[PATCH] sis190: resurrection
authorFrancois Romieu <romieu@fr.zoreil.com>
Sat, 30 Jul 2005 11:08:43 +0000 (13:08 +0200)
committerJeff Garzik <jgarzik@pobox.com>
Sat, 30 Jul 2005 22:20:59 +0000 (18:20 -0400)
Raise the sis190 driver from the dead

The driver handles the integrated network device found on SiS 965L
chipset. It follows the classical (non-napi) interrupt-driven model
and provides minimal ethtool support.

The code comes from a heavy cleanup/rewrite of the original code
which was removed from the kernel on 14/04/2004. Since the r8169
driver does not work too bad and there will probably be (at least)
a few months of improvements/testing/fixing, I made the code as
close as possible to the r8169 one.

Pascal Chapperon <pascal.chapperon@wanadoo.fr> deserves some special
credit for testing and bug-catching. Many thanks to Lars Vahlenberg
as well.

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
drivers/net/Kconfig
drivers/net/Makefile
drivers/net/sis190.c [new file with mode: 0644]

index 8a835eb58808726eb8c7c478f6bfd3c3a450db6f..765fbb29d386033620ef89cbb74ab6a7426c8ee8 100644 (file)
@@ -1921,6 +1921,16 @@ config R8169_VLAN
          
          If in doubt, say Y.
 
+config SIS190
+      tristate "SiS190 gigabit ethernet support"
+      depends on PCI
+      select CRC32
+      ---help---
+        Say Y here if you have a SiS 190 PCI Gigabit Ethernet adapter.
+
+        To compile this driver as a module, choose M here: the module
+        will be called sis190.  This is recommended.
+
 config SKGE
        tristate "New SysKonnect GigaEthernet support (EXPERIMENTAL)"
        depends on PCI && EXPERIMENTAL
index 63c6d1e6d4d982937ada0fdaca07003ec29ef9e6..67b280af425e4f60f877498b00b24a7356fee011 100644 (file)
@@ -42,6 +42,7 @@ obj-$(CONFIG_EEPRO100) += eepro100.o
 obj-$(CONFIG_E100) += e100.o
 obj-$(CONFIG_TLAN) += tlan.o
 obj-$(CONFIG_EPIC100) += epic100.o
+obj-$(CONFIG_SIS190) += sis190.o
 obj-$(CONFIG_SIS900) += sis900.o
 obj-$(CONFIG_YELLOWFIN) += yellowfin.o
 obj-$(CONFIG_ACENIC) += acenic.o
diff --git a/drivers/net/sis190.c b/drivers/net/sis190.c
new file mode 100644 (file)
index 0000000..fd303e7
--- /dev/null
@@ -0,0 +1,1359 @@
+/*
+   sis190.c: Silicon Integrated Systems SiS190 ethernet driver
+
+   Copyright (c) 2003 K.M. Liu <kmliu@sis.com>
+   Copyright (c) 2003, 2004 Jeff Garzik <jgarzik@pobox.com>
+   Copyright (c) 2003, 2004, 2005 Francois Romieu <romieu@fr.zoreil.com>
+
+   Based on r8169.c, tg3.c, 8139cp.c, skge.c and probably even epic100.c.
+
+   This software may be used and distributed according to the terms of
+   the GNU General Public License (GPL), incorporated herein by reference.
+   Drivers based on or derived from this code fall under the GPL and must
+   retain the authorship, copyright and license notice.  This file is not
+   a complete program and may only be used when the entire operating
+   system is licensed under the GPL.
+
+   See the file COPYING in this distribution for more information.
+
+ */
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/ethtool.h>
+#include <linux/pci.h>
+#include <linux/mii.h>
+#include <linux/delay.h>
+#include <linux/crc32.h>
+#include <linux/dma-mapping.h>
+#include <asm/irq.h>
+
+#define net_drv(p, arg...)     if (netif_msg_drv(p)) \
+                                       printk(arg)
+#define net_probe(p, arg...)   if (netif_msg_probe(p)) \
+                                       printk(arg)
+#define net_link(p, arg...)    if (netif_msg_link(p)) \
+                                       printk(arg)
+#define net_intr(p, arg...)    if (netif_msg_intr(p)) \
+                                       printk(arg)
+#define net_tx_err(p, arg...)  if (netif_msg_tx_err(p)) \
+                                       printk(arg)
+
+#ifdef CONFIG_SIS190_NAPI
+#define NAPI_SUFFIX    "-NAPI"
+#else
+#define NAPI_SUFFIX    ""
+#endif
+
+#define DRV_VERSION            "1.2" NAPI_SUFFIX
+#define DRV_NAME               "sis190"
+#define SIS190_DRIVER_NAME     DRV_NAME " Gigabit Ethernet driver " DRV_VERSION
+#define PFX DRV_NAME ": "
+
+#ifdef CONFIG_SIS190_NAPI
+#define sis190_rx_skb                  netif_receive_skb
+#define sis190_rx_quota(count, quota)  min(count, quota)
+#else
+#define sis190_rx_skb                  netif_rx
+#define sis190_rx_quota(count, quota)  count
+#endif
+
+#define MAC_ADDR_LEN           6
+
+#define NUM_TX_DESC            64
+#define NUM_RX_DESC            64
+#define TX_RING_BYTES          (NUM_TX_DESC * sizeof(struct TxDesc))
+#define RX_RING_BYTES          (NUM_RX_DESC * sizeof(struct RxDesc))
+#define RX_BUF_SIZE            1536
+
+#define SIS190_REGS_SIZE       0x80
+#define SIS190_TX_TIMEOUT      (6*HZ)
+#define SIS190_PHY_TIMEOUT     (10*HZ)
+#define SIS190_MSG_DEFAULT     (NETIF_MSG_DRV | NETIF_MSG_PROBE | \
+                                NETIF_MSG_LINK | NETIF_MSG_IFUP | \
+                                NETIF_MSG_IFDOWN)
+
+/* Enhanced PHY access register bit definitions */
+#define EhnMIIread             0x0000
+#define EhnMIIwrite            0x0020
+#define EhnMIIdataShift                16
+#define EhnMIIpmdShift         6       /* 7016 only */
+#define EhnMIIregShift         11
+#define EhnMIIreq              0x0010
+#define EhnMIInotDone          0x0010
+
+/* Write/read MMIO register */
+#define SIS_W8(reg, val)       writeb ((val), ioaddr + (reg))
+#define SIS_W16(reg, val)      writew ((val), ioaddr + (reg))
+#define SIS_W32(reg, val)      writel ((val), ioaddr + (reg))
+#define SIS_R8(reg)            readb (ioaddr + (reg))
+#define SIS_R16(reg)           readw (ioaddr + (reg))
+#define SIS_R32(reg)           readl (ioaddr + (reg))
+
+#define SIS_PCI_COMMIT()       SIS_R32(IntrControl)
+
+enum sis190_registers {
+       TxControl               = 0x00,
+       TxDescStartAddr         = 0x04,
+       TxNextDescAddr          = 0x0c, // unused
+       RxControl               = 0x10,
+       RxDescStartAddr         = 0x14,
+       RxNextDescAddr          = 0x1c, // unused
+       IntrStatus              = 0x20,
+       IntrMask                = 0x24,
+       IntrControl             = 0x28,
+       IntrTimer               = 0x2c, // unused
+       PMControl               = 0x30, // unused
+       ROMControl              = 0x38,
+       ROMInterface            = 0x3c,
+       StationControl          = 0x40,
+       GMIIControl             = 0x44,
+       TxMacControl            = 0x50,
+       RxMacControl            = 0x60,
+       RxMacAddr               = 0x62,
+       RxHashTable             = 0x68,
+       // Undocumented         = 0x6c,
+       RxWakeOnLan             = 0x70,
+       // Undocumented         = 0x74,
+       RxMPSControl            = 0x78, // unused
+};
+
+enum sis190_register_content {
+       /* IntrStatus */
+       SoftInt                 = 0x40000000,   // unused
+       Timeup                  = 0x20000000,   // unused
+       PauseFrame              = 0x00080000,   // unused
+       MagicPacket             = 0x00040000,   // unused
+       WakeupFrame             = 0x00020000,   // unused
+       LinkChange              = 0x00010000,
+       RxQEmpty                = 0x00000080,
+       RxQInt                  = 0x00000040,
+       TxQ1Empty               = 0x00000020,   // unused
+       TxQ1Int                 = 0x00000010,
+       TxQ0Empty               = 0x00000008,   // unused
+       TxQ0Int                 = 0x00000004,
+       RxHalt                  = 0x00000002,
+       TxHalt                  = 0x00000001,
+
+       /* RxStatusDesc */
+       RxRES                   = 0x00200000,   // unused
+       RxCRC                   = 0x00080000,
+       RxRUNT                  = 0x00100000,   // unused
+       RxRWT                   = 0x00400000,   // unused
+
+       /* {Rx/Tx}CmdBits */
+       CmdReset                = 0x10,
+       CmdRxEnb                = 0x08,         // unused
+       CmdTxEnb                = 0x01,
+       RxBufEmpty              = 0x01,         // unused
+
+       /* Cfg9346Bits */
+       Cfg9346_Lock            = 0x00,         // unused
+       Cfg9346_Unlock          = 0xc0,         // unused
+
+       /* RxMacControl */
+       AcceptErr               = 0x20,         // unused
+       AcceptRunt              = 0x10,         // unused
+       AcceptBroadcast         = 0x0800,
+       AcceptMulticast         = 0x0400,
+       AcceptMyPhys            = 0x0200,
+       AcceptAllPhys           = 0x0100,
+
+       /* RxConfigBits */
+       RxCfgFIFOShift          = 13,
+       RxCfgDMAShift           = 8,            // 0x1a in RxControl ?
+
+       /* TxConfigBits */
+       TxInterFrameGapShift    = 24,
+       TxDMAShift              = 8, /* DMA burst value (0-7) is shift this many bits */
+
+       /* StationControl */
+       _1000bpsF               = 0x1c00,
+       _1000bpsH               = 0x0c00,
+       _100bpsF                = 0x1800,
+       _100bpsH                = 0x0800,
+       _10bpsF                 = 0x1400,
+       _10bpsH                 = 0x0400,
+
+       LinkStatus              = 0x02,         // unused
+       FullDup                 = 0x01,         // unused
+
+       /* TBICSRBit */
+       TBILinkOK               = 0x02000000,   // unused
+};
+
+struct TxDesc {
+       u32 PSize;
+       u32 status;
+       u32 addr;
+       u32 size;
+};
+
+struct RxDesc {
+       u32 PSize;
+       u32 status;
+       u32 addr;
+       u32 size;
+};
+
+enum _DescStatusBit {
+       /* _Desc.status */
+       OWNbit          = 0x80000000,
+       INTbit          = 0x40000000,
+       DEFbit          = 0x00200000,
+       CRCbit          = 0x00020000,
+       PADbit          = 0x00010000,
+       /* _Desc.size */
+       RingEnd         = (1 << 31),
+       /* _Desc.PSize */
+       RxSizeMask      = 0x0000ffff
+};
+
+struct sis190_private {
+       void __iomem *mmio_addr;
+       struct pci_dev *pci_dev;
+       struct net_device_stats stats;
+       spinlock_t lock;
+       u32 rx_buf_sz;
+       u32 cur_rx;
+       u32 cur_tx;
+       u32 dirty_rx;
+       u32 dirty_tx;
+       dma_addr_t rx_dma;
+       dma_addr_t tx_dma;
+       struct RxDesc *RxDescRing;
+       struct TxDesc *TxDescRing;
+       struct sk_buff *Rx_skbuff[NUM_RX_DESC];
+       struct sk_buff *Tx_skbuff[NUM_TX_DESC];
+       struct work_struct phy_task;
+       struct timer_list timer;
+       u32 msg_enable;
+};
+
+const static struct {
+       const char *name;
+       u8 version;             /* depend on docs */
+       u32 RxConfigMask;       /* clear the bits supported by this chip */
+} sis_chip_info[] = {
+       { DRV_NAME, 0x00, 0xff7e1880, },
+};
+
+static struct pci_device_id sis190_pci_tbl[] __devinitdata = {
+       { PCI_DEVICE(PCI_VENDOR_ID_SI, 0x0190), 0, 0, 0 },
+       { 0, },
+};
+
+MODULE_DEVICE_TABLE(pci, sis190_pci_tbl);
+
+static int rx_copybreak = 200;
+
+static struct {
+       u32 msg_enable;
+} debug = { -1 };
+
+MODULE_DESCRIPTION("SiS sis190 Gigabit Ethernet driver");
+module_param(rx_copybreak, int, 0);
+MODULE_PARM_DESC(rx_copybreak, "Copy breakpoint for copy-only-tiny-frames");
+module_param_named(debug, debug.msg_enable, int, 0);
+MODULE_PARM_DESC(debug, "Debug verbosity level (0=none, ..., 16=all)");
+MODULE_AUTHOR("K.M. Liu <kmliu@sis.com>, Ueimor <romieu@fr.zoreil.com>");
+MODULE_VERSION(DRV_VERSION);
+MODULE_LICENSE("GPL");
+
+static const u32 sis190_intr_mask =
+       RxQEmpty | RxQInt | TxQ1Int | TxQ0Int | RxHalt | TxHalt;
+
+/*
+ * Maximum number of multicast addresses to filter (vs. Rx-all-multicast).
+ * The chips use a 64 element hash table based on the Ethernet CRC.
+ */
+static int multicast_filter_limit = 32;
+
+static void __mdio_cmd(void __iomem *ioaddr, u32 ctl)
+{
+       unsigned int i;
+
+       SIS_W32(GMIIControl, ctl);
+
+       msleep(1);
+
+       for (i = 0; i < 100; i++) {
+               if (!(SIS_R32(GMIIControl) & EhnMIInotDone))
+                       break;
+               msleep(1);
+       }
+
+       if (i > 999)
+               printk(KERN_ERR PFX "PHY command failed !\n");
+}
+
+static void mdio_write(void __iomem *ioaddr, int reg, int val)
+{
+       u32 pmd = 1;
+
+       __mdio_cmd(ioaddr, EhnMIIreq | EhnMIIwrite |
+               (((u32) reg) << EhnMIIregShift) | (pmd << EhnMIIpmdShift) |
+               (((u32) val) << EhnMIIdataShift));
+}
+
+static int mdio_read(void __iomem *ioaddr, int reg)
+{
+       u32 pmd = 1;
+
+       __mdio_cmd(ioaddr, EhnMIIreq | EhnMIIread |
+               (((u32) reg) << EhnMIIregShift) | (pmd << EhnMIIpmdShift));
+
+       return (u16) (SIS_R32(GMIIControl) >> EhnMIIdataShift);
+}
+
+static int sis190_read_eeprom(void __iomem *ioaddr, u32 reg)
+{
+       unsigned int i;
+       u16 data;
+       u32 val;
+
+       if (!(SIS_R32(ROMControl) & 0x0002))
+               return 0;
+
+       val = (0x0080 | (0x2 << 8) | (reg << 10));
+
+       SIS_W32(ROMInterface, val);
+
+       for (i = 0; i < 200; i++) {
+               if (!(SIS_R32(ROMInterface) & 0x0080))
+                       break;
+               msleep(1);
+       }
+
+       data = (u16) ((SIS_R32(ROMInterface) & 0xffff0000) >> 16);
+
+       return data;
+}
+
+static void sis190_irq_mask_and_ack(void __iomem *ioaddr)
+{
+       SIS_W32(IntrMask, 0x00);
+       SIS_W32(IntrStatus, 0xffffffff);
+       SIS_PCI_COMMIT();
+}
+
+static void sis190_asic_down(void __iomem *ioaddr)
+{
+       /* Stop the chip's Tx and Rx DMA processes. */
+
+       SIS_W32(TxControl, 0x1a00);
+       SIS_W32(RxControl, 0x1a00);
+
+       sis190_irq_mask_and_ack(ioaddr);
+}
+
+static void sis190_mark_as_last_descriptor(struct RxDesc *desc)
+{
+       desc->size |= cpu_to_le32(RingEnd);
+}
+
+static inline void sis190_give_to_asic(struct RxDesc *desc, u32 rx_buf_sz)
+{
+       u32 eor = le32_to_cpu(desc->size) & RingEnd;
+
+       desc->PSize = 0x0;
+       desc->size = cpu_to_le32(rx_buf_sz | eor);
+       wmb();
+       desc->status = cpu_to_le32(OWNbit | INTbit);
+}
+
+static inline void sis190_map_to_asic(struct RxDesc *desc, dma_addr_t mapping,
+                                     u32 rx_buf_sz)
+{
+       desc->addr = cpu_to_le32(mapping);
+       sis190_give_to_asic(desc, rx_buf_sz);
+}
+
+static inline void sis190_make_unusable_by_asic(struct RxDesc *desc)
+{
+       desc->PSize = 0x0;
+       desc->addr = 0xdeadbeef;
+       desc->size &= cpu_to_le32(RingEnd);
+       wmb();
+       desc->status = 0x0;
+}
+
+static int sis190_alloc_rx_skb(struct pci_dev *pdev, struct sk_buff **sk_buff,
+                              struct RxDesc *desc, u32 rx_buf_sz)
+{
+       struct sk_buff *skb;
+       dma_addr_t mapping;
+       int ret = 0;
+
+       skb = dev_alloc_skb(rx_buf_sz);
+       if (!skb)
+               goto err_out;
+
+       *sk_buff = skb;
+
+       mapping = pci_map_single(pdev, skb->data, rx_buf_sz,
+                                PCI_DMA_FROMDEVICE);
+
+       sis190_map_to_asic(desc, mapping, rx_buf_sz);
+out:
+       return ret;
+
+err_out:
+       ret = -ENOMEM;
+       sis190_make_unusable_by_asic(desc);
+       goto out;
+}
+
+static u32 sis190_rx_fill(struct sis190_private *tp, struct net_device *dev,
+                         u32 start, u32 end)
+{
+       u32 cur;
+
+       for (cur = start; cur < end; cur++) {
+               int ret, i = cur % NUM_RX_DESC;
+
+               if (tp->Rx_skbuff[i])
+                       continue;
+
+               ret = sis190_alloc_rx_skb(tp->pci_dev, tp->Rx_skbuff + i,
+                                         tp->RxDescRing + i, tp->rx_buf_sz);
+               if (ret < 0)
+                       break;
+       }
+       return cur - start;
+}
+
+static inline int sis190_try_rx_copy(struct sk_buff **sk_buff, int pkt_size,
+                                    struct RxDesc *desc, int rx_buf_sz)
+{
+       int ret = -1;
+
+       if (pkt_size < rx_copybreak) {
+               struct sk_buff *skb;
+
+               skb = dev_alloc_skb(pkt_size + NET_IP_ALIGN);
+               if (skb) {
+                       skb_reserve(skb, NET_IP_ALIGN);
+                       eth_copy_and_sum(skb, sk_buff[0]->data, pkt_size, 0);
+                       *sk_buff = skb;
+                       sis190_give_to_asic(desc, rx_buf_sz);
+                       ret = 0;
+               }
+       }
+       return ret;
+}
+
+static int sis190_rx_interrupt(struct net_device *dev,
+                              struct sis190_private *tp, void __iomem *ioaddr)
+{
+       struct net_device_stats *stats = &tp->stats;
+       u32 rx_left, cur_rx = tp->cur_rx;
+       u32 delta, count;
+
+       rx_left = NUM_RX_DESC + tp->dirty_rx - cur_rx;
+       rx_left = sis190_rx_quota(rx_left, (u32) dev->quota);
+
+       for (; rx_left > 0; rx_left--, cur_rx++) {
+               unsigned int entry = cur_rx % NUM_RX_DESC;
+               struct RxDesc *desc = tp->RxDescRing + entry;
+               u32 status;
+
+               if (desc->status & OWNbit)
+                       break;
+
+               status = le32_to_cpu(desc->PSize);
+
+               // net_intr(tp, KERN_INFO "%s: Rx PSize = %08x.\n", dev->name,
+               //       status);
+
+               if (status & RxCRC) {
+                       net_intr(tp, KERN_INFO "%s: bad crc. status = %08x.\n",
+                                dev->name, status);
+                       stats->rx_errors++;
+                       stats->rx_crc_errors++;
+                       sis190_give_to_asic(desc, tp->rx_buf_sz);
+               } else if (!(status & PADbit)) {
+                       net_intr(tp, KERN_INFO "%s: bad pad. status = %08x.\n",
+                                dev->name, status);
+                       stats->rx_errors++;
+                       stats->rx_length_errors++;
+                       sis190_give_to_asic(desc, tp->rx_buf_sz);
+               } else {
+                       struct sk_buff *skb = tp->Rx_skbuff[entry];
+                       int pkt_size = (status & RxSizeMask) - 4;
+                       void (*pci_action)(struct pci_dev *, dma_addr_t,
+                               size_t, int) = pci_dma_sync_single_for_device;
+
+                       if (unlikely(pkt_size > tp->rx_buf_sz)) {
+                               net_intr(tp, KERN_INFO
+                                        "%s: (frag) status = %08x.\n",
+                                        dev->name, status);
+                               stats->rx_dropped++;
+                               stats->rx_length_errors++;
+                               sis190_give_to_asic(desc, tp->rx_buf_sz);
+                               continue;
+                       }
+
+                       pci_dma_sync_single_for_cpu(tp->pci_dev,
+                               le32_to_cpu(desc->addr), tp->rx_buf_sz,
+                               PCI_DMA_FROMDEVICE);
+
+                       if (sis190_try_rx_copy(&skb, pkt_size, desc,
+                                              tp->rx_buf_sz)) {
+                               pci_action = pci_unmap_single;
+                               tp->Rx_skbuff[entry] = NULL;
+                               sis190_make_unusable_by_asic(desc);
+                       }
+
+                       pci_action(tp->pci_dev, le32_to_cpu(desc->addr),
+                                  tp->rx_buf_sz, PCI_DMA_FROMDEVICE);
+
+                       skb->dev = dev;
+                       skb_put(skb, pkt_size);
+                       skb->protocol = eth_type_trans(skb, dev);
+
+                       sis190_rx_skb(skb);
+
+                       dev->last_rx = jiffies;
+                       stats->rx_bytes += pkt_size;
+                       stats->rx_packets++;
+               }
+       }
+       count = cur_rx - tp->cur_rx;
+       tp->cur_rx = cur_rx;
+
+       delta = sis190_rx_fill(tp, dev, tp->dirty_rx, tp->cur_rx);
+       if (!delta && count && netif_msg_intr(tp))
+               printk(KERN_INFO "%s: no Rx buffer allocated.\n", dev->name);
+       tp->dirty_rx += delta;
+
+       if (((tp->dirty_rx + NUM_RX_DESC) == tp->cur_rx) && netif_msg_intr(tp))
+               printk(KERN_EMERG "%s: Rx buffers exhausted.\n", dev->name);
+
+       return count;
+}
+
+static void sis190_unmap_tx_skb(struct pci_dev *pdev, struct sk_buff *skb,
+                               struct TxDesc *desc)
+{
+       unsigned int len;
+
+       len = skb->len < ETH_ZLEN ? ETH_ZLEN : skb->len;
+
+       pci_unmap_single(pdev, le32_to_cpu(desc->addr), len, PCI_DMA_TODEVICE);
+
+       memset(desc, 0x00, sizeof(*desc));
+}
+
+static void sis190_tx_interrupt(struct net_device *dev,
+                               struct sis190_private *tp, void __iomem *ioaddr)
+{
+       u32 pending, dirty_tx = tp->dirty_tx;
+       /*
+        * It would not be needed if queueing was allowed to be enabled
+        * again too early (hint: think preempt and unclocked smp systems).
+        */
+       unsigned int queue_stopped;
+
+       smp_rmb();
+       pending = tp->cur_tx - dirty_tx;
+       queue_stopped = (pending == NUM_TX_DESC);
+
+       for (; pending; pending--, dirty_tx++) {
+               unsigned int entry = dirty_tx % NUM_TX_DESC;
+               struct TxDesc *txd = tp->TxDescRing + entry;
+               struct sk_buff *skb;
+
+               if (le32_to_cpu(txd->status) & OWNbit)
+                       break;
+
+               skb = tp->Tx_skbuff[entry];
+
+               tp->stats.tx_packets++;
+               tp->stats.tx_bytes += skb->len;
+
+               sis190_unmap_tx_skb(tp->pci_dev, skb, txd);
+               tp->Tx_skbuff[entry] = NULL;
+               dev_kfree_skb_irq(skb);
+       }
+
+       if (tp->dirty_tx != dirty_tx) {
+               tp->dirty_tx = dirty_tx;
+               smp_wmb();
+               if (queue_stopped)
+                       netif_wake_queue(dev);
+       }
+}
+
+/*
+ * The interrupt handler does all of the Rx thread work and cleans up after
+ * the Tx thread.
+ */
+static irqreturn_t sis190_interrupt(int irq, void *__dev, struct pt_regs *regs)
+{
+       struct net_device *dev = __dev;
+       struct sis190_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       unsigned int handled = 0;
+       u32 status;
+
+       status = SIS_R32(IntrStatus);
+
+       if ((status == 0xffffffff) || !status)
+               goto out;
+
+       handled = 1;
+
+       if (unlikely(!netif_running(dev))) {
+               sis190_asic_down(ioaddr);
+               goto out;
+       }
+
+       SIS_W32(IntrStatus, status);
+
+       // net_intr(tp, KERN_INFO "%s: status = %08x.\n", dev->name, status);
+
+       if (status & LinkChange) {
+               net_intr(tp, KERN_INFO "%s: link change.\n", dev->name);
+               schedule_work(&tp->phy_task);
+       }
+
+       if (status & RxQInt)
+               sis190_rx_interrupt(dev, tp, ioaddr);
+
+       if (status & TxQ0Int)
+               sis190_tx_interrupt(dev, tp, ioaddr);
+out:
+       return IRQ_RETVAL(handled);
+}
+
+static void sis190_free_rx_skb(struct sis190_private *tp,
+                              struct sk_buff **sk_buff, struct RxDesc *desc)
+{
+       struct pci_dev *pdev = tp->pci_dev;
+
+       pci_unmap_single(pdev, le32_to_cpu(desc->addr), tp->rx_buf_sz,
+                        PCI_DMA_FROMDEVICE);
+       dev_kfree_skb(*sk_buff);
+       *sk_buff = NULL;
+       sis190_make_unusable_by_asic(desc);
+}
+
+static void sis190_rx_clear(struct sis190_private *tp)
+{
+       unsigned int i;
+
+       for (i = 0; i < NUM_RX_DESC; i++) {
+               if (!tp->Rx_skbuff[i])
+                       continue;
+               sis190_free_rx_skb(tp, tp->Rx_skbuff + i, tp->RxDescRing + i);
+       }
+}
+
+static void sis190_init_ring_indexes(struct sis190_private *tp)
+{
+       tp->dirty_tx = tp->dirty_rx = tp->cur_tx = tp->cur_rx = 0;
+}
+
+static int sis190_init_ring(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+
+       sis190_init_ring_indexes(tp);
+
+       memset(tp->Tx_skbuff, 0x0, NUM_TX_DESC * sizeof(struct sk_buff *));
+       memset(tp->Rx_skbuff, 0x0, NUM_RX_DESC * sizeof(struct sk_buff *));
+
+       if (sis190_rx_fill(tp, dev, 0, NUM_RX_DESC) != NUM_RX_DESC)
+               goto err_rx_clear;
+
+       sis190_mark_as_last_descriptor(tp->RxDescRing + NUM_RX_DESC - 1);
+
+       return 0;
+
+err_rx_clear:
+       sis190_rx_clear(tp);
+       return -ENOMEM;
+}
+
+static void sis190_set_rx_mode(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       unsigned long flags;
+       u32 mc_filter[2];       /* Multicast hash filter */
+       u16 rx_mode;
+
+       if (dev->flags & IFF_PROMISC) {
+               /* Unconditionally log net taps. */
+               net_drv(tp, KERN_NOTICE "%s: Promiscuous mode enabled.\n",
+                       dev->name);
+               rx_mode =
+                       AcceptBroadcast | AcceptMulticast | AcceptMyPhys |
+                       AcceptAllPhys;
+               mc_filter[1] = mc_filter[0] = 0xffffffff;
+       } else if ((dev->mc_count > multicast_filter_limit) ||
+                  (dev->flags & IFF_ALLMULTI)) {
+               /* Too many to filter perfectly -- accept all multicasts. */
+               rx_mode = AcceptBroadcast | AcceptMulticast | AcceptMyPhys;
+               mc_filter[1] = mc_filter[0] = 0xffffffff;
+       } else {
+               struct dev_mc_list *mclist;
+               unsigned int i;
+
+               rx_mode = AcceptBroadcast | AcceptMyPhys;
+               mc_filter[1] = mc_filter[0] = 0;
+               for (i = 0, mclist = dev->mc_list; mclist && i < dev->mc_count;
+                    i++, mclist = mclist->next) {
+                       int bit_nr =
+                               ether_crc(ETH_ALEN, mclist->dmi_addr) >> 26;
+                       mc_filter[bit_nr >> 5] |= 1 << (bit_nr & 31);
+                       rx_mode |= AcceptMulticast;
+               }
+       }
+
+       spin_lock_irqsave(&tp->lock, flags);
+
+       SIS_W16(RxMacControl, rx_mode | 0x2);
+       SIS_W32(RxHashTable, mc_filter[0]);
+       SIS_W32(RxHashTable + 4, mc_filter[1]);
+
+       spin_unlock_irqrestore(&tp->lock, flags);
+}
+
+static void sis190_soft_reset(void __iomem *ioaddr)
+{
+       SIS_W32(IntrControl, 0x8000);
+       SIS_PCI_COMMIT();
+       msleep(1);
+       SIS_W32(IntrControl, 0x0);
+       sis190_asic_down(ioaddr);
+       msleep(1);
+}
+
+static void sis190_hw_start(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+
+       sis190_soft_reset(ioaddr);
+
+       SIS_W32(TxDescStartAddr, tp->tx_dma);
+       SIS_W32(RxDescStartAddr, tp->rx_dma);
+
+       SIS_W32(IntrStatus, 0xffffffff);
+       SIS_W32(IntrMask, 0x0);
+       /*
+        * Default is 100Mbps.
+        * A bit strange: 100Mbps is 0x1801 elsewhere -- FR 2005/06/09
+        */
+       SIS_W16(StationControl, 0x1901);
+       SIS_W32(GMIIControl, 0x0);
+       SIS_W32(TxMacControl, 0x60);
+       SIS_W16(RxMacControl, 0x02);
+       SIS_W32(RxHashTable, 0x0);
+       SIS_W32(0x6c, 0x0);
+       SIS_W32(RxWakeOnLan, 0x0);
+       SIS_W32(0x74, 0x0);
+
+       SIS_PCI_COMMIT();
+
+       sis190_set_rx_mode(dev);
+
+       /* Enable all known interrupts by setting the interrupt mask. */
+       SIS_W32(IntrMask, sis190_intr_mask);
+
+       SIS_W32(TxControl, 0x1a00 | CmdTxEnb);
+       SIS_W32(RxControl, 0x1a1d);
+
+       netif_start_queue(dev);
+}
+
+static void sis190_phy_task(void * data)
+{
+       struct net_device *dev = data;
+       struct sis190_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       u16 val;
+
+       val = mdio_read(ioaddr, MII_BMCR);
+       if (val & BMCR_RESET) {
+               // FIXME: needlessly high ?  -- FR 02/07/2005
+               mod_timer(&tp->timer, jiffies + HZ/10);
+       } else if (!(mdio_read(ioaddr, MII_BMSR) & BMSR_ANEGCOMPLETE)) {
+               net_link(tp, KERN_WARNING "%s: PHY reset until link up.\n",
+                        dev->name);
+               mdio_write(ioaddr, MII_BMCR, val | BMCR_RESET);
+               mod_timer(&tp->timer, jiffies + SIS190_PHY_TIMEOUT);
+       } else {
+               /* Rejoice ! */
+               struct {
+                       int val;
+                       const char *msg;
+                       u16 ctl;
+               } reg31[] = {
+                       { LPA_1000XFULL | LPA_SLCT,
+                               "1000 Mbps Full Duplex",
+                               0x01 | _1000bpsF },
+                       { LPA_1000XHALF | LPA_SLCT,
+                               "1000 Mbps Half Duplex",
+                               0x01 | _1000bpsH },
+                       { LPA_100FULL,
+                               "100 Mbps Full Duplex",
+                               0x01 | _100bpsF },
+                       { LPA_100HALF,
+                               "100 Mbps Half Duplex",
+                               0x01 | _100bpsH },
+                       { LPA_10FULL,
+                               "10 Mbps Full Duplex",
+                               0x01 | _10bpsF },
+                       { LPA_10HALF,
+                               "10 Mbps Half Duplex",
+                               0x01 | _10bpsH },
+                       { 0, "unknown", 0x0000 }
+               }, *p;
+
+               val = mdio_read(ioaddr, 0x1f);
+               net_link(tp, KERN_INFO "%s: mii ext = %04x.\n", dev->name, val);
+
+               val = mdio_read(ioaddr, MII_LPA);
+               net_link(tp, KERN_INFO "%s: mii lpa = %04x.\n", dev->name, val);
+
+               for (p = reg31; p->ctl; p++) {
+                       if ((val & p->val) == p->val)
+                               break;
+               }
+               if (p->ctl)
+                       SIS_W16(StationControl, p->ctl);
+               net_link(tp, KERN_INFO "%s: link on %s mode.\n", dev->name,
+                        p->msg);
+               netif_carrier_on(dev);
+       }
+}
+
+static void sis190_phy_timer(unsigned long __opaque)
+{
+       struct net_device *dev = (struct net_device *)__opaque;
+       struct sis190_private *tp = netdev_priv(dev);
+
+       if (likely(netif_running(dev)))
+               schedule_work(&tp->phy_task);
+}
+
+static inline void sis190_delete_timer(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+
+       del_timer_sync(&tp->timer);
+}
+
+static inline void sis190_request_timer(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       struct timer_list *timer = &tp->timer;
+
+       init_timer(timer);
+       timer->expires = jiffies + SIS190_PHY_TIMEOUT;
+       timer->data = (unsigned long)dev;
+       timer->function = sis190_phy_timer;
+       add_timer(timer);
+}
+
+static void sis190_set_rxbufsize(struct sis190_private *tp,
+                                struct net_device *dev)
+{
+       unsigned int mtu = dev->mtu;
+
+       tp->rx_buf_sz = (mtu > RX_BUF_SIZE) ? mtu + ETH_HLEN + 8 : RX_BUF_SIZE;
+}
+
+static int sis190_open(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       struct pci_dev *pdev = tp->pci_dev;
+       int rc = -ENOMEM;
+
+       sis190_set_rxbufsize(tp, dev);
+
+       /*
+        * Rx and Tx descriptors need 256 bytes alignment.
+        * pci_alloc_consistent() guarantees a stronger alignment.
+        */
+       tp->TxDescRing = pci_alloc_consistent(pdev, TX_RING_BYTES, &tp->tx_dma);
+       if (!tp->TxDescRing)
+               goto out;
+
+       tp->RxDescRing = pci_alloc_consistent(pdev, RX_RING_BYTES, &tp->rx_dma);
+       if (!tp->RxDescRing)
+               goto err_free_tx_0;
+
+       rc = sis190_init_ring(dev);
+       if (rc < 0)
+               goto err_free_rx_1;
+
+       INIT_WORK(&tp->phy_task, sis190_phy_task, dev);
+
+       sis190_request_timer(dev);
+
+       rc = request_irq(dev->irq, sis190_interrupt, SA_SHIRQ, dev->name, dev);
+       if (rc < 0)
+               goto err_release_timer_2;
+
+       sis190_hw_start(dev);
+out:
+       return rc;
+
+err_release_timer_2:
+       sis190_delete_timer(dev);
+       sis190_rx_clear(tp);
+err_free_rx_1:
+       pci_free_consistent(tp->pci_dev, RX_RING_BYTES, tp->RxDescRing,
+               tp->rx_dma);
+err_free_tx_0:
+       pci_free_consistent(tp->pci_dev, TX_RING_BYTES, tp->TxDescRing,
+               tp->tx_dma);
+       goto out;
+}
+
+static void sis190_tx_clear(struct sis190_private *tp)
+{
+       unsigned int i;
+
+       for (i = 0; i < NUM_TX_DESC; i++) {
+               struct sk_buff *skb = tp->Tx_skbuff[i];
+
+               if (!skb)
+                       continue;
+
+               sis190_unmap_tx_skb(tp->pci_dev, skb, tp->TxDescRing + i);
+               tp->Tx_skbuff[i] = NULL;
+               dev_kfree_skb(skb);
+
+               tp->stats.tx_dropped++;
+       }
+       tp->cur_tx = tp->dirty_tx = 0;
+}
+
+static void sis190_down(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       unsigned int poll_locked = 0;
+
+       sis190_delete_timer(dev);
+
+       netif_stop_queue(dev);
+
+       flush_scheduled_work();
+
+       do {
+               spin_lock_irq(&tp->lock);
+
+               sis190_asic_down(ioaddr);
+
+               spin_unlock_irq(&tp->lock);
+
+               synchronize_irq(dev->irq);
+
+               if (!poll_locked) {
+                       netif_poll_disable(dev);
+                       poll_locked++;
+               }
+
+               synchronize_sched();
+
+       } while (SIS_R32(IntrMask));
+
+       sis190_tx_clear(tp);
+       sis190_rx_clear(tp);
+}
+
+static int sis190_close(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       struct pci_dev *pdev = tp->pci_dev;
+
+       sis190_down(dev);
+
+       free_irq(dev->irq, dev);
+
+       netif_poll_enable(dev);
+
+       pci_free_consistent(pdev, TX_RING_BYTES, tp->TxDescRing, tp->tx_dma);
+       pci_free_consistent(pdev, RX_RING_BYTES, tp->RxDescRing, tp->rx_dma);
+
+       tp->TxDescRing = NULL;
+       tp->RxDescRing = NULL;
+
+       return 0;
+}
+
+static int sis190_start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       u32 len, entry, dirty_tx;
+       struct TxDesc *desc;
+       dma_addr_t mapping;
+
+       if (unlikely(skb->len < ETH_ZLEN)) {
+               skb = skb_padto(skb, ETH_ZLEN);
+               if (!skb) {
+                       tp->stats.tx_dropped++;
+                       goto out;
+               }
+               len = ETH_ZLEN;
+       } else {
+               len = skb->len;
+       }
+
+       entry = tp->cur_tx % NUM_TX_DESC;
+       desc = tp->TxDescRing + entry;
+
+       if (unlikely(le32_to_cpu(desc->status) & OWNbit)) {
+               netif_stop_queue(dev);
+               net_tx_err(tp, KERN_ERR PFX
+                          "%s: BUG! Tx Ring full when queue awake!\n",
+                          dev->name);
+               return NETDEV_TX_BUSY;
+       }
+
+       mapping = pci_map_single(tp->pci_dev, skb->data, len, PCI_DMA_TODEVICE);
+
+       tp->Tx_skbuff[entry] = skb;
+
+       desc->PSize = cpu_to_le32(len);
+       desc->addr = cpu_to_le32(mapping);
+
+       desc->size = cpu_to_le32(len);
+       if (entry == (NUM_TX_DESC - 1))
+               desc->size |= cpu_to_le32(RingEnd);
+
+       wmb();
+
+       desc->status = cpu_to_le32(OWNbit | INTbit | DEFbit | CRCbit | PADbit);
+
+       tp->cur_tx++;
+
+       smp_wmb();
+
+       SIS_W32(TxControl, 0x1a00 | CmdReset | CmdTxEnb);
+
+       dev->trans_start = jiffies;
+
+       dirty_tx = tp->dirty_tx;
+       if ((tp->cur_tx - NUM_TX_DESC) == dirty_tx) {
+               netif_stop_queue(dev);
+               smp_rmb();
+               if (dirty_tx != tp->dirty_tx)
+                       netif_wake_queue(dev);
+       }
+out:
+       return NETDEV_TX_OK;
+}
+
+static struct net_device_stats *sis190_get_stats(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+
+       return &tp->stats;
+}
+
+static void sis190_release_board(struct pci_dev *pdev)
+{
+       struct net_device *dev = pci_get_drvdata(pdev);
+       struct sis190_private *tp = netdev_priv(dev);
+
+       iounmap(tp->mmio_addr);
+       pci_release_regions(pdev);
+       pci_disable_device(pdev);
+       free_netdev(dev);
+}
+
+static struct net_device * __devinit sis190_init_board(struct pci_dev *pdev)
+{
+       struct sis190_private *tp;
+       struct net_device *dev;
+       void __iomem *ioaddr;
+       int rc;
+
+       dev = alloc_etherdev(sizeof(*tp));
+       if (!dev) {
+               net_drv(&debug, KERN_ERR PFX "unable to alloc new ethernet\n");
+               rc = -ENOMEM;
+               goto err_out_0;
+       }
+
+       SET_MODULE_OWNER(dev);
+       SET_NETDEV_DEV(dev, &pdev->dev);
+
+       tp = netdev_priv(dev);
+       tp->msg_enable = netif_msg_init(debug.msg_enable, SIS190_MSG_DEFAULT);
+
+       rc = pci_enable_device(pdev);
+       if (rc < 0) {
+               net_probe(tp, KERN_ERR "%s: enable failure\n", pci_name(pdev));
+               goto err_free_dev_1;
+       }
+
+       rc = -ENODEV;
+
+       if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
+               net_probe(tp, KERN_ERR "%s: region #0 is no MMIO resource.\n",
+                         pci_name(pdev));
+               goto err_pci_disable_2;
+       }
+       if (pci_resource_len(pdev, 0) < SIS190_REGS_SIZE) {
+               net_probe(tp, KERN_ERR "%s: invalid PCI region size(s).\n",
+                         pci_name(pdev));
+               goto err_pci_disable_2;
+       }
+
+       rc = pci_request_regions(pdev, DRV_NAME);
+       if (rc < 0) {
+               net_probe(tp, KERN_ERR PFX "%s: could not request regions.\n",
+                         pci_name(pdev));
+               goto err_pci_disable_2;
+       }
+
+       rc = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
+       if (rc < 0) {
+               net_probe(tp, KERN_ERR "%s: DMA configuration failed.\n",
+                         pci_name(pdev));
+               goto err_free_res_3;
+       }
+
+       pci_set_master(pdev);
+
+       ioaddr = ioremap(pci_resource_start(pdev, 0), SIS190_REGS_SIZE);
+       if (!ioaddr) {
+               net_probe(tp, KERN_ERR "%s: cannot remap MMIO, aborting\n",
+                         pci_name(pdev));
+               rc = -EIO;
+               goto err_free_res_3;
+       }
+
+       tp->pci_dev = pdev;
+       tp->mmio_addr = ioaddr;
+
+       sis190_irq_mask_and_ack(ioaddr);
+
+       sis190_soft_reset(ioaddr);
+out:
+       return dev;
+
+err_free_res_3:
+       pci_release_regions(pdev);
+err_pci_disable_2:
+       pci_disable_device(pdev);
+err_free_dev_1:
+       free_netdev(dev);
+err_out_0:
+       dev = ERR_PTR(rc);
+       goto out;
+}
+
+static void sis190_tx_timeout(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       u8 tmp8;
+
+       /* Disable Tx, if not already */
+       tmp8 = SIS_R8(TxControl);
+       if (tmp8 & CmdTxEnb)
+               SIS_W8(TxControl, tmp8 & ~CmdTxEnb);
+
+       /* Disable interrupts by clearing the interrupt mask. */
+       SIS_W32(IntrMask, 0x0000);
+
+       /* Stop a shared interrupt from scavenging while we are. */
+       spin_lock_irq(&tp->lock);
+       sis190_tx_clear(tp);
+       spin_unlock_irq(&tp->lock);
+
+       /* ...and finally, reset everything. */
+       sis190_hw_start(dev);
+
+       netif_wake_queue(dev);
+}
+
+static void sis190_set_speed_auto(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       int val;
+
+       net_link(tp, KERN_INFO "%s: Enabling Auto-negotiation.\n", dev->name);
+
+       val = mdio_read(ioaddr, MII_ADVERTISE);
+
+       // Enable 10/100 Full/Half Mode, leave MII_ADVERTISE bit4:0
+       // unchanged.
+       mdio_write(ioaddr, MII_ADVERTISE, (val & ADVERTISE_SLCT) |
+                  ADVERTISE_100FULL | ADVERTISE_10FULL |
+                  ADVERTISE_100HALF | ADVERTISE_10HALF);
+
+       // Enable 1000 Full Mode.
+       mdio_write(ioaddr, MII_CTRL1000, ADVERTISE_1000FULL);
+
+       // Enable auto-negotiation and restart auto-negotiation.
+       mdio_write(ioaddr, MII_BMCR,
+                  BMCR_ANENABLE | BMCR_ANRESTART | BMCR_RESET);
+}
+
+static void sis190_get_drvinfo(struct net_device *dev,
+                              struct ethtool_drvinfo *info)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+
+       strcpy(info->driver, DRV_NAME);
+       strcpy(info->version, DRV_VERSION);
+       strcpy(info->bus_info, pci_name(tp->pci_dev));
+}
+
+static int sis190_get_regs_len(struct net_device *dev)
+{
+       return SIS190_REGS_SIZE;
+}
+
+static void sis190_get_regs(struct net_device *dev, struct ethtool_regs *regs,
+                           void *p)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+       unsigned long flags;
+
+       if (regs->len > SIS190_REGS_SIZE)
+               regs->len = SIS190_REGS_SIZE;
+
+       spin_lock_irqsave(&tp->lock, flags);
+       memcpy_fromio(p, tp->mmio_addr, regs->len);
+       spin_unlock_irqrestore(&tp->lock, flags);
+}
+
+static u32 sis190_get_msglevel(struct net_device *dev)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+
+       return tp->msg_enable;
+}
+
+static void sis190_set_msglevel(struct net_device *dev, u32 value)
+{
+       struct sis190_private *tp = netdev_priv(dev);
+
+       tp->msg_enable = value;
+}
+
+static struct ethtool_ops sis190_ethtool_ops = {
+       .get_drvinfo    = sis190_get_drvinfo,
+       .get_regs_len   = sis190_get_regs_len,
+       .get_regs       = sis190_get_regs,
+       .get_link       = ethtool_op_get_link,
+       .get_msglevel   = sis190_get_msglevel,
+       .set_msglevel   = sis190_set_msglevel,
+};
+
+static int __devinit sis190_init_one(struct pci_dev *pdev,
+                                    const struct pci_device_id *ent)
+{
+       static int printed_version = 0;
+       struct sis190_private *tp;
+       struct net_device *dev;
+       void __iomem *ioaddr;
+       int i, rc;
+
+       if (!printed_version) {
+               net_drv(&debug, KERN_INFO SIS190_DRIVER_NAME " loaded.\n");
+               printed_version = 1;
+       }
+
+       dev = sis190_init_board(pdev);
+       if (IS_ERR(dev)) {
+               rc = PTR_ERR(dev);
+               goto out;
+       }
+
+       tp = netdev_priv(dev);
+       ioaddr = tp->mmio_addr;
+
+       /* Get MAC address */
+       /* Read node address from the EEPROM */
+
+       if (SIS_R32(ROMControl) & 0x4) {
+               for (i = 0; i < 3; i++) {
+                       SIS_W16(RxMacAddr + 2*i,
+                               sis190_read_eeprom(ioaddr, 3 + i));
+               }
+       }
+
+       for (i = 0; i < MAC_ADDR_LEN; i++)
+               dev->dev_addr[i] = SIS_R8(RxMacAddr + i);
+
+       INIT_WORK(&tp->phy_task, sis190_phy_task, dev);
+
+       dev->open = sis190_open;
+       dev->stop = sis190_close;
+       dev->get_stats = sis190_get_stats;
+       dev->tx_timeout = sis190_tx_timeout;
+       dev->watchdog_timeo = SIS190_TX_TIMEOUT;
+       dev->hard_start_xmit = sis190_start_xmit;
+       dev->set_multicast_list = sis190_set_rx_mode;
+       SET_ETHTOOL_OPS(dev, &sis190_ethtool_ops);
+       dev->irq = pdev->irq;
+       dev->base_addr = (unsigned long) 0xdead;
+
+       spin_lock_init(&tp->lock);
+       rc = register_netdev(dev);
+       if (rc < 0) {
+               sis190_release_board(pdev);
+               goto out;
+       }
+
+       pci_set_drvdata(pdev, dev);
+
+       net_probe(tp, KERN_INFO "%s: %s at %p (IRQ: %d), "
+              "%2.2x:%2.2x:%2.2x:%2.2x:%2.2x:%2.2x\n",
+              pci_name(pdev), sis_chip_info[ent->driver_data].name,
+              ioaddr, dev->irq,
+              dev->dev_addr[0], dev->dev_addr[1],
+              dev->dev_addr[2], dev->dev_addr[3],
+              dev->dev_addr[4], dev->dev_addr[5]);
+
+       netif_carrier_off(dev);
+
+       sis190_set_speed_auto(dev);
+out:
+       return rc;
+}
+
+static void __devexit sis190_remove_one(struct pci_dev *pdev)
+{
+       struct net_device *dev = pci_get_drvdata(pdev);
+
+       unregister_netdev(dev);
+       sis190_release_board(pdev);
+       pci_set_drvdata(pdev, NULL);
+}
+
+static struct pci_driver sis190_pci_driver = {
+       .name           = DRV_NAME,
+       .id_table       = sis190_pci_tbl,
+       .probe          = sis190_init_one,
+       .remove         = __devexit_p(sis190_remove_one),
+};
+
+static int __init sis190_init_module(void)
+{
+       return pci_module_init(&sis190_pci_driver);
+}
+
+static void __exit sis190_cleanup_module(void)
+{
+       pci_unregister_driver(&sis190_pci_driver);
+}
+
+module_init(sis190_init_module);
+module_exit(sis190_cleanup_module);