Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / ethernet / smsc / smc91x.c
diff --git a/kernel/drivers/net/ethernet/smsc/smc91x.c b/kernel/drivers/net/ethernet/smsc/smc91x.c
new file mode 100644 (file)
index 0000000..630f0b7
--- /dev/null
@@ -0,0 +1,2476 @@
+/*
+ * smc91x.c
+ * This is a driver for SMSC's 91C9x/91C1xx single-chip Ethernet devices.
+ *
+ * Copyright (C) 1996 by Erik Stahlman
+ * Copyright (C) 2001 Standard Microsystems Corporation
+ *     Developed by Simple Network Magic Corporation
+ * Copyright (C) 2003 Monta Vista Software, Inc.
+ *     Unified SMC91x driver by Nicolas Pitre
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, see <http://www.gnu.org/licenses/>.
+ *
+ * Arguments:
+ *     io      = for the base address
+ *     irq     = for the IRQ
+ *     nowait  = 0 for normal wait states, 1 eliminates additional wait states
+ *
+ * original author:
+ *     Erik Stahlman <erik@vt.edu>
+ *
+ * hardware multicast code:
+ *    Peter Cammaert <pc@denkart.be>
+ *
+ * contributors:
+ *     Daris A Nevil <dnevil@snmc.com>
+ *      Nicolas Pitre <nico@fluxnic.net>
+ *     Russell King <rmk@arm.linux.org.uk>
+ *
+ * History:
+ *   08/20/00  Arnaldo Melo       fix kfree(skb) in smc_hardware_send_packet
+ *   12/15/00  Christian Jullien  fix "Warning: kfree_skb on hard IRQ"
+ *   03/16/01  Daris A Nevil      modified smc9194.c for use with LAN91C111
+ *   08/22/01  Scott Anderson     merge changes from smc9194 to smc91111
+ *   08/21/01  Pramod B Bhardwaj  added support for RevB of LAN91C111
+ *   12/20/01  Jeff Sutherland    initial port to Xscale PXA with DMA support
+ *   04/07/03  Nicolas Pitre      unified SMC91x driver, killed irq races,
+ *                                more bus abstraction, big cleanup, etc.
+ *   29/09/03  Russell King       - add driver model support
+ *                                - ethtool support
+ *                                - convert to use generic MII interface
+ *                                - add link up/down notification
+ *                                - don't try to handle full negotiation in
+ *                                  smc_phy_configure
+ *                                - clean up (and fix stack overrun) in PHY
+ *                                  MII read/write functions
+ *   22/09/04  Nicolas Pitre      big update (see commit log for details)
+ */
+static const char version[] =
+       "smc91x.c: v1.1, sep 22 2004 by Nicolas Pitre <nico@fluxnic.net>";
+
+/* Debugging level */
+#ifndef SMC_DEBUG
+#define SMC_DEBUG              0
+#endif
+
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/delay.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/errno.h>
+#include <linux/ioport.h>
+#include <linux/crc32.h>
+#include <linux/platform_device.h>
+#include <linux/spinlock.h>
+#include <linux/ethtool.h>
+#include <linux/mii.h>
+#include <linux/workqueue.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/of_gpio.h>
+
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/skbuff.h>
+
+#include <asm/io.h>
+
+#include "smc91x.h"
+
+#if defined(CONFIG_ASSABET_NEPONSET)
+#include <mach/assabet.h>
+#include <mach/neponset.h>
+#endif
+
+#ifndef SMC_NOWAIT
+# define SMC_NOWAIT            0
+#endif
+static int nowait = SMC_NOWAIT;
+module_param(nowait, int, 0400);
+MODULE_PARM_DESC(nowait, "set to 1 for no wait state");
+
+/*
+ * Transmit timeout, default 5 seconds.
+ */
+static int watchdog = 1000;
+module_param(watchdog, int, 0400);
+MODULE_PARM_DESC(watchdog, "transmit timeout in milliseconds");
+
+MODULE_LICENSE("GPL");
+MODULE_ALIAS("platform:smc91x");
+
+/*
+ * The internal workings of the driver.  If you are changing anything
+ * here with the SMC stuff, you should have the datasheet and know
+ * what you are doing.
+ */
+#define CARDNAME "smc91x"
+
+/*
+ * Use power-down feature of the chip
+ */
+#define POWER_DOWN             1
+
+/*
+ * Wait time for memory to be free.  This probably shouldn't be
+ * tuned that much, as waiting for this means nothing else happens
+ * in the system
+ */
+#define MEMORY_WAIT_TIME       16
+
+/*
+ * The maximum number of processing loops allowed for each call to the
+ * IRQ handler.
+ */
+#define MAX_IRQ_LOOPS          8
+
+/*
+ * This selects whether TX packets are sent one by one to the SMC91x internal
+ * memory and throttled until transmission completes.  This may prevent
+ * RX overruns a litle by keeping much of the memory free for RX packets
+ * but to the expense of reduced TX throughput and increased IRQ overhead.
+ * Note this is not a cure for a too slow data bus or too high IRQ latency.
+ */
+#define THROTTLE_TX_PKTS       0
+
+/*
+ * The MII clock high/low times.  2x this number gives the MII clock period
+ * in microseconds. (was 50, but this gives 6.4ms for each MII transaction!)
+ */
+#define MII_DELAY              1
+
+#define DBG(n, dev, fmt, ...)                                  \
+       do {                                                    \
+               if (SMC_DEBUG >= (n))                           \
+                       netdev_dbg(dev, fmt, ##__VA_ARGS__);    \
+       } while (0)
+
+#define PRINTK(dev, fmt, ...)                                  \
+       do {                                                    \
+               if (SMC_DEBUG > 0)                              \
+                       netdev_info(dev, fmt, ##__VA_ARGS__);   \
+               else                                            \
+                       netdev_dbg(dev, fmt, ##__VA_ARGS__);    \
+       } while (0)
+
+#if SMC_DEBUG > 3
+static void PRINT_PKT(u_char *buf, int length)
+{
+       int i;
+       int remainder;
+       int lines;
+
+       lines = length / 16;
+       remainder = length % 16;
+
+       for (i = 0; i < lines ; i ++) {
+               int cur;
+               printk(KERN_DEBUG);
+               for (cur = 0; cur < 8; cur++) {
+                       u_char a, b;
+                       a = *buf++;
+                       b = *buf++;
+                       pr_cont("%02x%02x ", a, b);
+               }
+               pr_cont("\n");
+       }
+       printk(KERN_DEBUG);
+       for (i = 0; i < remainder/2 ; i++) {
+               u_char a, b;
+               a = *buf++;
+               b = *buf++;
+               pr_cont("%02x%02x ", a, b);
+       }
+       pr_cont("\n");
+}
+#else
+static inline void PRINT_PKT(u_char *buf, int length) { }
+#endif
+
+
+/* this enables an interrupt in the interrupt mask register */
+#define SMC_ENABLE_INT(lp, x) do {                                     \
+       unsigned char mask;                                             \
+       unsigned long smc_enable_flags;                                 \
+       spin_lock_irqsave(&lp->lock, smc_enable_flags);                 \
+       mask = SMC_GET_INT_MASK(lp);                                    \
+       mask |= (x);                                                    \
+       SMC_SET_INT_MASK(lp, mask);                                     \
+       spin_unlock_irqrestore(&lp->lock, smc_enable_flags);            \
+} while (0)
+
+/* this disables an interrupt from the interrupt mask register */
+#define SMC_DISABLE_INT(lp, x) do {                                    \
+       unsigned char mask;                                             \
+       unsigned long smc_disable_flags;                                \
+       spin_lock_irqsave(&lp->lock, smc_disable_flags);                \
+       mask = SMC_GET_INT_MASK(lp);                                    \
+       mask &= ~(x);                                                   \
+       SMC_SET_INT_MASK(lp, mask);                                     \
+       spin_unlock_irqrestore(&lp->lock, smc_disable_flags);           \
+} while (0)
+
+/*
+ * Wait while MMU is busy.  This is usually in the order of a few nanosecs
+ * if at all, but let's avoid deadlocking the system if the hardware
+ * decides to go south.
+ */
+#define SMC_WAIT_MMU_BUSY(lp) do {                                     \
+       if (unlikely(SMC_GET_MMU_CMD(lp) & MC_BUSY)) {          \
+               unsigned long timeout = jiffies + 2;                    \
+               while (SMC_GET_MMU_CMD(lp) & MC_BUSY) {         \
+                       if (time_after(jiffies, timeout)) {             \
+                               netdev_dbg(dev, "timeout %s line %d\n", \
+                                          __FILE__, __LINE__);         \
+                               break;                                  \
+                       }                                               \
+                       cpu_relax();                                    \
+               }                                                       \
+       }                                                               \
+} while (0)
+
+
+/*
+ * this does a soft reset on the device
+ */
+static void smc_reset(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned int ctl, cfg;
+       struct sk_buff *pending_skb;
+
+       DBG(2, dev, "%s\n", __func__);
+
+       /* Disable all interrupts, block TX tasklet */
+       spin_lock_irq(&lp->lock);
+       SMC_SELECT_BANK(lp, 2);
+       SMC_SET_INT_MASK(lp, 0);
+       pending_skb = lp->pending_tx_skb;
+       lp->pending_tx_skb = NULL;
+       spin_unlock_irq(&lp->lock);
+
+       /* free any pending tx skb */
+       if (pending_skb) {
+               dev_kfree_skb(pending_skb);
+               dev->stats.tx_errors++;
+               dev->stats.tx_aborted_errors++;
+       }
+
+       /*
+        * This resets the registers mostly to defaults, but doesn't
+        * affect EEPROM.  That seems unnecessary
+        */
+       SMC_SELECT_BANK(lp, 0);
+       SMC_SET_RCR(lp, RCR_SOFTRST);
+
+       /*
+        * Setup the Configuration Register
+        * This is necessary because the CONFIG_REG is not affected
+        * by a soft reset
+        */
+       SMC_SELECT_BANK(lp, 1);
+
+       cfg = CONFIG_DEFAULT;
+
+       /*
+        * Setup for fast accesses if requested.  If the card/system
+        * can't handle it then there will be no recovery except for
+        * a hard reset or power cycle
+        */
+       if (lp->cfg.flags & SMC91X_NOWAIT)
+               cfg |= CONFIG_NO_WAIT;
+
+       /*
+        * Release from possible power-down state
+        * Configuration register is not affected by Soft Reset
+        */
+       cfg |= CONFIG_EPH_POWER_EN;
+
+       SMC_SET_CONFIG(lp, cfg);
+
+       /* this should pause enough for the chip to be happy */
+       /*
+        * elaborate?  What does the chip _need_? --jgarzik
+        *
+        * This seems to be undocumented, but something the original
+        * driver(s) have always done.  Suspect undocumented timing
+        * info/determined empirically. --rmk
+        */
+       udelay(1);
+
+       /* Disable transmit and receive functionality */
+       SMC_SELECT_BANK(lp, 0);
+       SMC_SET_RCR(lp, RCR_CLEAR);
+       SMC_SET_TCR(lp, TCR_CLEAR);
+
+       SMC_SELECT_BANK(lp, 1);
+       ctl = SMC_GET_CTL(lp) | CTL_LE_ENABLE;
+
+       /*
+        * Set the control register to automatically release successfully
+        * transmitted packets, to make the best use out of our limited
+        * memory
+        */
+       if(!THROTTLE_TX_PKTS)
+               ctl |= CTL_AUTO_RELEASE;
+       else
+               ctl &= ~CTL_AUTO_RELEASE;
+       SMC_SET_CTL(lp, ctl);
+
+       /* Reset the MMU */
+       SMC_SELECT_BANK(lp, 2);
+       SMC_SET_MMU_CMD(lp, MC_RESET);
+       SMC_WAIT_MMU_BUSY(lp);
+}
+
+/*
+ * Enable Interrupts, Receive, and Transmit
+ */
+static void smc_enable(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       int mask;
+
+       DBG(2, dev, "%s\n", __func__);
+
+       /* see the header file for options in TCR/RCR DEFAULT */
+       SMC_SELECT_BANK(lp, 0);
+       SMC_SET_TCR(lp, lp->tcr_cur_mode);
+       SMC_SET_RCR(lp, lp->rcr_cur_mode);
+
+       SMC_SELECT_BANK(lp, 1);
+       SMC_SET_MAC_ADDR(lp, dev->dev_addr);
+
+       /* now, enable interrupts */
+       mask = IM_EPH_INT|IM_RX_OVRN_INT|IM_RCV_INT;
+       if (lp->version >= (CHIP_91100 << 4))
+               mask |= IM_MDINT;
+       SMC_SELECT_BANK(lp, 2);
+       SMC_SET_INT_MASK(lp, mask);
+
+       /*
+        * From this point the register bank must _NOT_ be switched away
+        * to something else than bank 2 without proper locking against
+        * races with any tasklet or interrupt handlers until smc_shutdown()
+        * or smc_reset() is called.
+        */
+}
+
+/*
+ * this puts the device in an inactive state
+ */
+static void smc_shutdown(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       struct sk_buff *pending_skb;
+
+       DBG(2, dev, "%s: %s\n", CARDNAME, __func__);
+
+       /* no more interrupts for me */
+       spin_lock_irq(&lp->lock);
+       SMC_SELECT_BANK(lp, 2);
+       SMC_SET_INT_MASK(lp, 0);
+       pending_skb = lp->pending_tx_skb;
+       lp->pending_tx_skb = NULL;
+       spin_unlock_irq(&lp->lock);
+       if (pending_skb)
+               dev_kfree_skb(pending_skb);
+
+       /* and tell the card to stay away from that nasty outside world */
+       SMC_SELECT_BANK(lp, 0);
+       SMC_SET_RCR(lp, RCR_CLEAR);
+       SMC_SET_TCR(lp, TCR_CLEAR);
+
+#ifdef POWER_DOWN
+       /* finally, shut the chip down */
+       SMC_SELECT_BANK(lp, 1);
+       SMC_SET_CONFIG(lp, SMC_GET_CONFIG(lp) & ~CONFIG_EPH_POWER_EN);
+#endif
+}
+
+/*
+ * This is the procedure to handle the receipt of a packet.
+ */
+static inline void  smc_rcv(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned int packet_number, status, packet_len;
+
+       DBG(3, dev, "%s\n", __func__);
+
+       packet_number = SMC_GET_RXFIFO(lp);
+       if (unlikely(packet_number & RXFIFO_REMPTY)) {
+               PRINTK(dev, "smc_rcv with nothing on FIFO.\n");
+               return;
+       }
+
+       /* read from start of packet */
+       SMC_SET_PTR(lp, PTR_READ | PTR_RCV | PTR_AUTOINC);
+
+       /* First two words are status and packet length */
+       SMC_GET_PKT_HDR(lp, status, packet_len);
+       packet_len &= 0x07ff;  /* mask off top bits */
+       DBG(2, dev, "RX PNR 0x%x STATUS 0x%04x LENGTH 0x%04x (%d)\n",
+           packet_number, status, packet_len, packet_len);
+
+       back:
+       if (unlikely(packet_len < 6 || status & RS_ERRORS)) {
+               if (status & RS_TOOLONG && packet_len <= (1514 + 4 + 6)) {
+                       /* accept VLAN packets */
+                       status &= ~RS_TOOLONG;
+                       goto back;
+               }
+               if (packet_len < 6) {
+                       /* bloody hardware */
+                       netdev_err(dev, "fubar (rxlen %u status %x\n",
+                                  packet_len, status);
+                       status |= RS_TOOSHORT;
+               }
+               SMC_WAIT_MMU_BUSY(lp);
+               SMC_SET_MMU_CMD(lp, MC_RELEASE);
+               dev->stats.rx_errors++;
+               if (status & RS_ALGNERR)
+                       dev->stats.rx_frame_errors++;
+               if (status & (RS_TOOSHORT | RS_TOOLONG))
+                       dev->stats.rx_length_errors++;
+               if (status & RS_BADCRC)
+                       dev->stats.rx_crc_errors++;
+       } else {
+               struct sk_buff *skb;
+               unsigned char *data;
+               unsigned int data_len;
+
+               /* set multicast stats */
+               if (status & RS_MULTICAST)
+                       dev->stats.multicast++;
+
+               /*
+                * Actual payload is packet_len - 6 (or 5 if odd byte).
+                * We want skb_reserve(2) and the final ctrl word
+                * (2 bytes, possibly containing the payload odd byte).
+                * Furthermore, we add 2 bytes to allow rounding up to
+                * multiple of 4 bytes on 32 bit buses.
+                * Hence packet_len - 6 + 2 + 2 + 2.
+                */
+               skb = netdev_alloc_skb(dev, packet_len);
+               if (unlikely(skb == NULL)) {
+                       SMC_WAIT_MMU_BUSY(lp);
+                       SMC_SET_MMU_CMD(lp, MC_RELEASE);
+                       dev->stats.rx_dropped++;
+                       return;
+               }
+
+               /* Align IP header to 32 bits */
+               skb_reserve(skb, 2);
+
+               /* BUG: the LAN91C111 rev A never sets this bit. Force it. */
+               if (lp->version == 0x90)
+                       status |= RS_ODDFRAME;
+
+               /*
+                * If odd length: packet_len - 5,
+                * otherwise packet_len - 6.
+                * With the trailing ctrl byte it's packet_len - 4.
+                */
+               data_len = packet_len - ((status & RS_ODDFRAME) ? 5 : 6);
+               data = skb_put(skb, data_len);
+               SMC_PULL_DATA(lp, data, packet_len - 4);
+
+               SMC_WAIT_MMU_BUSY(lp);
+               SMC_SET_MMU_CMD(lp, MC_RELEASE);
+
+               PRINT_PKT(data, packet_len - 4);
+
+               skb->protocol = eth_type_trans(skb, dev);
+               netif_rx(skb);
+               dev->stats.rx_packets++;
+               dev->stats.rx_bytes += data_len;
+       }
+}
+
+#ifdef CONFIG_SMP
+/*
+ * On SMP we have the following problem:
+ *
+ *     A = smc_hardware_send_pkt()
+ *     B = smc_hard_start_xmit()
+ *     C = smc_interrupt()
+ *
+ * A and B can never be executed simultaneously.  However, at least on UP,
+ * it is possible (and even desirable) for C to interrupt execution of
+ * A or B in order to have better RX reliability and avoid overruns.
+ * C, just like A and B, must have exclusive access to the chip and
+ * each of them must lock against any other concurrent access.
+ * Unfortunately this is not possible to have C suspend execution of A or
+ * B taking place on another CPU. On UP this is no an issue since A and B
+ * are run from softirq context and C from hard IRQ context, and there is
+ * no other CPU where concurrent access can happen.
+ * If ever there is a way to force at least B and C to always be executed
+ * on the same CPU then we could use read/write locks to protect against
+ * any other concurrent access and C would always interrupt B. But life
+ * isn't that easy in a SMP world...
+ */
+#define smc_special_trylock(lock, flags)                               \
+({                                                                     \
+       int __ret;                                                      \
+       local_irq_save(flags);                                          \
+       __ret = spin_trylock(lock);                                     \
+       if (!__ret)                                                     \
+               local_irq_restore(flags);                               \
+       __ret;                                                          \
+})
+#define smc_special_lock(lock, flags)          spin_lock_irqsave(lock, flags)
+#define smc_special_unlock(lock, flags)        spin_unlock_irqrestore(lock, flags)
+#else
+#define smc_special_trylock(lock, flags)       (flags == flags)
+#define smc_special_lock(lock, flags)          do { flags = 0; } while (0)
+#define smc_special_unlock(lock, flags)        do { flags = 0; } while (0)
+#endif
+
+/*
+ * This is called to actually send a packet to the chip.
+ */
+static void smc_hardware_send_pkt(unsigned long data)
+{
+       struct net_device *dev = (struct net_device *)data;
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       struct sk_buff *skb;
+       unsigned int packet_no, len;
+       unsigned char *buf;
+       unsigned long flags;
+
+       DBG(3, dev, "%s\n", __func__);
+
+       if (!smc_special_trylock(&lp->lock, flags)) {
+               netif_stop_queue(dev);
+               tasklet_schedule(&lp->tx_task);
+               return;
+       }
+
+       skb = lp->pending_tx_skb;
+       if (unlikely(!skb)) {
+               smc_special_unlock(&lp->lock, flags);
+               return;
+       }
+       lp->pending_tx_skb = NULL;
+
+       packet_no = SMC_GET_AR(lp);
+       if (unlikely(packet_no & AR_FAILED)) {
+               netdev_err(dev, "Memory allocation failed.\n");
+               dev->stats.tx_errors++;
+               dev->stats.tx_fifo_errors++;
+               smc_special_unlock(&lp->lock, flags);
+               goto done;
+       }
+
+       /* point to the beginning of the packet */
+       SMC_SET_PN(lp, packet_no);
+       SMC_SET_PTR(lp, PTR_AUTOINC);
+
+       buf = skb->data;
+       len = skb->len;
+       DBG(2, dev, "TX PNR 0x%x LENGTH 0x%04x (%d) BUF 0x%p\n",
+           packet_no, len, len, buf);
+       PRINT_PKT(buf, len);
+
+       /*
+        * Send the packet length (+6 for status words, length, and ctl.
+        * The card will pad to 64 bytes with zeroes if packet is too small.
+        */
+       SMC_PUT_PKT_HDR(lp, 0, len + 6);
+
+       /* send the actual data */
+       SMC_PUSH_DATA(lp, buf, len & ~1);
+
+       /* Send final ctl word with the last byte if there is one */
+       SMC_outw(((len & 1) ? (0x2000 | buf[len-1]) : 0), ioaddr, DATA_REG(lp));
+
+       /*
+        * If THROTTLE_TX_PKTS is set, we stop the queue here. This will
+        * have the effect of having at most one packet queued for TX
+        * in the chip's memory at all time.
+        *
+        * If THROTTLE_TX_PKTS is not set then the queue is stopped only
+        * when memory allocation (MC_ALLOC) does not succeed right away.
+        */
+       if (THROTTLE_TX_PKTS)
+               netif_stop_queue(dev);
+
+       /* queue the packet for TX */
+       SMC_SET_MMU_CMD(lp, MC_ENQUEUE);
+       smc_special_unlock(&lp->lock, flags);
+
+       dev->trans_start = jiffies;
+       dev->stats.tx_packets++;
+       dev->stats.tx_bytes += len;
+
+       SMC_ENABLE_INT(lp, IM_TX_INT | IM_TX_EMPTY_INT);
+
+done:  if (!THROTTLE_TX_PKTS)
+               netif_wake_queue(dev);
+
+       dev_consume_skb_any(skb);
+}
+
+/*
+ * Since I am not sure if I will have enough room in the chip's ram
+ * to store the packet, I call this routine which either sends it
+ * now, or set the card to generates an interrupt when ready
+ * for the packet.
+ */
+static int smc_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned int numPages, poll_count, status;
+       unsigned long flags;
+
+       DBG(3, dev, "%s\n", __func__);
+
+       BUG_ON(lp->pending_tx_skb != NULL);
+
+       /*
+        * The MMU wants the number of pages to be the number of 256 bytes
+        * 'pages', minus 1 (since a packet can't ever have 0 pages :))
+        *
+        * The 91C111 ignores the size bits, but earlier models don't.
+        *
+        * Pkt size for allocating is data length +6 (for additional status
+        * words, length and ctl)
+        *
+        * If odd size then last byte is included in ctl word.
+        */
+       numPages = ((skb->len & ~1) + (6 - 1)) >> 8;
+       if (unlikely(numPages > 7)) {
+               netdev_warn(dev, "Far too big packet error.\n");
+               dev->stats.tx_errors++;
+               dev->stats.tx_dropped++;
+               dev_kfree_skb_any(skb);
+               return NETDEV_TX_OK;
+       }
+
+       smc_special_lock(&lp->lock, flags);
+
+       /* now, try to allocate the memory */
+       SMC_SET_MMU_CMD(lp, MC_ALLOC | numPages);
+
+       /*
+        * Poll the chip for a short amount of time in case the
+        * allocation succeeds quickly.
+        */
+       poll_count = MEMORY_WAIT_TIME;
+       do {
+               status = SMC_GET_INT(lp);
+               if (status & IM_ALLOC_INT) {
+                       SMC_ACK_INT(lp, IM_ALLOC_INT);
+                       break;
+               }
+       } while (--poll_count);
+
+       smc_special_unlock(&lp->lock, flags);
+
+       lp->pending_tx_skb = skb;
+       if (!poll_count) {
+               /* oh well, wait until the chip finds memory later */
+               netif_stop_queue(dev);
+               DBG(2, dev, "TX memory allocation deferred.\n");
+               SMC_ENABLE_INT(lp, IM_ALLOC_INT);
+       } else {
+               /*
+                * Allocation succeeded: push packet to the chip's own memory
+                * immediately.
+                */
+               smc_hardware_send_pkt((unsigned long)dev);
+       }
+
+       return NETDEV_TX_OK;
+}
+
+/*
+ * This handles a TX interrupt, which is only called when:
+ * - a TX error occurred, or
+ * - CTL_AUTO_RELEASE is not set and TX of a packet completed.
+ */
+static void smc_tx(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned int saved_packet, packet_no, tx_status, pkt_len;
+
+       DBG(3, dev, "%s\n", __func__);
+
+       /* If the TX FIFO is empty then nothing to do */
+       packet_no = SMC_GET_TXFIFO(lp);
+       if (unlikely(packet_no & TXFIFO_TEMPTY)) {
+               PRINTK(dev, "smc_tx with nothing on FIFO.\n");
+               return;
+       }
+
+       /* select packet to read from */
+       saved_packet = SMC_GET_PN(lp);
+       SMC_SET_PN(lp, packet_no);
+
+       /* read the first word (status word) from this packet */
+       SMC_SET_PTR(lp, PTR_AUTOINC | PTR_READ);
+       SMC_GET_PKT_HDR(lp, tx_status, pkt_len);
+       DBG(2, dev, "TX STATUS 0x%04x PNR 0x%02x\n",
+           tx_status, packet_no);
+
+       if (!(tx_status & ES_TX_SUC))
+               dev->stats.tx_errors++;
+
+       if (tx_status & ES_LOSTCARR)
+               dev->stats.tx_carrier_errors++;
+
+       if (tx_status & (ES_LATCOL | ES_16COL)) {
+               PRINTK(dev, "%s occurred on last xmit\n",
+                      (tx_status & ES_LATCOL) ?
+                       "late collision" : "too many collisions");
+               dev->stats.tx_window_errors++;
+               if (!(dev->stats.tx_window_errors & 63) && net_ratelimit()) {
+                       netdev_info(dev, "unexpectedly large number of bad collisions. Please check duplex setting.\n");
+               }
+       }
+
+       /* kill the packet */
+       SMC_WAIT_MMU_BUSY(lp);
+       SMC_SET_MMU_CMD(lp, MC_FREEPKT);
+
+       /* Don't restore Packet Number Reg until busy bit is cleared */
+       SMC_WAIT_MMU_BUSY(lp);
+       SMC_SET_PN(lp, saved_packet);
+
+       /* re-enable transmit */
+       SMC_SELECT_BANK(lp, 0);
+       SMC_SET_TCR(lp, lp->tcr_cur_mode);
+       SMC_SELECT_BANK(lp, 2);
+}
+
+
+/*---PHY CONTROL AND CONFIGURATION-----------------------------------------*/
+
+static void smc_mii_out(struct net_device *dev, unsigned int val, int bits)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned int mii_reg, mask;
+
+       mii_reg = SMC_GET_MII(lp) & ~(MII_MCLK | MII_MDOE | MII_MDO);
+       mii_reg |= MII_MDOE;
+
+       for (mask = 1 << (bits - 1); mask; mask >>= 1) {
+               if (val & mask)
+                       mii_reg |= MII_MDO;
+               else
+                       mii_reg &= ~MII_MDO;
+
+               SMC_SET_MII(lp, mii_reg);
+               udelay(MII_DELAY);
+               SMC_SET_MII(lp, mii_reg | MII_MCLK);
+               udelay(MII_DELAY);
+       }
+}
+
+static unsigned int smc_mii_in(struct net_device *dev, int bits)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned int mii_reg, mask, val;
+
+       mii_reg = SMC_GET_MII(lp) & ~(MII_MCLK | MII_MDOE | MII_MDO);
+       SMC_SET_MII(lp, mii_reg);
+
+       for (mask = 1 << (bits - 1), val = 0; mask; mask >>= 1) {
+               if (SMC_GET_MII(lp) & MII_MDI)
+                       val |= mask;
+
+               SMC_SET_MII(lp, mii_reg);
+               udelay(MII_DELAY);
+               SMC_SET_MII(lp, mii_reg | MII_MCLK);
+               udelay(MII_DELAY);
+       }
+
+       return val;
+}
+
+/*
+ * Reads a register from the MII Management serial interface
+ */
+static int smc_phy_read(struct net_device *dev, int phyaddr, int phyreg)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned int phydata;
+
+       SMC_SELECT_BANK(lp, 3);
+
+       /* Idle - 32 ones */
+       smc_mii_out(dev, 0xffffffff, 32);
+
+       /* Start code (01) + read (10) + phyaddr + phyreg */
+       smc_mii_out(dev, 6 << 10 | phyaddr << 5 | phyreg, 14);
+
+       /* Turnaround (2bits) + phydata */
+       phydata = smc_mii_in(dev, 18);
+
+       /* Return to idle state */
+       SMC_SET_MII(lp, SMC_GET_MII(lp) & ~(MII_MCLK|MII_MDOE|MII_MDO));
+
+       DBG(3, dev, "%s: phyaddr=0x%x, phyreg=0x%x, phydata=0x%x\n",
+           __func__, phyaddr, phyreg, phydata);
+
+       SMC_SELECT_BANK(lp, 2);
+       return phydata;
+}
+
+/*
+ * Writes a register to the MII Management serial interface
+ */
+static void smc_phy_write(struct net_device *dev, int phyaddr, int phyreg,
+                         int phydata)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+
+       SMC_SELECT_BANK(lp, 3);
+
+       /* Idle - 32 ones */
+       smc_mii_out(dev, 0xffffffff, 32);
+
+       /* Start code (01) + write (01) + phyaddr + phyreg + turnaround + phydata */
+       smc_mii_out(dev, 5 << 28 | phyaddr << 23 | phyreg << 18 | 2 << 16 | phydata, 32);
+
+       /* Return to idle state */
+       SMC_SET_MII(lp, SMC_GET_MII(lp) & ~(MII_MCLK|MII_MDOE|MII_MDO));
+
+       DBG(3, dev, "%s: phyaddr=0x%x, phyreg=0x%x, phydata=0x%x\n",
+           __func__, phyaddr, phyreg, phydata);
+
+       SMC_SELECT_BANK(lp, 2);
+}
+
+/*
+ * Finds and reports the PHY address
+ */
+static void smc_phy_detect(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       int phyaddr;
+
+       DBG(2, dev, "%s\n", __func__);
+
+       lp->phy_type = 0;
+
+       /*
+        * Scan all 32 PHY addresses if necessary, starting at
+        * PHY#1 to PHY#31, and then PHY#0 last.
+        */
+       for (phyaddr = 1; phyaddr < 33; ++phyaddr) {
+               unsigned int id1, id2;
+
+               /* Read the PHY identifiers */
+               id1 = smc_phy_read(dev, phyaddr & 31, MII_PHYSID1);
+               id2 = smc_phy_read(dev, phyaddr & 31, MII_PHYSID2);
+
+               DBG(3, dev, "phy_id1=0x%x, phy_id2=0x%x\n",
+                   id1, id2);
+
+               /* Make sure it is a valid identifier */
+               if (id1 != 0x0000 && id1 != 0xffff && id1 != 0x8000 &&
+                   id2 != 0x0000 && id2 != 0xffff && id2 != 0x8000) {
+                       /* Save the PHY's address */
+                       lp->mii.phy_id = phyaddr & 31;
+                       lp->phy_type = id1 << 16 | id2;
+                       break;
+               }
+       }
+}
+
+/*
+ * Sets the PHY to a configuration as determined by the user
+ */
+static int smc_phy_fixed(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       int phyaddr = lp->mii.phy_id;
+       int bmcr, cfg1;
+
+       DBG(3, dev, "%s\n", __func__);
+
+       /* Enter Link Disable state */
+       cfg1 = smc_phy_read(dev, phyaddr, PHY_CFG1_REG);
+       cfg1 |= PHY_CFG1_LNKDIS;
+       smc_phy_write(dev, phyaddr, PHY_CFG1_REG, cfg1);
+
+       /*
+        * Set our fixed capabilities
+        * Disable auto-negotiation
+        */
+       bmcr = 0;
+
+       if (lp->ctl_rfduplx)
+               bmcr |= BMCR_FULLDPLX;
+
+       if (lp->ctl_rspeed == 100)
+               bmcr |= BMCR_SPEED100;
+
+       /* Write our capabilities to the phy control register */
+       smc_phy_write(dev, phyaddr, MII_BMCR, bmcr);
+
+       /* Re-Configure the Receive/Phy Control register */
+       SMC_SELECT_BANK(lp, 0);
+       SMC_SET_RPC(lp, lp->rpc_cur_mode);
+       SMC_SELECT_BANK(lp, 2);
+
+       return 1;
+}
+
+/**
+ * smc_phy_reset - reset the phy
+ * @dev: net device
+ * @phy: phy address
+ *
+ * Issue a software reset for the specified PHY and
+ * wait up to 100ms for the reset to complete.  We should
+ * not access the PHY for 50ms after issuing the reset.
+ *
+ * The time to wait appears to be dependent on the PHY.
+ *
+ * Must be called with lp->lock locked.
+ */
+static int smc_phy_reset(struct net_device *dev, int phy)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       unsigned int bmcr;
+       int timeout;
+
+       smc_phy_write(dev, phy, MII_BMCR, BMCR_RESET);
+
+       for (timeout = 2; timeout; timeout--) {
+               spin_unlock_irq(&lp->lock);
+               msleep(50);
+               spin_lock_irq(&lp->lock);
+
+               bmcr = smc_phy_read(dev, phy, MII_BMCR);
+               if (!(bmcr & BMCR_RESET))
+                       break;
+       }
+
+       return bmcr & BMCR_RESET;
+}
+
+/**
+ * smc_phy_powerdown - powerdown phy
+ * @dev: net device
+ *
+ * Power down the specified PHY
+ */
+static void smc_phy_powerdown(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       unsigned int bmcr;
+       int phy = lp->mii.phy_id;
+
+       if (lp->phy_type == 0)
+               return;
+
+       /* We need to ensure that no calls to smc_phy_configure are
+          pending.
+       */
+       cancel_work_sync(&lp->phy_configure);
+
+       bmcr = smc_phy_read(dev, phy, MII_BMCR);
+       smc_phy_write(dev, phy, MII_BMCR, bmcr | BMCR_PDOWN);
+}
+
+/**
+ * smc_phy_check_media - check the media status and adjust TCR
+ * @dev: net device
+ * @init: set true for initialisation
+ *
+ * Select duplex mode depending on negotiation state.  This
+ * also updates our carrier state.
+ */
+static void smc_phy_check_media(struct net_device *dev, int init)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+
+       if (mii_check_media(&lp->mii, netif_msg_link(lp), init)) {
+               /* duplex state has changed */
+               if (lp->mii.full_duplex) {
+                       lp->tcr_cur_mode |= TCR_SWFDUP;
+               } else {
+                       lp->tcr_cur_mode &= ~TCR_SWFDUP;
+               }
+
+               SMC_SELECT_BANK(lp, 0);
+               SMC_SET_TCR(lp, lp->tcr_cur_mode);
+       }
+}
+
+/*
+ * Configures the specified PHY through the MII management interface
+ * using Autonegotiation.
+ * Calls smc_phy_fixed() if the user has requested a certain config.
+ * If RPC ANEG bit is set, the media selection is dependent purely on
+ * the selection by the MII (either in the MII BMCR reg or the result
+ * of autonegotiation.)  If the RPC ANEG bit is cleared, the selection
+ * is controlled by the RPC SPEED and RPC DPLX bits.
+ */
+static void smc_phy_configure(struct work_struct *work)
+{
+       struct smc_local *lp =
+               container_of(work, struct smc_local, phy_configure);
+       struct net_device *dev = lp->dev;
+       void __iomem *ioaddr = lp->base;
+       int phyaddr = lp->mii.phy_id;
+       int my_phy_caps; /* My PHY capabilities */
+       int my_ad_caps; /* My Advertised capabilities */
+       int status;
+
+       DBG(3, dev, "smc_program_phy()\n");
+
+       spin_lock_irq(&lp->lock);
+
+       /*
+        * We should not be called if phy_type is zero.
+        */
+       if (lp->phy_type == 0)
+               goto smc_phy_configure_exit;
+
+       if (smc_phy_reset(dev, phyaddr)) {
+               netdev_info(dev, "PHY reset timed out\n");
+               goto smc_phy_configure_exit;
+       }
+
+       /*
+        * Enable PHY Interrupts (for register 18)
+        * Interrupts listed here are disabled
+        */
+       smc_phy_write(dev, phyaddr, PHY_MASK_REG,
+               PHY_INT_LOSSSYNC | PHY_INT_CWRD | PHY_INT_SSD |
+               PHY_INT_ESD | PHY_INT_RPOL | PHY_INT_JAB |
+               PHY_INT_SPDDET | PHY_INT_DPLXDET);
+
+       /* Configure the Receive/Phy Control register */
+       SMC_SELECT_BANK(lp, 0);
+       SMC_SET_RPC(lp, lp->rpc_cur_mode);
+
+       /* If the user requested no auto neg, then go set his request */
+       if (lp->mii.force_media) {
+               smc_phy_fixed(dev);
+               goto smc_phy_configure_exit;
+       }
+
+       /* Copy our capabilities from MII_BMSR to MII_ADVERTISE */
+       my_phy_caps = smc_phy_read(dev, phyaddr, MII_BMSR);
+
+       if (!(my_phy_caps & BMSR_ANEGCAPABLE)) {
+               netdev_info(dev, "Auto negotiation NOT supported\n");
+               smc_phy_fixed(dev);
+               goto smc_phy_configure_exit;
+       }
+
+       my_ad_caps = ADVERTISE_CSMA; /* I am CSMA capable */
+
+       if (my_phy_caps & BMSR_100BASE4)
+               my_ad_caps |= ADVERTISE_100BASE4;
+       if (my_phy_caps & BMSR_100FULL)
+               my_ad_caps |= ADVERTISE_100FULL;
+       if (my_phy_caps & BMSR_100HALF)
+               my_ad_caps |= ADVERTISE_100HALF;
+       if (my_phy_caps & BMSR_10FULL)
+               my_ad_caps |= ADVERTISE_10FULL;
+       if (my_phy_caps & BMSR_10HALF)
+               my_ad_caps |= ADVERTISE_10HALF;
+
+       /* Disable capabilities not selected by our user */
+       if (lp->ctl_rspeed != 100)
+               my_ad_caps &= ~(ADVERTISE_100BASE4|ADVERTISE_100FULL|ADVERTISE_100HALF);
+
+       if (!lp->ctl_rfduplx)
+               my_ad_caps &= ~(ADVERTISE_100FULL|ADVERTISE_10FULL);
+
+       /* Update our Auto-Neg Advertisement Register */
+       smc_phy_write(dev, phyaddr, MII_ADVERTISE, my_ad_caps);
+       lp->mii.advertising = my_ad_caps;
+
+       /*
+        * Read the register back.  Without this, it appears that when
+        * auto-negotiation is restarted, sometimes it isn't ready and
+        * the link does not come up.
+        */
+       status = smc_phy_read(dev, phyaddr, MII_ADVERTISE);
+
+       DBG(2, dev, "phy caps=%x\n", my_phy_caps);
+       DBG(2, dev, "phy advertised caps=%x\n", my_ad_caps);
+
+       /* Restart auto-negotiation process in order to advertise my caps */
+       smc_phy_write(dev, phyaddr, MII_BMCR, BMCR_ANENABLE | BMCR_ANRESTART);
+
+       smc_phy_check_media(dev, 1);
+
+smc_phy_configure_exit:
+       SMC_SELECT_BANK(lp, 2);
+       spin_unlock_irq(&lp->lock);
+}
+
+/*
+ * smc_phy_interrupt
+ *
+ * Purpose:  Handle interrupts relating to PHY register 18. This is
+ *  called from the "hard" interrupt handler under our private spinlock.
+ */
+static void smc_phy_interrupt(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       int phyaddr = lp->mii.phy_id;
+       int phy18;
+
+       DBG(2, dev, "%s\n", __func__);
+
+       if (lp->phy_type == 0)
+               return;
+
+       for(;;) {
+               smc_phy_check_media(dev, 0);
+
+               /* Read PHY Register 18, Status Output */
+               phy18 = smc_phy_read(dev, phyaddr, PHY_INT_REG);
+               if ((phy18 & PHY_INT_INT) == 0)
+                       break;
+       }
+}
+
+/*--- END PHY CONTROL AND CONFIGURATION-------------------------------------*/
+
+static void smc_10bt_check_media(struct net_device *dev, int init)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned int old_carrier, new_carrier;
+
+       old_carrier = netif_carrier_ok(dev) ? 1 : 0;
+
+       SMC_SELECT_BANK(lp, 0);
+       new_carrier = (SMC_GET_EPH_STATUS(lp) & ES_LINK_OK) ? 1 : 0;
+       SMC_SELECT_BANK(lp, 2);
+
+       if (init || (old_carrier != new_carrier)) {
+               if (!new_carrier) {
+                       netif_carrier_off(dev);
+               } else {
+                       netif_carrier_on(dev);
+               }
+               if (netif_msg_link(lp))
+                       netdev_info(dev, "link %s\n",
+                                   new_carrier ? "up" : "down");
+       }
+}
+
+static void smc_eph_interrupt(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned int ctl;
+
+       smc_10bt_check_media(dev, 0);
+
+       SMC_SELECT_BANK(lp, 1);
+       ctl = SMC_GET_CTL(lp);
+       SMC_SET_CTL(lp, ctl & ~CTL_LE_ENABLE);
+       SMC_SET_CTL(lp, ctl);
+       SMC_SELECT_BANK(lp, 2);
+}
+
+/*
+ * This is the main routine of the driver, to handle the device when
+ * it needs some attention.
+ */
+static irqreturn_t smc_interrupt(int irq, void *dev_id)
+{
+       struct net_device *dev = dev_id;
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       int status, mask, timeout, card_stats;
+       int saved_pointer;
+
+       DBG(3, dev, "%s\n", __func__);
+
+       spin_lock(&lp->lock);
+
+       /* A preamble may be used when there is a potential race
+        * between the interruptible transmit functions and this
+        * ISR. */
+       SMC_INTERRUPT_PREAMBLE;
+
+       saved_pointer = SMC_GET_PTR(lp);
+       mask = SMC_GET_INT_MASK(lp);
+       SMC_SET_INT_MASK(lp, 0);
+
+       /* set a timeout value, so I don't stay here forever */
+       timeout = MAX_IRQ_LOOPS;
+
+       do {
+               status = SMC_GET_INT(lp);
+
+               DBG(2, dev, "INT 0x%02x MASK 0x%02x MEM 0x%04x FIFO 0x%04x\n",
+                   status, mask,
+                   ({ int meminfo; SMC_SELECT_BANK(lp, 0);
+                      meminfo = SMC_GET_MIR(lp);
+                      SMC_SELECT_BANK(lp, 2); meminfo; }),
+                   SMC_GET_FIFO(lp));
+
+               status &= mask;
+               if (!status)
+                       break;
+
+               if (status & IM_TX_INT) {
+                       /* do this before RX as it will free memory quickly */
+                       DBG(3, dev, "TX int\n");
+                       smc_tx(dev);
+                       SMC_ACK_INT(lp, IM_TX_INT);
+                       if (THROTTLE_TX_PKTS)
+                               netif_wake_queue(dev);
+               } else if (status & IM_RCV_INT) {
+                       DBG(3, dev, "RX irq\n");
+                       smc_rcv(dev);
+               } else if (status & IM_ALLOC_INT) {
+                       DBG(3, dev, "Allocation irq\n");
+                       tasklet_hi_schedule(&lp->tx_task);
+                       mask &= ~IM_ALLOC_INT;
+               } else if (status & IM_TX_EMPTY_INT) {
+                       DBG(3, dev, "TX empty\n");
+                       mask &= ~IM_TX_EMPTY_INT;
+
+                       /* update stats */
+                       SMC_SELECT_BANK(lp, 0);
+                       card_stats = SMC_GET_COUNTER(lp);
+                       SMC_SELECT_BANK(lp, 2);
+
+                       /* single collisions */
+                       dev->stats.collisions += card_stats & 0xF;
+                       card_stats >>= 4;
+
+                       /* multiple collisions */
+                       dev->stats.collisions += card_stats & 0xF;
+               } else if (status & IM_RX_OVRN_INT) {
+                       DBG(1, dev, "RX overrun (EPH_ST 0x%04x)\n",
+                           ({ int eph_st; SMC_SELECT_BANK(lp, 0);
+                              eph_st = SMC_GET_EPH_STATUS(lp);
+                              SMC_SELECT_BANK(lp, 2); eph_st; }));
+                       SMC_ACK_INT(lp, IM_RX_OVRN_INT);
+                       dev->stats.rx_errors++;
+                       dev->stats.rx_fifo_errors++;
+               } else if (status & IM_EPH_INT) {
+                       smc_eph_interrupt(dev);
+               } else if (status & IM_MDINT) {
+                       SMC_ACK_INT(lp, IM_MDINT);
+                       smc_phy_interrupt(dev);
+               } else if (status & IM_ERCV_INT) {
+                       SMC_ACK_INT(lp, IM_ERCV_INT);
+                       PRINTK(dev, "UNSUPPORTED: ERCV INTERRUPT\n");
+               }
+       } while (--timeout);
+
+       /* restore register states */
+       SMC_SET_PTR(lp, saved_pointer);
+       SMC_SET_INT_MASK(lp, mask);
+       spin_unlock(&lp->lock);
+
+#ifndef CONFIG_NET_POLL_CONTROLLER
+       if (timeout == MAX_IRQ_LOOPS)
+               PRINTK(dev, "spurious interrupt (mask = 0x%02x)\n",
+                      mask);
+#endif
+       DBG(3, dev, "Interrupt done (%d loops)\n",
+           MAX_IRQ_LOOPS - timeout);
+
+       /*
+        * We return IRQ_HANDLED unconditionally here even if there was
+        * nothing to do.  There is a possibility that a packet might
+        * get enqueued into the chip right after TX_EMPTY_INT is raised
+        * but just before the CPU acknowledges the IRQ.
+        * Better take an unneeded IRQ in some occasions than complexifying
+        * the code for all cases.
+        */
+       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 smc_poll_controller(struct net_device *dev)
+{
+       disable_irq(dev->irq);
+       smc_interrupt(dev->irq, dev);
+       enable_irq(dev->irq);
+}
+#endif
+
+/* Our watchdog timed out. Called by the networking layer */
+static void smc_timeout(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       int status, mask, eph_st, meminfo, fifo;
+
+       DBG(2, dev, "%s\n", __func__);
+
+       spin_lock_irq(&lp->lock);
+       status = SMC_GET_INT(lp);
+       mask = SMC_GET_INT_MASK(lp);
+       fifo = SMC_GET_FIFO(lp);
+       SMC_SELECT_BANK(lp, 0);
+       eph_st = SMC_GET_EPH_STATUS(lp);
+       meminfo = SMC_GET_MIR(lp);
+       SMC_SELECT_BANK(lp, 2);
+       spin_unlock_irq(&lp->lock);
+       PRINTK(dev, "TX timeout (INT 0x%02x INTMASK 0x%02x MEM 0x%04x FIFO 0x%04x EPH_ST 0x%04x)\n",
+              status, mask, meminfo, fifo, eph_st);
+
+       smc_reset(dev);
+       smc_enable(dev);
+
+       /*
+        * Reconfiguring the PHY doesn't seem like a bad idea here, but
+        * smc_phy_configure() calls msleep() which calls schedule_timeout()
+        * which calls schedule().  Hence we use a work queue.
+        */
+       if (lp->phy_type != 0)
+               schedule_work(&lp->phy_configure);
+
+       /* We can accept TX packets again */
+       dev->trans_start = jiffies; /* prevent tx timeout */
+       netif_wake_queue(dev);
+}
+
+/*
+ * This routine will, depending on the values passed to it,
+ * either make it accept multicast packets, go into
+ * promiscuous mode (for TCPDUMP and cousins) or accept
+ * a select set of multicast packets
+ */
+static void smc_set_multicast_list(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+       unsigned char multicast_table[8];
+       int update_multicast = 0;
+
+       DBG(2, dev, "%s\n", __func__);
+
+       if (dev->flags & IFF_PROMISC) {
+               DBG(2, dev, "RCR_PRMS\n");
+               lp->rcr_cur_mode |= RCR_PRMS;
+       }
+
+/* BUG?  I never disable promiscuous mode if multicasting was turned on.
+   Now, I turn off promiscuous mode, but I don't do anything to multicasting
+   when promiscuous mode is turned on.
+*/
+
+       /*
+        * Here, I am setting this to accept all multicast packets.
+        * I don't need to zero the multicast table, because the flag is
+        * checked before the table is
+        */
+       else if (dev->flags & IFF_ALLMULTI || netdev_mc_count(dev) > 16) {
+               DBG(2, dev, "RCR_ALMUL\n");
+               lp->rcr_cur_mode |= RCR_ALMUL;
+       }
+
+       /*
+        * This sets the internal hardware table to filter out unwanted
+        * multicast packets before they take up memory.
+        *
+        * The SMC chip uses a hash table where the high 6 bits of the CRC of
+        * address are the offset into the table.  If that bit is 1, then the
+        * multicast packet is accepted.  Otherwise, it's dropped silently.
+        *
+        * To use the 6 bits as an offset into the table, the high 3 bits are
+        * the number of the 8 bit register, while the low 3 bits are the bit
+        * within that register.
+        */
+       else if (!netdev_mc_empty(dev)) {
+               struct netdev_hw_addr *ha;
+
+               /* table for flipping the order of 3 bits */
+               static const unsigned char invert3[] = {0, 4, 2, 6, 1, 5, 3, 7};
+
+               /* start with a table of all zeros: reject all */
+               memset(multicast_table, 0, sizeof(multicast_table));
+
+               netdev_for_each_mc_addr(ha, dev) {
+                       int position;
+
+                       /* only use the low order bits */
+                       position = crc32_le(~0, ha->addr, 6) & 0x3f;
+
+                       /* do some messy swapping to put the bit in the right spot */
+                       multicast_table[invert3[position&7]] |=
+                               (1<<invert3[(position>>3)&7]);
+               }
+
+               /* be sure I get rid of flags I might have set */
+               lp->rcr_cur_mode &= ~(RCR_PRMS | RCR_ALMUL);
+
+               /* now, the table can be loaded into the chipset */
+               update_multicast = 1;
+       } else  {
+               DBG(2, dev, "~(RCR_PRMS|RCR_ALMUL)\n");
+               lp->rcr_cur_mode &= ~(RCR_PRMS | RCR_ALMUL);
+
+               /*
+                * since I'm disabling all multicast entirely, I need to
+                * clear the multicast list
+                */
+               memset(multicast_table, 0, sizeof(multicast_table));
+               update_multicast = 1;
+       }
+
+       spin_lock_irq(&lp->lock);
+       SMC_SELECT_BANK(lp, 0);
+       SMC_SET_RCR(lp, lp->rcr_cur_mode);
+       if (update_multicast) {
+               SMC_SELECT_BANK(lp, 3);
+               SMC_SET_MCAST(lp, multicast_table);
+       }
+       SMC_SELECT_BANK(lp, 2);
+       spin_unlock_irq(&lp->lock);
+}
+
+
+/*
+ * Open and Initialize the board
+ *
+ * Set up everything, reset the card, etc..
+ */
+static int
+smc_open(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+
+       DBG(2, dev, "%s\n", __func__);
+
+       /* Setup the default Register Modes */
+       lp->tcr_cur_mode = TCR_DEFAULT;
+       lp->rcr_cur_mode = RCR_DEFAULT;
+       lp->rpc_cur_mode = RPC_DEFAULT |
+                               lp->cfg.leda << RPC_LSXA_SHFT |
+                               lp->cfg.ledb << RPC_LSXB_SHFT;
+
+       /*
+        * If we are not using a MII interface, we need to
+        * monitor our own carrier signal to detect faults.
+        */
+       if (lp->phy_type == 0)
+               lp->tcr_cur_mode |= TCR_MON_CSN;
+
+       /* reset the hardware */
+       smc_reset(dev);
+       smc_enable(dev);
+
+       /* Configure the PHY, initialize the link state */
+       if (lp->phy_type != 0)
+               smc_phy_configure(&lp->phy_configure);
+       else {
+               spin_lock_irq(&lp->lock);
+               smc_10bt_check_media(dev, 1);
+               spin_unlock_irq(&lp->lock);
+       }
+
+       netif_start_queue(dev);
+       return 0;
+}
+
+/*
+ * smc_close
+ *
+ * this makes the board clean up everything that it can
+ * and not talk to the outside world.   Caused by
+ * an 'ifconfig ethX down'
+ */
+static int smc_close(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+
+       DBG(2, dev, "%s\n", __func__);
+
+       netif_stop_queue(dev);
+       netif_carrier_off(dev);
+
+       /* clear everything */
+       smc_shutdown(dev);
+       tasklet_kill(&lp->tx_task);
+       smc_phy_powerdown(dev);
+       return 0;
+}
+
+/*
+ * Ethtool support
+ */
+static int
+smc_ethtool_getsettings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       int ret;
+
+       cmd->maxtxpkt = 1;
+       cmd->maxrxpkt = 1;
+
+       if (lp->phy_type != 0) {
+               spin_lock_irq(&lp->lock);
+               ret = mii_ethtool_gset(&lp->mii, cmd);
+               spin_unlock_irq(&lp->lock);
+       } else {
+               cmd->supported = SUPPORTED_10baseT_Half |
+                                SUPPORTED_10baseT_Full |
+                                SUPPORTED_TP | SUPPORTED_AUI;
+
+               if (lp->ctl_rspeed == 10)
+                       ethtool_cmd_speed_set(cmd, SPEED_10);
+               else if (lp->ctl_rspeed == 100)
+                       ethtool_cmd_speed_set(cmd, SPEED_100);
+
+               cmd->autoneg = AUTONEG_DISABLE;
+               cmd->transceiver = XCVR_INTERNAL;
+               cmd->port = 0;
+               cmd->duplex = lp->tcr_cur_mode & TCR_SWFDUP ? DUPLEX_FULL : DUPLEX_HALF;
+
+               ret = 0;
+       }
+
+       return ret;
+}
+
+static int
+smc_ethtool_setsettings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       int ret;
+
+       if (lp->phy_type != 0) {
+               spin_lock_irq(&lp->lock);
+               ret = mii_ethtool_sset(&lp->mii, cmd);
+               spin_unlock_irq(&lp->lock);
+       } else {
+               if (cmd->autoneg != AUTONEG_DISABLE ||
+                   cmd->speed != SPEED_10 ||
+                   (cmd->duplex != DUPLEX_HALF && cmd->duplex != DUPLEX_FULL) ||
+                   (cmd->port != PORT_TP && cmd->port != PORT_AUI))
+                       return -EINVAL;
+
+//             lp->port = cmd->port;
+               lp->ctl_rfduplx = cmd->duplex == DUPLEX_FULL;
+
+//             if (netif_running(dev))
+//                     smc_set_port(dev);
+
+               ret = 0;
+       }
+
+       return ret;
+}
+
+static void
+smc_ethtool_getdrvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
+{
+       strlcpy(info->driver, CARDNAME, sizeof(info->driver));
+       strlcpy(info->version, version, sizeof(info->version));
+       strlcpy(info->bus_info, dev_name(dev->dev.parent),
+               sizeof(info->bus_info));
+}
+
+static int smc_ethtool_nwayreset(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       int ret = -EINVAL;
+
+       if (lp->phy_type != 0) {
+               spin_lock_irq(&lp->lock);
+               ret = mii_nway_restart(&lp->mii);
+               spin_unlock_irq(&lp->lock);
+       }
+
+       return ret;
+}
+
+static u32 smc_ethtool_getmsglevel(struct net_device *dev)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       return lp->msg_enable;
+}
+
+static void smc_ethtool_setmsglevel(struct net_device *dev, u32 level)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       lp->msg_enable = level;
+}
+
+static int smc_write_eeprom_word(struct net_device *dev, u16 addr, u16 word)
+{
+       u16 ctl;
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+
+       spin_lock_irq(&lp->lock);
+       /* load word into GP register */
+       SMC_SELECT_BANK(lp, 1);
+       SMC_SET_GP(lp, word);
+       /* set the address to put the data in EEPROM */
+       SMC_SELECT_BANK(lp, 2);
+       SMC_SET_PTR(lp, addr);
+       /* tell it to write */
+       SMC_SELECT_BANK(lp, 1);
+       ctl = SMC_GET_CTL(lp);
+       SMC_SET_CTL(lp, ctl | (CTL_EEPROM_SELECT | CTL_STORE));
+       /* wait for it to finish */
+       do {
+               udelay(1);
+       } while (SMC_GET_CTL(lp) & CTL_STORE);
+       /* clean up */
+       SMC_SET_CTL(lp, ctl);
+       SMC_SELECT_BANK(lp, 2);
+       spin_unlock_irq(&lp->lock);
+       return 0;
+}
+
+static int smc_read_eeprom_word(struct net_device *dev, u16 addr, u16 *word)
+{
+       u16 ctl;
+       struct smc_local *lp = netdev_priv(dev);
+       void __iomem *ioaddr = lp->base;
+
+       spin_lock_irq(&lp->lock);
+       /* set the EEPROM address to get the data from */
+       SMC_SELECT_BANK(lp, 2);
+       SMC_SET_PTR(lp, addr | PTR_READ);
+       /* tell it to load */
+       SMC_SELECT_BANK(lp, 1);
+       SMC_SET_GP(lp, 0xffff); /* init to known */
+       ctl = SMC_GET_CTL(lp);
+       SMC_SET_CTL(lp, ctl | (CTL_EEPROM_SELECT | CTL_RELOAD));
+       /* wait for it to finish */
+       do {
+               udelay(1);
+       } while (SMC_GET_CTL(lp) & CTL_RELOAD);
+       /* read word from GP register */
+       *word = SMC_GET_GP(lp);
+       /* clean up */
+       SMC_SET_CTL(lp, ctl);
+       SMC_SELECT_BANK(lp, 2);
+       spin_unlock_irq(&lp->lock);
+       return 0;
+}
+
+static int smc_ethtool_geteeprom_len(struct net_device *dev)
+{
+       return 0x23 * 2;
+}
+
+static int smc_ethtool_geteeprom(struct net_device *dev,
+               struct ethtool_eeprom *eeprom, u8 *data)
+{
+       int i;
+       int imax;
+
+       DBG(1, dev, "Reading %d bytes at %d(0x%x)\n",
+               eeprom->len, eeprom->offset, eeprom->offset);
+       imax = smc_ethtool_geteeprom_len(dev);
+       for (i = 0; i < eeprom->len; i += 2) {
+               int ret;
+               u16 wbuf;
+               int offset = i + eeprom->offset;
+               if (offset > imax)
+                       break;
+               ret = smc_read_eeprom_word(dev, offset >> 1, &wbuf);
+               if (ret != 0)
+                       return ret;
+               DBG(2, dev, "Read 0x%x from 0x%x\n", wbuf, offset >> 1);
+               data[i] = (wbuf >> 8) & 0xff;
+               data[i+1] = wbuf & 0xff;
+       }
+       return 0;
+}
+
+static int smc_ethtool_seteeprom(struct net_device *dev,
+               struct ethtool_eeprom *eeprom, u8 *data)
+{
+       int i;
+       int imax;
+
+       DBG(1, dev, "Writing %d bytes to %d(0x%x)\n",
+           eeprom->len, eeprom->offset, eeprom->offset);
+       imax = smc_ethtool_geteeprom_len(dev);
+       for (i = 0; i < eeprom->len; i += 2) {
+               int ret;
+               u16 wbuf;
+               int offset = i + eeprom->offset;
+               if (offset > imax)
+                       break;
+               wbuf = (data[i] << 8) | data[i + 1];
+               DBG(2, dev, "Writing 0x%x to 0x%x\n", wbuf, offset >> 1);
+               ret = smc_write_eeprom_word(dev, offset >> 1, wbuf);
+               if (ret != 0)
+                       return ret;
+       }
+       return 0;
+}
+
+
+static const struct ethtool_ops smc_ethtool_ops = {
+       .get_settings   = smc_ethtool_getsettings,
+       .set_settings   = smc_ethtool_setsettings,
+       .get_drvinfo    = smc_ethtool_getdrvinfo,
+
+       .get_msglevel   = smc_ethtool_getmsglevel,
+       .set_msglevel   = smc_ethtool_setmsglevel,
+       .nway_reset     = smc_ethtool_nwayreset,
+       .get_link       = ethtool_op_get_link,
+       .get_eeprom_len = smc_ethtool_geteeprom_len,
+       .get_eeprom     = smc_ethtool_geteeprom,
+       .set_eeprom     = smc_ethtool_seteeprom,
+};
+
+static const struct net_device_ops smc_netdev_ops = {
+       .ndo_open               = smc_open,
+       .ndo_stop               = smc_close,
+       .ndo_start_xmit         = smc_hard_start_xmit,
+       .ndo_tx_timeout         = smc_timeout,
+       .ndo_set_rx_mode        = smc_set_multicast_list,
+       .ndo_change_mtu         = eth_change_mtu,
+       .ndo_validate_addr      = eth_validate_addr,
+       .ndo_set_mac_address    = eth_mac_addr,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = smc_poll_controller,
+#endif
+};
+
+/*
+ * smc_findirq
+ *
+ * This routine has a simple purpose -- make the SMC chip generate an
+ * interrupt, so an auto-detect routine can detect it, and find the IRQ,
+ */
+/*
+ * does this still work?
+ *
+ * I just deleted auto_irq.c, since it was never built...
+ *   --jgarzik
+ */
+static int smc_findirq(struct smc_local *lp)
+{
+       void __iomem *ioaddr = lp->base;
+       int timeout = 20;
+       unsigned long cookie;
+
+       DBG(2, lp->dev, "%s: %s\n", CARDNAME, __func__);
+
+       cookie = probe_irq_on();
+
+       /*
+        * What I try to do here is trigger an ALLOC_INT. This is done
+        * by allocating a small chunk of memory, which will give an interrupt
+        * when done.
+        */
+       /* enable ALLOCation interrupts ONLY */
+       SMC_SELECT_BANK(lp, 2);
+       SMC_SET_INT_MASK(lp, IM_ALLOC_INT);
+
+       /*
+        * Allocate 512 bytes of memory.  Note that the chip was just
+        * reset so all the memory is available
+        */
+       SMC_SET_MMU_CMD(lp, MC_ALLOC | 1);
+
+       /*
+        * Wait until positive that the interrupt has been generated
+        */
+       do {
+               int int_status;
+               udelay(10);
+               int_status = SMC_GET_INT(lp);
+               if (int_status & IM_ALLOC_INT)
+                       break;          /* got the interrupt */
+       } while (--timeout);
+
+       /*
+        * there is really nothing that I can do here if timeout fails,
+        * as autoirq_report will return a 0 anyway, which is what I
+        * want in this case.   Plus, the clean up is needed in both
+        * cases.
+        */
+
+       /* and disable all interrupts again */
+       SMC_SET_INT_MASK(lp, 0);
+
+       /* and return what I found */
+       return probe_irq_off(cookie);
+}
+
+/*
+ * Function: smc_probe(unsigned long ioaddr)
+ *
+ * Purpose:
+ *     Tests to see if a given ioaddr points to an SMC91x chip.
+ *     Returns a 0 on success
+ *
+ * Algorithm:
+ *     (1) see if the high byte of BANK_SELECT is 0x33
+ *     (2) compare the ioaddr with the base register's address
+ *     (3) see if I recognize the chip ID in the appropriate register
+ *
+ * Here I do typical initialization tasks.
+ *
+ * o  Initialize the structure if needed
+ * o  print out my vanity message if not done so already
+ * o  print out what type of hardware is detected
+ * o  print out the ethernet address
+ * o  find the IRQ
+ * o  set up my private data
+ * o  configure the dev structure with my subroutines
+ * o  actually GRAB the irq.
+ * o  GRAB the region
+ */
+static int smc_probe(struct net_device *dev, void __iomem *ioaddr,
+                    unsigned long irq_flags)
+{
+       struct smc_local *lp = netdev_priv(dev);
+       int retval;
+       unsigned int val, revision_register;
+       const char *version_string;
+
+       DBG(2, dev, "%s: %s\n", CARDNAME, __func__);
+
+       /* First, see if the high byte is 0x33 */
+       val = SMC_CURRENT_BANK(lp);
+       DBG(2, dev, "%s: bank signature probe returned 0x%04x\n",
+           CARDNAME, val);
+       if ((val & 0xFF00) != 0x3300) {
+               if ((val & 0xFF) == 0x33) {
+                       netdev_warn(dev,
+                                   "%s: Detected possible byte-swapped interface at IOADDR %p\n",
+                                   CARDNAME, ioaddr);
+               }
+               retval = -ENODEV;
+               goto err_out;
+       }
+
+       /*
+        * The above MIGHT indicate a device, but I need to write to
+        * further test this.
+        */
+       SMC_SELECT_BANK(lp, 0);
+       val = SMC_CURRENT_BANK(lp);
+       if ((val & 0xFF00) != 0x3300) {
+               retval = -ENODEV;
+               goto err_out;
+       }
+
+       /*
+        * well, we've already written once, so hopefully another
+        * time won't hurt.  This time, I need to switch the bank
+        * register to bank 1, so I can access the base address
+        * register
+        */
+       SMC_SELECT_BANK(lp, 1);
+       val = SMC_GET_BASE(lp);
+       val = ((val & 0x1F00) >> 3) << SMC_IO_SHIFT;
+       if (((unsigned long)ioaddr & (0x3e0 << SMC_IO_SHIFT)) != val) {
+               netdev_warn(dev, "%s: IOADDR %p doesn't match configuration (%x).\n",
+                           CARDNAME, ioaddr, val);
+       }
+
+       /*
+        * check if the revision register is something that I
+        * recognize.  These might need to be added to later,
+        * as future revisions could be added.
+        */
+       SMC_SELECT_BANK(lp, 3);
+       revision_register = SMC_GET_REV(lp);
+       DBG(2, dev, "%s: revision = 0x%04x\n", CARDNAME, revision_register);
+       version_string = chip_ids[ (revision_register >> 4) & 0xF];
+       if (!version_string || (revision_register & 0xff00) != 0x3300) {
+               /* I don't recognize this chip, so... */
+               netdev_warn(dev, "%s: IO %p: Unrecognized revision register 0x%04x, Contact author.\n",
+                           CARDNAME, ioaddr, revision_register);
+
+               retval = -ENODEV;
+               goto err_out;
+       }
+
+       /* At this point I'll assume that the chip is an SMC91x. */
+       pr_info_once("%s\n", version);
+
+       /* fill in some of the fields */
+       dev->base_addr = (unsigned long)ioaddr;
+       lp->base = ioaddr;
+       lp->version = revision_register & 0xff;
+       spin_lock_init(&lp->lock);
+
+       /* Get the MAC address */
+       SMC_SELECT_BANK(lp, 1);
+       SMC_GET_MAC_ADDR(lp, dev->dev_addr);
+
+       /* now, reset the chip, and put it into a known state */
+       smc_reset(dev);
+
+       /*
+        * If dev->irq is 0, then the device has to be banged on to see
+        * what the IRQ is.
+        *
+        * This banging doesn't always detect the IRQ, for unknown reasons.
+        * a workaround is to reset the chip and try again.
+        *
+        * Interestingly, the DOS packet driver *SETS* the IRQ on the card to
+        * be what is requested on the command line.   I don't do that, mostly
+        * because the card that I have uses a non-standard method of accessing
+        * the IRQs, and because this _should_ work in most configurations.
+        *
+        * Specifying an IRQ is done with the assumption that the user knows
+        * what (s)he is doing.  No checking is done!!!!
+        */
+       if (dev->irq < 1) {
+               int trials;
+
+               trials = 3;
+               while (trials--) {
+                       dev->irq = smc_findirq(lp);
+                       if (dev->irq)
+                               break;
+                       /* kick the card and try again */
+                       smc_reset(dev);
+               }
+       }
+       if (dev->irq == 0) {
+               netdev_warn(dev, "Couldn't autodetect your IRQ. Use irq=xx.\n");
+               retval = -ENODEV;
+               goto err_out;
+       }
+       dev->irq = irq_canonicalize(dev->irq);
+
+       dev->watchdog_timeo = msecs_to_jiffies(watchdog);
+       dev->netdev_ops = &smc_netdev_ops;
+       dev->ethtool_ops = &smc_ethtool_ops;
+
+       tasklet_init(&lp->tx_task, smc_hardware_send_pkt, (unsigned long)dev);
+       INIT_WORK(&lp->phy_configure, smc_phy_configure);
+       lp->dev = dev;
+       lp->mii.phy_id_mask = 0x1f;
+       lp->mii.reg_num_mask = 0x1f;
+       lp->mii.force_media = 0;
+       lp->mii.full_duplex = 0;
+       lp->mii.dev = dev;
+       lp->mii.mdio_read = smc_phy_read;
+       lp->mii.mdio_write = smc_phy_write;
+
+       /*
+        * Locate the phy, if any.
+        */
+       if (lp->version >= (CHIP_91100 << 4))
+               smc_phy_detect(dev);
+
+       /* then shut everything down to save power */
+       smc_shutdown(dev);
+       smc_phy_powerdown(dev);
+
+       /* Set default parameters */
+       lp->msg_enable = NETIF_MSG_LINK;
+       lp->ctl_rfduplx = 0;
+       lp->ctl_rspeed = 10;
+
+       if (lp->version >= (CHIP_91100 << 4)) {
+               lp->ctl_rfduplx = 1;
+               lp->ctl_rspeed = 100;
+       }
+
+       /* Grab the IRQ */
+       retval = request_irq(dev->irq, smc_interrupt, irq_flags, dev->name, dev);
+       if (retval)
+               goto err_out;
+
+#ifdef CONFIG_ARCH_PXA
+#  ifdef SMC_USE_PXA_DMA
+       lp->cfg.flags |= SMC91X_USE_DMA;
+#  endif
+       if (lp->cfg.flags & SMC91X_USE_DMA) {
+               int dma = pxa_request_dma(dev->name, DMA_PRIO_LOW,
+                                         smc_pxa_dma_irq, NULL);
+               if (dma >= 0)
+                       dev->dma = dma;
+       }
+#endif
+
+       retval = register_netdev(dev);
+       if (retval == 0) {
+               /* now, print out the card info, in a short format.. */
+               netdev_info(dev, "%s (rev %d) at %p IRQ %d",
+                           version_string, revision_register & 0x0f,
+                           lp->base, dev->irq);
+
+               if (dev->dma != (unsigned char)-1)
+                       pr_cont(" DMA %d", dev->dma);
+
+               pr_cont("%s%s\n",
+                       lp->cfg.flags & SMC91X_NOWAIT ? " [nowait]" : "",
+                       THROTTLE_TX_PKTS ? " [throttle_tx]" : "");
+
+               if (!is_valid_ether_addr(dev->dev_addr)) {
+                       netdev_warn(dev, "Invalid ethernet MAC address. Please set using ifconfig\n");
+               } else {
+                       /* Print the Ethernet address */
+                       netdev_info(dev, "Ethernet addr: %pM\n",
+                                   dev->dev_addr);
+               }
+
+               if (lp->phy_type == 0) {
+                       PRINTK(dev, "No PHY found\n");
+               } else if ((lp->phy_type & 0xfffffff0) == 0x0016f840) {
+                       PRINTK(dev, "PHY LAN83C183 (LAN91C111 Internal)\n");
+               } else if ((lp->phy_type & 0xfffffff0) == 0x02821c50) {
+                       PRINTK(dev, "PHY LAN83C180\n");
+               }
+       }
+
+err_out:
+#ifdef CONFIG_ARCH_PXA
+       if (retval && dev->dma != (unsigned char)-1)
+               pxa_free_dma(dev->dma);
+#endif
+       return retval;
+}
+
+static int smc_enable_device(struct platform_device *pdev)
+{
+       struct net_device *ndev = platform_get_drvdata(pdev);
+       struct smc_local *lp = netdev_priv(ndev);
+       unsigned long flags;
+       unsigned char ecor, ecsr;
+       void __iomem *addr;
+       struct resource * res;
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "smc91x-attrib");
+       if (!res)
+               return 0;
+
+       /*
+        * Map the attribute space.  This is overkill, but clean.
+        */
+       addr = ioremap(res->start, ATTRIB_SIZE);
+       if (!addr)
+               return -ENOMEM;
+
+       /*
+        * Reset the device.  We must disable IRQs around this
+        * since a reset causes the IRQ line become active.
+        */
+       local_irq_save(flags);
+       ecor = readb(addr + (ECOR << SMC_IO_SHIFT)) & ~ECOR_RESET;
+       writeb(ecor | ECOR_RESET, addr + (ECOR << SMC_IO_SHIFT));
+       readb(addr + (ECOR << SMC_IO_SHIFT));
+
+       /*
+        * Wait 100us for the chip to reset.
+        */
+       udelay(100);
+
+       /*
+        * The device will ignore all writes to the enable bit while
+        * reset is asserted, even if the reset bit is cleared in the
+        * same write.  Must clear reset first, then enable the device.
+        */
+       writeb(ecor, addr + (ECOR << SMC_IO_SHIFT));
+       writeb(ecor | ECOR_ENABLE, addr + (ECOR << SMC_IO_SHIFT));
+
+       /*
+        * Set the appropriate byte/word mode.
+        */
+       ecsr = readb(addr + (ECSR << SMC_IO_SHIFT)) & ~ECSR_IOIS8;
+       if (!SMC_16BIT(lp))
+               ecsr |= ECSR_IOIS8;
+       writeb(ecsr, addr + (ECSR << SMC_IO_SHIFT));
+       local_irq_restore(flags);
+
+       iounmap(addr);
+
+       /*
+        * Wait for the chip to wake up.  We could poll the control
+        * register in the main register space, but that isn't mapped
+        * yet.  We know this is going to take 750us.
+        */
+       msleep(1);
+
+       return 0;
+}
+
+static int smc_request_attrib(struct platform_device *pdev,
+                             struct net_device *ndev)
+{
+       struct resource * res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "smc91x-attrib");
+       struct smc_local *lp __maybe_unused = netdev_priv(ndev);
+
+       if (!res)
+               return 0;
+
+       if (!request_mem_region(res->start, ATTRIB_SIZE, CARDNAME))
+               return -EBUSY;
+
+       return 0;
+}
+
+static void smc_release_attrib(struct platform_device *pdev,
+                              struct net_device *ndev)
+{
+       struct resource * res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "smc91x-attrib");
+       struct smc_local *lp __maybe_unused = netdev_priv(ndev);
+
+       if (res)
+               release_mem_region(res->start, ATTRIB_SIZE);
+}
+
+static inline void smc_request_datacs(struct platform_device *pdev, struct net_device *ndev)
+{
+       if (SMC_CAN_USE_DATACS) {
+               struct resource * res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "smc91x-data32");
+               struct smc_local *lp = netdev_priv(ndev);
+
+               if (!res)
+                       return;
+
+               if(!request_mem_region(res->start, SMC_DATA_EXTENT, CARDNAME)) {
+                       netdev_info(ndev, "%s: failed to request datacs memory region.\n",
+                                   CARDNAME);
+                       return;
+               }
+
+               lp->datacs = ioremap(res->start, SMC_DATA_EXTENT);
+       }
+}
+
+static void smc_release_datacs(struct platform_device *pdev, struct net_device *ndev)
+{
+       if (SMC_CAN_USE_DATACS) {
+               struct smc_local *lp = netdev_priv(ndev);
+               struct resource * res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "smc91x-data32");
+
+               if (lp->datacs)
+                       iounmap(lp->datacs);
+
+               lp->datacs = NULL;
+
+               if (res)
+                       release_mem_region(res->start, SMC_DATA_EXTENT);
+       }
+}
+
+#if IS_BUILTIN(CONFIG_OF)
+static const struct of_device_id smc91x_match[] = {
+       { .compatible = "smsc,lan91c94", },
+       { .compatible = "smsc,lan91c111", },
+       {},
+};
+MODULE_DEVICE_TABLE(of, smc91x_match);
+
+/**
+ * of_try_set_control_gpio - configure a gpio if it exists
+ */
+static int try_toggle_control_gpio(struct device *dev,
+                                  struct gpio_desc **desc,
+                                  const char *name, int index,
+                                  int value, unsigned int nsdelay)
+{
+       struct gpio_desc *gpio = *desc;
+       enum gpiod_flags flags = value ? GPIOD_OUT_LOW : GPIOD_OUT_HIGH;
+
+       gpio = devm_gpiod_get_index_optional(dev, name, index, flags);
+       if (IS_ERR(gpio))
+               return PTR_ERR(gpio);
+
+       if (gpio) {
+               if (nsdelay)
+                       usleep_range(nsdelay, 2 * nsdelay);
+               gpiod_set_value_cansleep(gpio, value);
+       }
+       *desc = gpio;
+
+       return 0;
+}
+#endif
+
+/*
+ * smc_init(void)
+ *   Input parameters:
+ *     dev->base_addr == 0, try to find all possible locations
+ *     dev->base_addr > 0x1ff, this is the address to check
+ *     dev->base_addr == <anything else>, return failure code
+ *
+ *   Output:
+ *     0 --> there is a device
+ *     anything else, error
+ */
+static int smc_drv_probe(struct platform_device *pdev)
+{
+       struct smc91x_platdata *pd = dev_get_platdata(&pdev->dev);
+       const struct of_device_id *match = NULL;
+       struct smc_local *lp;
+       struct net_device *ndev;
+       struct resource *res;
+       unsigned int __iomem *addr;
+       unsigned long irq_flags = SMC_IRQ_FLAGS;
+       unsigned long irq_resflags;
+       int ret;
+
+       ndev = alloc_etherdev(sizeof(struct smc_local));
+       if (!ndev) {
+               ret = -ENOMEM;
+               goto out;
+       }
+       SET_NETDEV_DEV(ndev, &pdev->dev);
+
+       /* get configuration from platform data, only allow use of
+        * bus width if both SMC_CAN_USE_xxx and SMC91X_USE_xxx are set.
+        */
+
+       lp = netdev_priv(ndev);
+       lp->cfg.flags = 0;
+
+       if (pd) {
+               memcpy(&lp->cfg, pd, sizeof(lp->cfg));
+               lp->io_shift = SMC91X_IO_SHIFT(lp->cfg.flags);
+       }
+
+#if IS_BUILTIN(CONFIG_OF)
+       match = of_match_device(of_match_ptr(smc91x_match), &pdev->dev);
+       if (match) {
+               struct device_node *np = pdev->dev.of_node;
+               u32 val;
+
+               /* Optional pwrdwn GPIO configured? */
+               ret = try_toggle_control_gpio(&pdev->dev, &lp->power_gpio,
+                                             "power", 0, 0, 100);
+               if (ret)
+                       return ret;
+
+               /*
+                * Optional reset GPIO configured? Minimum 100 ns reset needed
+                * according to LAN91C96 datasheet page 14.
+                */
+               ret = try_toggle_control_gpio(&pdev->dev, &lp->reset_gpio,
+                                             "reset", 0, 0, 100);
+               if (ret)
+                       return ret;
+
+               /*
+                * Need to wait for optional EEPROM to load, max 750 us according
+                * to LAN91C96 datasheet page 55.
+                */
+               if (lp->reset_gpio)
+                       usleep_range(750, 1000);
+
+               /* Combination of IO widths supported, default to 16-bit */
+               if (!of_property_read_u32(np, "reg-io-width", &val)) {
+                       if (val & 1)
+                               lp->cfg.flags |= SMC91X_USE_8BIT;
+                       if ((val == 0) || (val & 2))
+                               lp->cfg.flags |= SMC91X_USE_16BIT;
+                       if (val & 4)
+                               lp->cfg.flags |= SMC91X_USE_32BIT;
+               } else {
+                       lp->cfg.flags |= SMC91X_USE_16BIT;
+               }
+       }
+#endif
+
+       if (!pd && !match) {
+               lp->cfg.flags |= (SMC_CAN_USE_8BIT)  ? SMC91X_USE_8BIT  : 0;
+               lp->cfg.flags |= (SMC_CAN_USE_16BIT) ? SMC91X_USE_16BIT : 0;
+               lp->cfg.flags |= (SMC_CAN_USE_32BIT) ? SMC91X_USE_32BIT : 0;
+               lp->cfg.flags |= (nowait) ? SMC91X_NOWAIT : 0;
+       }
+
+       if (!lp->cfg.leda && !lp->cfg.ledb) {
+               lp->cfg.leda = RPC_LSA_DEFAULT;
+               lp->cfg.ledb = RPC_LSB_DEFAULT;
+       }
+
+       ndev->dma = (unsigned char)-1;
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "smc91x-regs");
+       if (!res)
+               res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!res) {
+               ret = -ENODEV;
+               goto out_free_netdev;
+       }
+
+
+       if (!request_mem_region(res->start, SMC_IO_EXTENT, CARDNAME)) {
+               ret = -EBUSY;
+               goto out_free_netdev;
+       }
+
+       ndev->irq = platform_get_irq(pdev, 0);
+       if (ndev->irq <= 0) {
+               ret = -ENODEV;
+               goto out_release_io;
+       }
+       /*
+        * If this platform does not specify any special irqflags, or if
+        * the resource supplies a trigger, override the irqflags with
+        * the trigger flags from the resource.
+        */
+       irq_resflags = irqd_get_trigger_type(irq_get_irq_data(ndev->irq));
+       if (irq_flags == -1 || irq_resflags & IRQF_TRIGGER_MASK)
+               irq_flags = irq_resflags & IRQF_TRIGGER_MASK;
+
+       ret = smc_request_attrib(pdev, ndev);
+       if (ret)
+               goto out_release_io;
+#if defined(CONFIG_ASSABET_NEPONSET)
+       if (machine_is_assabet() && machine_has_neponset())
+               neponset_ncr_set(NCR_ENET_OSC_EN);
+#endif
+       platform_set_drvdata(pdev, ndev);
+       ret = smc_enable_device(pdev);
+       if (ret)
+               goto out_release_attrib;
+
+       addr = ioremap(res->start, SMC_IO_EXTENT);
+       if (!addr) {
+               ret = -ENOMEM;
+               goto out_release_attrib;
+       }
+
+#ifdef CONFIG_ARCH_PXA
+       {
+               struct smc_local *lp = netdev_priv(ndev);
+               lp->device = &pdev->dev;
+               lp->physaddr = res->start;
+       }
+#endif
+
+       ret = smc_probe(ndev, addr, irq_flags);
+       if (ret != 0)
+               goto out_iounmap;
+
+       smc_request_datacs(pdev, ndev);
+
+       return 0;
+
+ out_iounmap:
+       iounmap(addr);
+ out_release_attrib:
+       smc_release_attrib(pdev, ndev);
+ out_release_io:
+       release_mem_region(res->start, SMC_IO_EXTENT);
+ out_free_netdev:
+       free_netdev(ndev);
+ out:
+       pr_info("%s: not found (%d).\n", CARDNAME, ret);
+
+       return ret;
+}
+
+static int smc_drv_remove(struct platform_device *pdev)
+{
+       struct net_device *ndev = platform_get_drvdata(pdev);
+       struct smc_local *lp = netdev_priv(ndev);
+       struct resource *res;
+
+       unregister_netdev(ndev);
+
+       free_irq(ndev->irq, ndev);
+
+#ifdef CONFIG_ARCH_PXA
+       if (ndev->dma != (unsigned char)-1)
+               pxa_free_dma(ndev->dma);
+#endif
+       iounmap(lp->base);
+
+       smc_release_datacs(pdev,ndev);
+       smc_release_attrib(pdev,ndev);
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "smc91x-regs");
+       if (!res)
+               res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       release_mem_region(res->start, SMC_IO_EXTENT);
+
+       free_netdev(ndev);
+
+       return 0;
+}
+
+static int smc_drv_suspend(struct device *dev)
+{
+       struct platform_device *pdev = to_platform_device(dev);
+       struct net_device *ndev = platform_get_drvdata(pdev);
+
+       if (ndev) {
+               if (netif_running(ndev)) {
+                       netif_device_detach(ndev);
+                       smc_shutdown(ndev);
+                       smc_phy_powerdown(ndev);
+               }
+       }
+       return 0;
+}
+
+static int smc_drv_resume(struct device *dev)
+{
+       struct platform_device *pdev = to_platform_device(dev);
+       struct net_device *ndev = platform_get_drvdata(pdev);
+
+       if (ndev) {
+               struct smc_local *lp = netdev_priv(ndev);
+               smc_enable_device(pdev);
+               if (netif_running(ndev)) {
+                       smc_reset(ndev);
+                       smc_enable(ndev);
+                       if (lp->phy_type != 0)
+                               smc_phy_configure(&lp->phy_configure);
+                       netif_device_attach(ndev);
+               }
+       }
+       return 0;
+}
+
+static struct dev_pm_ops smc_drv_pm_ops = {
+       .suspend        = smc_drv_suspend,
+       .resume         = smc_drv_resume,
+};
+
+static struct platform_driver smc_driver = {
+       .probe          = smc_drv_probe,
+       .remove         = smc_drv_remove,
+       .driver         = {
+               .name   = CARDNAME,
+               .pm     = &smc_drv_pm_ops,
+               .of_match_table = of_match_ptr(smc91x_match),
+       },
+};
+
+module_platform_driver(smc_driver);