+               BUG_ON(n_desc >= TX_DESCS);
+               desc = tx_desc_ptr(port, n_desc);
+               debug_desc(phys, desc);
+
+               if (port->tx_buff_tab[n_desc]) { /* not the draining packet */
+                       port->stat.tx_packets++;
+                       port->stat.tx_bytes += desc->pkt_len;
+
+                       dma_unmap_tx(port, desc);
+#if DEBUG_TX
+                       printk(KERN_DEBUG "%s: eth_txdone_irq free %p\n",
+                              port->netdev->name, port->tx_buff_tab[n_desc]);
+#endif
+                       free_buffer_irq(port->tx_buff_tab[n_desc]);
+                       port->tx_buff_tab[n_desc] = NULL;
+               }
+
+               start = qmgr_stat_empty(port->plat->txreadyq);
+               queue_put_desc(port->plat->txreadyq, phys, desc);
+               if (start) {
+#if DEBUG_TX
+                       printk(KERN_DEBUG "%s: eth_txdone_irq xmit ready\n",
+                              port->netdev->name);
+#endif
+                       netif_wake_queue(port->netdev);
+               }
+       }
+}
+
+static int eth_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct port *port = netdev_priv(dev);
+       unsigned int txreadyq = port->plat->txreadyq;
+       int len, offset, bytes, n;
+       void *mem;
+       u32 phys;
+       struct desc *desc;
+
+#if DEBUG_TX
+       printk(KERN_DEBUG "%s: eth_xmit\n", dev->name);
+#endif
+
+       if (unlikely(skb->len > MAX_MRU)) {
+               dev_kfree_skb(skb);
+               port->stat.tx_errors++;
+               return NETDEV_TX_OK;
+       }
+
+       debug_pkt(dev, "eth_xmit", skb->data, skb->len);
+
+       len = skb->len;
+#ifdef __ARMEB__
+       offset = 0; /* no need to keep alignment */
+       bytes = len;
+       mem = skb->data;
+#else
+       offset = (int)skb->data & 3; /* keep 32-bit alignment */
+       bytes = ALIGN(offset + len, 4);
+       if (!(mem = kmalloc(bytes, GFP_ATOMIC))) {
+               dev_kfree_skb(skb);
+               port->stat.tx_dropped++;
+               return NETDEV_TX_OK;
+       }
+       memcpy_swab32(mem, (u32 *)((int)skb->data & ~3), bytes / 4);
+       dev_kfree_skb(skb);
+#endif
+
+       phys = dma_map_single(&dev->dev, mem, bytes, DMA_TO_DEVICE);
+       if (dma_mapping_error(phys)) {
+#ifdef __ARMEB__
+               dev_kfree_skb(skb);
+#else
+               kfree(mem);
+#endif
+               port->stat.tx_dropped++;
+               return NETDEV_TX_OK;
+       }
+
+       n = queue_get_desc(txreadyq, port, 1);
+       BUG_ON(n < 0);
+       desc = tx_desc_ptr(port, n);
+
+#ifdef __ARMEB__
+       port->tx_buff_tab[n] = skb;
+#else
+       port->tx_buff_tab[n] = mem;
+#endif
+       desc->data = phys + offset;
+       desc->buf_len = desc->pkt_len = len;
+
+       /* NPE firmware pads short frames with zeros internally */
+       wmb();
+       queue_put_desc(TX_QUEUE(port->id), tx_desc_phys(port, n), desc);
+       dev->trans_start = jiffies;
+
+       if (qmgr_stat_empty(txreadyq)) {
+#if DEBUG_TX
+               printk(KERN_DEBUG "%s: eth_xmit queue full\n", dev->name);
+#endif
+               netif_stop_queue(dev);
+               /* we could miss TX ready interrupt */
+               if (!qmgr_stat_empty(txreadyq)) {
+#if DEBUG_TX
+                       printk(KERN_DEBUG "%s: eth_xmit ready again\n",
+                              dev->name);
+#endif
+                       netif_wake_queue(dev);
+               }
+       }
+
+#if DEBUG_TX
+       printk(KERN_DEBUG "%s: eth_xmit end\n", dev->name);
+#endif
+       return NETDEV_TX_OK;
+}
+
+
+static struct net_device_stats *eth_stats(struct net_device *dev)
+{
+       struct port *port = netdev_priv(dev);
+       return &port->stat;
+}
+
+static void eth_set_mcast_list(struct net_device *dev)
+{
+       struct port *port = netdev_priv(dev);
+       struct dev_mc_list *mclist = dev->mc_list;
+       u8 diffs[ETH_ALEN], *addr;
+       int cnt = dev->mc_count, i;
+
+       if ((dev->flags & IFF_PROMISC) || !mclist || !cnt) {
+               __raw_writel(DEFAULT_RX_CNTRL0 & ~RX_CNTRL0_ADDR_FLTR_EN,
+                            &port->regs->rx_control[0]);
+               return;
+       }
+
+       memset(diffs, 0, ETH_ALEN);
+       addr = mclist->dmi_addr; /* first MAC address */
+
+       while (--cnt && (mclist = mclist->next))
+               for (i = 0; i < ETH_ALEN; i++)
+                       diffs[i] |= addr[i] ^ mclist->dmi_addr[i];
+
+       for (i = 0; i < ETH_ALEN; i++) {
+               __raw_writel(addr[i], &port->regs->mcast_addr[i]);
+               __raw_writel(~diffs[i], &port->regs->mcast_mask[i]);
+       }
+
+       __raw_writel(DEFAULT_RX_CNTRL0 | RX_CNTRL0_ADDR_FLTR_EN,
+                    &port->regs->rx_control[0]);
+}
+
+
+static int eth_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
+{
+       struct port *port = netdev_priv(dev);
+       unsigned int duplex_chg;
+       int err;
+
+       if (!netif_running(dev))
+               return -EINVAL;
+       err = generic_mii_ioctl(&port->mii, if_mii(req), cmd, &duplex_chg);
+       if (duplex_chg)
+               eth_set_duplex(port);
+       return err;
+}
+
+
+static int request_queues(struct port *port)
+{
+       int err;
+
+       err = qmgr_request_queue(RXFREE_QUEUE(port->id), RX_DESCS, 0, 0);
+       if (err)
+               return err;
+
+       err = qmgr_request_queue(port->plat->rxq, RX_DESCS, 0, 0);
+       if (err)
+               goto rel_rxfree;
+
+       err = qmgr_request_queue(TX_QUEUE(port->id), TX_DESCS, 0, 0);
+       if (err)
+               goto rel_rx;
+
+       err = qmgr_request_queue(port->plat->txreadyq, TX_DESCS, 0, 0);
+       if (err)
+               goto rel_tx;
+
+       /* TX-done queue handles skbs sent out by the NPEs */
+       if (!ports_open) {
+               err = qmgr_request_queue(TXDONE_QUEUE, TXDONE_QUEUE_LEN, 0, 0);
+               if (err)
+                       goto rel_txready;
+       }
+       return 0;
+
+rel_txready:
+       qmgr_release_queue(port->plat->txreadyq);
+rel_tx:
+       qmgr_release_queue(TX_QUEUE(port->id));
+rel_rx:
+       qmgr_release_queue(port->plat->rxq);
+rel_rxfree:
+       qmgr_release_queue(RXFREE_QUEUE(port->id));
+       printk(KERN_DEBUG "%s: unable to request hardware queues\n",
+              port->netdev->name);
+       return err;
+}
+
+static void release_queues(struct port *port)
+{
+       qmgr_release_queue(RXFREE_QUEUE(port->id));
+       qmgr_release_queue(port->plat->rxq);
+       qmgr_release_queue(TX_QUEUE(port->id));
+       qmgr_release_queue(port->plat->txreadyq);
+
+       if (!ports_open)
+               qmgr_release_queue(TXDONE_QUEUE);
+}
+
+static int init_queues(struct port *port)
+{
+       int i;
+
+       if (!ports_open)
+               if (!(dma_pool = dma_pool_create(DRV_NAME, NULL,
+                                                POOL_ALLOC_SIZE, 32, 0)))
+                       return -ENOMEM;
+
+       if (!(port->desc_tab = dma_pool_alloc(dma_pool, GFP_KERNEL,
+                                             &port->desc_tab_phys)))
+               return -ENOMEM;
+       memset(port->desc_tab, 0, POOL_ALLOC_SIZE);
+       memset(port->rx_buff_tab, 0, sizeof(port->rx_buff_tab)); /* tables */
+       memset(port->tx_buff_tab, 0, sizeof(port->tx_buff_tab));
+
+       /* Setup RX buffers */
+       for (i = 0; i < RX_DESCS; i++) {
+               struct desc *desc = rx_desc_ptr(port, i);
+               buffer_t *buff; /* skb or kmalloc()ated memory */
+               void *data;
+#ifdef __ARMEB__
+               if (!(buff = netdev_alloc_skb(port->netdev, RX_BUFF_SIZE)))
+                       return -ENOMEM;
+               data = buff->data;
+#else
+               if (!(buff = kmalloc(RX_BUFF_SIZE, GFP_KERNEL)))
+                       return -ENOMEM;
+               data = buff;
+#endif
+               desc->buf_len = MAX_MRU;
+               desc->data = dma_map_single(&port->netdev->dev, data,
+                                           RX_BUFF_SIZE, DMA_FROM_DEVICE);
+               if (dma_mapping_error(desc->data)) {
+                       free_buffer(buff);
+                       return -EIO;
+               }
+               desc->data += NET_IP_ALIGN;
+               port->rx_buff_tab[i] = buff;
+       }
+
+       return 0;
+}
+
+static void destroy_queues(struct port *port)
+{
+       int i;
+
+       if (port->desc_tab) {
+               for (i = 0; i < RX_DESCS; i++) {
+                       struct desc *desc = rx_desc_ptr(port, i);
+                       buffer_t *buff = port->rx_buff_tab[i];
+                       if (buff) {
+                               dma_unmap_single(&port->netdev->dev,
+                                                desc->data - NET_IP_ALIGN,
+                                                RX_BUFF_SIZE, DMA_FROM_DEVICE);
+                               free_buffer(buff);
+                       }
+               }
+               for (i = 0; i < TX_DESCS; i++) {
+                       struct desc *desc = tx_desc_ptr(port, i);
+                       buffer_t *buff = port->tx_buff_tab[i];
+                       if (buff) {
+                               dma_unmap_tx(port, desc);
+                               free_buffer(buff);
+                       }
+               }
+               dma_pool_free(dma_pool, port->desc_tab, port->desc_tab_phys);
+               port->desc_tab = NULL;
+       }
+
+       if (!ports_open && dma_pool) {
+               dma_pool_destroy(dma_pool);
+               dma_pool = NULL;
+       }
+}
+
+static int eth_open(struct net_device *dev)
+{
+       struct port *port = netdev_priv(dev);
+       struct npe *npe = port->npe;
+       struct msg msg;
+       int i, err;
+
+       if (!npe_running(npe)) {
+               err = npe_load_firmware(npe, npe_name(npe), &dev->dev);
+               if (err)
+                       return err;
+
+               if (npe_recv_message(npe, &msg, "ETH_GET_STATUS")) {
+                       printk(KERN_ERR "%s: %s not responding\n", dev->name,
+                              npe_name(npe));
+                       return -EIO;
+               }
+       }
+
+       mdio_write(dev, port->plat->phy, MII_BMCR, port->mii_bmcr);
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd = NPE_VLAN_SETRXQOSENTRY;
+       msg.eth_id = port->id;
+       msg.byte5 = port->plat->rxq | 0x80;
+       msg.byte7 = port->plat->rxq << 4;
+       for (i = 0; i < 8; i++) {
+               msg.byte3 = i;
+               if (npe_send_recv_message(port->npe, &msg, "ETH_SET_RXQ"))
+                       return -EIO;
+       }
+
+       msg.cmd = NPE_EDB_SETPORTADDRESS;
+       msg.eth_id = PHYSICAL_ID(port->id);
+       msg.byte2 = dev->dev_addr[0];
+       msg.byte3 = dev->dev_addr[1];
+       msg.byte4 = dev->dev_addr[2];
+       msg.byte5 = dev->dev_addr[3];
+       msg.byte6 = dev->dev_addr[4];
+       msg.byte7 = dev->dev_addr[5];
+       if (npe_send_recv_message(port->npe, &msg, "ETH_SET_MAC"))
+               return -EIO;
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd = NPE_FW_SETFIREWALLMODE;
+       msg.eth_id = port->id;
+       if (npe_send_recv_message(port->npe, &msg, "ETH_SET_FIREWALL_MODE"))
+               return -EIO;
+
+       if ((err = request_queues(port)) != 0)
+               return err;
+
+       if ((err = init_queues(port)) != 0) {
+               destroy_queues(port);
+               release_queues(port);
+               return err;
+       }
+
+       for (i = 0; i < ETH_ALEN; i++)
+               __raw_writel(dev->dev_addr[i], &port->regs->hw_addr[i]);
+       __raw_writel(0x08, &port->regs->random_seed);
+       __raw_writel(0x12, &port->regs->partial_empty_threshold);
+       __raw_writel(0x30, &port->regs->partial_full_threshold);
+       __raw_writel(0x08, &port->regs->tx_start_bytes);
+       __raw_writel(0x15, &port->regs->tx_deferral);
+       __raw_writel(0x08, &port->regs->tx_2part_deferral[0]);
+       __raw_writel(0x07, &port->regs->tx_2part_deferral[1]);
+       __raw_writel(0x80, &port->regs->slot_time);
+       __raw_writel(0x01, &port->regs->int_clock_threshold);
+
+       /* Populate queues with buffers, no failure after this point */
+       for (i = 0; i < TX_DESCS; i++)
+               queue_put_desc(port->plat->txreadyq,
+                              tx_desc_phys(port, i), tx_desc_ptr(port, i));
+
+       for (i = 0; i < RX_DESCS; i++)
+               queue_put_desc(RXFREE_QUEUE(port->id),
+                              rx_desc_phys(port, i), rx_desc_ptr(port, i));
+
+       __raw_writel(TX_CNTRL1_RETRIES, &port->regs->tx_control[1]);
+       __raw_writel(DEFAULT_TX_CNTRL0, &port->regs->tx_control[0]);
+       __raw_writel(0, &port->regs->rx_control[1]);
+       __raw_writel(DEFAULT_RX_CNTRL0, &port->regs->rx_control[0]);
+
+       napi_enable(&port->napi);
+       phy_check_media(port, 1);
+       eth_set_mcast_list(dev);
+       netif_start_queue(dev);
+       schedule_delayed_work(&port->mdio_thread, MDIO_INTERVAL);
+
+       qmgr_set_irq(port->plat->rxq, QUEUE_IRQ_SRC_NOT_EMPTY,
+                    eth_rx_irq, dev);
+       if (!ports_open) {
+               qmgr_set_irq(TXDONE_QUEUE, QUEUE_IRQ_SRC_NOT_EMPTY,
+                            eth_txdone_irq, NULL);
+               qmgr_enable_irq(TXDONE_QUEUE);
+       }
+       ports_open++;
+       /* we may already have RX data, enables IRQ */
+       netif_rx_schedule(dev, &port->napi);
+       return 0;
+}
+
+static int eth_close(struct net_device *dev)
+{
+       struct port *port = netdev_priv(dev);
+       struct msg msg;
+       int buffs = RX_DESCS; /* allocated RX buffers */
+       int i;
+
+       ports_open--;
+       qmgr_disable_irq(port->plat->rxq);
+       napi_disable(&port->napi);
+       netif_stop_queue(dev);
+
+       while (queue_get_desc(RXFREE_QUEUE(port->id), port, 0) >= 0)
+               buffs--;
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd = NPE_SETLOOPBACK_MODE;
+       msg.eth_id = port->id;
+       msg.byte3 = 1;
+       if (npe_send_recv_message(port->npe, &msg, "ETH_ENABLE_LOOPBACK"))
+               printk(KERN_CRIT "%s: unable to enable loopback\n", dev->name);
+
+       i = 0;
+       do {                    /* drain RX buffers */
+               while (queue_get_desc(port->plat->rxq, port, 0) >= 0)
+                       buffs--;
+               if (!buffs)
+                       break;
+               if (qmgr_stat_empty(TX_QUEUE(port->id))) {
+                       /* we have to inject some packet */
+                       struct desc *desc;
+                       u32 phys;
+                       int n = queue_get_desc(port->plat->txreadyq, port, 1);
+                       BUG_ON(n < 0);
+                       desc = tx_desc_ptr(port, n);
+                       phys = tx_desc_phys(port, n);
+                       desc->buf_len = desc->pkt_len = 1;
+                       wmb();
+                       queue_put_desc(TX_QUEUE(port->id), phys, desc);
+               }
+               udelay(1);
+       } while (++i < MAX_CLOSE_WAIT);
+
+       if (buffs)
+               printk(KERN_CRIT "%s: unable to drain RX queue, %i buffer(s)"
+                      " left in NPE\n", dev->name, buffs);
+#if DEBUG_CLOSE
+       if (!buffs)
+               printk(KERN_DEBUG "Draining RX queue took %i cycles\n", i);
+#endif
+
+       buffs = TX_DESCS;
+       while (queue_get_desc(TX_QUEUE(port->id), port, 1) >= 0)
+               buffs--; /* cancel TX */
+
+       i = 0;
+       do {
+               while (queue_get_desc(port->plat->txreadyq, port, 1) >= 0)
+                       buffs--;
+               if (!buffs)
+                       break;
+       } while (++i < MAX_CLOSE_WAIT);
+
+       if (buffs)
+               printk(KERN_CRIT "%s: unable to drain TX queue, %i buffer(s) "
+                      "left in NPE\n", dev->name, buffs);
+#if DEBUG_CLOSE
+       if (!buffs)
+               printk(KERN_DEBUG "Draining TX queues took %i cycles\n", i);
+#endif
+
+       msg.byte3 = 0;
+       if (npe_send_recv_message(port->npe, &msg, "ETH_DISABLE_LOOPBACK"))
+               printk(KERN_CRIT "%s: unable to disable loopback\n",
+                      dev->name);
+
+       port->mii_bmcr = mdio_read(dev, port->plat->phy, MII_BMCR) &
+               ~(BMCR_RESET | BMCR_PDOWN); /* may have been altered */
+       mdio_write(dev, port->plat->phy, MII_BMCR,
+                  port->mii_bmcr | BMCR_PDOWN);
+
+       if (!ports_open)
+               qmgr_disable_irq(TXDONE_QUEUE);
+       cancel_rearming_delayed_work(&port->mdio_thread);
+       destroy_queues(port);
+       release_queues(port);
+       return 0;
+}
+
+static int __devinit eth_init_one(struct platform_device *pdev)
+{
+       struct port *port;
+       struct net_device *dev;
+       struct eth_plat_info *plat = pdev->dev.platform_data;
+       u32 regs_phys;
+       int err;
+
+       if (!(dev = alloc_etherdev(sizeof(struct port))))
+               return -ENOMEM;
+
+       SET_NETDEV_DEV(dev, &pdev->dev);
+       port = netdev_priv(dev);
+       port->netdev = dev;
+       port->id = pdev->id;
+
+       switch (port->id) {
+       case IXP4XX_ETH_NPEA:
+               port->regs = (struct eth_regs __iomem *)IXP4XX_EthA_BASE_VIRT;
+               regs_phys  = IXP4XX_EthA_BASE_PHYS;
+               break;
+       case IXP4XX_ETH_NPEB:
+               port->regs = (struct eth_regs __iomem *)IXP4XX_EthB_BASE_VIRT;
+               regs_phys  = IXP4XX_EthB_BASE_PHYS;
+               break;
+       case IXP4XX_ETH_NPEC:
+               port->regs = (struct eth_regs __iomem *)IXP4XX_EthC_BASE_VIRT;
+               regs_phys  = IXP4XX_EthC_BASE_PHYS;
+               break;
+       default:
+               err = -ENOSYS;
+               goto err_free;
+       }
+
+       dev->open = eth_open;
+       dev->hard_start_xmit = eth_xmit;
+       dev->stop = eth_close;
+       dev->get_stats = eth_stats;
+       dev->do_ioctl = eth_ioctl;
+       dev->set_multicast_list = eth_set_mcast_list;
+       dev->tx_queue_len = 100;
+
+       netif_napi_add(dev, &port->napi, eth_poll, NAPI_WEIGHT);
+
+       if (!(port->npe = npe_request(NPE_ID(port->id)))) {
+               err = -EIO;
+               goto err_free;
+       }
+
+       if (register_netdev(dev)) {
+               err = -EIO;
+               goto err_npe_rel;
+       }
+
+       port->mem_res = request_mem_region(regs_phys, REGS_SIZE, dev->name);
+       if (!port->mem_res) {
+               err = -EBUSY;
+               goto err_unreg;
+       }
+
+       port->plat = plat;
+       npe_port_tab[NPE_ID(port->id)] = port;
+       memcpy(dev->dev_addr, plat->hwaddr, ETH_ALEN);
+
+       platform_set_drvdata(pdev, dev);
+
+       __raw_writel(DEFAULT_CORE_CNTRL | CORE_RESET,
+                    &port->regs->core_control);
+       udelay(50);
+       __raw_writel(DEFAULT_CORE_CNTRL, &port->regs->core_control);
+       udelay(50);
+
+       port->mii.dev = dev;
+       port->mii.mdio_read = mdio_read;
+       port->mii.mdio_write = mdio_write;
+       port->mii.phy_id = plat->phy;
+       port->mii.phy_id_mask = 0x1F;
+       port->mii.reg_num_mask = 0x1F;
+
+       printk(KERN_INFO "%s: MII PHY %i on %s\n", dev->name, plat->phy,
+              npe_name(port->npe));
+
+       phy_reset(dev, plat->phy);
+       port->mii_bmcr = mdio_read(dev, plat->phy, MII_BMCR) &
+               ~(BMCR_RESET | BMCR_PDOWN);
+       mdio_write(dev, plat->phy, MII_BMCR, port->mii_bmcr | BMCR_PDOWN);
+
+       INIT_DELAYED_WORK(&port->mdio_thread, mdio_thread);
+       return 0;
+
+err_unreg:
+       unregister_netdev(dev);
+err_npe_rel:
+       npe_release(port->npe);
+err_free:
+       free_netdev(dev);
+       return err;
+}
+
+static int __devexit eth_remove_one(struct platform_device *pdev)
+{
+       struct net_device *dev = platform_get_drvdata(pdev);
+       struct port *port = netdev_priv(dev);
+
+       unregister_netdev(dev);
+       npe_port_tab[NPE_ID(port->id)] = NULL;
+       platform_set_drvdata(pdev, NULL);
+       npe_release(port->npe);
+       release_resource(port->mem_res);
+       free_netdev(dev);
+       return 0;
+}
+
+static struct platform_driver drv = {
+       .driver.name    = DRV_NAME,
+       .probe          = eth_init_one,
+       .remove         = eth_remove_one,
+};
+
+static int __init eth_init_module(void)
+{
+       if (!(ixp4xx_read_feature_bits() & IXP4XX_FEATURE_NPEB_ETH0))
+               return -ENOSYS;
+
+       /* All MII PHY accesses use NPE-B Ethernet registers */
+       spin_lock_init(&mdio_lock);
+       mdio_regs = (struct eth_regs __iomem *)IXP4XX_EthB_BASE_VIRT;
+       __raw_writel(DEFAULT_CORE_CNTRL, &mdio_regs->core_control);
+
+       return platform_driver_register(&drv);
+}
+
+static void __exit eth_cleanup_module(void)
+{
+       platform_driver_unregister(&drv);
+}
+
+MODULE_AUTHOR("Krzysztof Halasa");
+MODULE_DESCRIPTION("Intel IXP4xx Ethernet driver");
+MODULE_LICENSE("GPL v2");
+MODULE_ALIAS("platform:ixp4xx_eth");
+module_init(eth_init_module);
+module_exit(eth_cleanup_module);