patch-2.4.22 linux-2.4.22/drivers/usb/ax8817x.c

Next file: linux-2.4.22/drivers/usb/catc.c
Previous file: linux-2.4.22/drivers/usb/aiptek.c
Back to the patch index
Back to the overall index

diff -urN linux-2.4.21/drivers/usb/ax8817x.c linux-2.4.22/drivers/usb/ax8817x.c
@@ -0,0 +1,1291 @@
+/*
+ * ASIX AX8817x USB 2.0 10/100/HomePNA Ethernet controller driver
+ *
+ * $Id: ax8817x.c,v 1.15 2003/06/15 18:45:21 dhollis Exp $
+ *
+ * Copyright (c) 2002-2003 TiVo Inc.
+ *
+ * This software may be used and distributed according to the terms
+ * of the GNU General Public License, incorporated herein by reference.
+ *
+ * History 
+ *
+ *	2003-06-28 - Dave Hollis <dhollis@davehollis.com>  1.0.2
+ *		* Added support for Intellinet
+ *	2003-06-12 - Dave Hollis <dhollis@davehollis.com>  1.0.1
+ *		* use usb_make_path for ethtool info
+ *		* Use crc32.h for crc functions
+ *		* Additional callback cases for other host ctrlrs
+ *		* Force minimum default rcv buffers
+ *
+ *	2003-06-12 - Dave Hollis <dhollis@davehollis.com>  1.0.0
+ *		* Backport removal of multiple tx_urb / lengthy 
+ *		  start_xmit routines, etc.
+ *		* ethtool driver name returns driver name, not
+ *		  long description
+ *
+ *	2003-06-05 - Dave Hollis <dhollis@davehollis.com>  0.9.9
+ *		* Cleanup unnecessary #if 0
+ *		* Fix coding style to match kernel style
+ *
+ *	2003-05-31 - Dave Hollis <dhollis@davehollis.com>  0.9.8
+ *		* Don't stop/start the queue in start_xmit
+ *		* Swallow URB status upon hard removal
+ *		* Cleanup remaining comments (kill // style)
+ *
+ *	2003-05-29 - Dave Hollis <dhollis@davehollis.com>  0.9.7
+ *		* Set module owner
+ *		* Follow-up on suggestions from David Brownell &
+ *		  Oliver Neukum which should help with robustness
+ *		* Use ether_crc from stock kernel if available
+ *
+ *	2003-05-28 - Dave Hollis <dhollis@davehollis.com>  0.9.6
+ *		* Added basic ethtool & mii support
+ *
+ *	2003-05-28 - Dave Hollis <dhollis@davehollis.com>  0.9.5
+ *		* Workout devrequest change to usb_ctrlrequest structure
+ *		* Replace FILL_BULK_URB macros to non-deprecated 
+ *		  usb_fill_bulk_urb macros
+ *		* Replace printks with equivalent macros
+ *		* Use defines for module description, version, author to
+ *		  simplify future changes
+ *
+ * Known Issues
+ *	usb-uhci.c: process_transfer: fixed toggle message to console
+ *		Possible bug in usb-uhci or bug in this driver that
+ *		makes it spit that out.  Doesn't seem to have harmful
+ *		effects.
+ *
+ * Todo
+ *	Fix mii/ethtool output
+*/
+
+#include <linux/slab.h>
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/usb.h>
+
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/ethtool.h>
+#include <linux/skbuff.h>
+#include <linux/mii.h>
+#include <asm/uaccess.h>
+#include <linux/crc32.h>
+
+/* Version Information */
+#define DRIVER_VERSION "v1.0.0"
+#define DRIVER_AUTHOR "TiVo, Inc."
+#define DRIVER_DESC "ASIX AX8817x USB Ethernet driver"
+
+MODULE_DESCRIPTION(DRIVER_DESC);
+MODULE_AUTHOR(DRIVER_AUTHOR);
+MODULE_LICENSE("GPL");
+
+
+#define AX_REQ_READ         ( USB_DIR_IN | USB_TYPE_VENDOR | USB_RECIP_DEVICE )
+#define AX_REQ_WRITE        ( USB_DIR_OUT | USB_TYPE_VENDOR | USB_RECIP_DEVICE )
+
+#define AX_CMD_SET_SW_MII           0x06
+#define AX_CMD_READ_MII_REG         0x07
+#define AX_CMD_WRITE_MII_REG        0x08
+#define AX_CMD_SET_HW_MII           0x0a
+#define AX_CMD_WRITE_RX_CTL         0x10
+#define AX_CMD_WRITE_MULTI_FILTER   0x16
+#define AX_CMD_READ_NODE_ID         0x17
+#define AX_CMD_READ_PHY_ID          0x19
+#define AX_CMD_WRITE_MEDIUM_MODE    0x1b
+#define AX_CMD_WRITE_GPIOS          0x1f
+
+#define AX_RX_MAX                   ETH_FRAME_LEN
+#define AX_TIMEOUT_CMD              ( HZ / 10 )
+#define AX_TIMEOUT_TX               ( HZ * 2 )
+#define AX_MAX_MCAST                64
+
+#define AX_DRV_STATE_INITIALIZING   0x00
+#define AX_DRV_STATE_RUNNING        0x01
+#define AX_DRV_STATE_EXITING        0x02
+
+#define AX_PHY_STATE_INITIALIZING   0x00
+#define AX_PHY_STATE_NO_LINK        0x01
+#define AX_PHY_STATE_POLLING_1      0x02
+#define AX_PHY_STATE_POLLING_2      0x03
+#define AX_PHY_STATE_POLLING_3      0x04
+#define AX_PHY_STATE_POLLING_4      0x05
+#define AX_PHY_STATE_SETTING_MAC    0x06
+#define AX_PHY_STATE_LINK           0x07
+#define AX_PHY_STATE_ABORT_POLL     0x08
+#define AX_PHY_STATE_ABORTING       0x09
+
+#define AX_MAX_PHY_RETRY            50
+
+#define AX_RX_URBS_DEFAULT	    2
+
+static const char driver_name[] = "ax8817x";
+static int n_rx_urbs = AX_RX_URBS_DEFAULT;
+
+MODULE_PARM(n_rx_urbs, "i");
+MODULE_PARM_DESC(n_rx_urbs,
+		 "Number of rx buffers to queue at once (def 2)");
+
+struct ax8817x_info;
+struct ax_cmd_req;
+typedef int (*ax_cmd_callback_t) (struct ax8817x_info *,
+				  struct ax_cmd_req *);
+
+struct ax_cmd_req {
+	struct list_head list;
+	ax_cmd_callback_t cmd_callback;
+	void *priv;
+	int status;
+	void *data;
+	int data_size;
+	int timeout;
+	struct usb_ctrlrequest devreq;
+};
+
+struct ax8817x_info {
+	struct usb_device *usb;
+	struct net_device *net;
+	struct urb **rx_urbs;
+	struct urb *tx_urb;
+	struct urb *int_urb;
+	u8 *int_buf;
+	struct urb *ctl_urb;
+	struct list_head ctl_queue;
+	spinlock_t ctl_lock;
+	atomic_t rx_refill_cnt;
+	int tx_head;
+	int tx_tail;
+	spinlock_t tx_lock;
+	struct net_device_stats stats;
+	struct ax_cmd_req phy_req;
+	u8 phy_id;
+	u8 phy_state;
+	u8 drv_state;
+};
+
+
+const struct usb_device_id ax8817x_id_table[] __devinitdata = {
+	/* Linksys USB200M */
+      {USB_DEVICE(0x077b, 0x2226), driver_info:0x00130103},
+	/* Hawking UF200, TRENDnet TU2-ET100 */
+      {USB_DEVICE(0x07b8, 0x420a), driver_info:0x001f1d1f},
+	/* NETGEAR FA120 */
+      {USB_DEVICE(0x0846, 0x1040), driver_info:0x00130103},
+	/* D-Link DUB-E100 */
+      {USB_DEVICE(0x2001, 0x1a00), driver_info:0x009f9d9f},
+	/* Intellinet USB Ethernet*/
+      {USB_DEVICE(0x0b95, 0x1720), driver_info:0x00130103},
+
+	{}
+};
+
+MODULE_DEVICE_TABLE(usb, ax8817x_id_table);
+
+static void ax_run_ctl_queue(struct ax8817x_info *, struct ax_cmd_req *,
+			     int);
+static void ax_rx_callback(struct urb *urb);
+
+static void ax_ctl_callback(struct urb *urb)
+{
+	struct ax8817x_info *ax_info =
+	    (struct ax8817x_info *) urb->context;
+
+	ax_run_ctl_queue(ax_info, NULL,
+			 urb->status ? urb->status : urb->actual_length);
+}
+
+/* 
+ * Queue a new ctl request, or dequeue the first in the list
+*/
+static void ax_run_ctl_queue(struct ax8817x_info *ax_info,
+			     struct ax_cmd_req *req, int status)
+{
+	struct ax_cmd_req *next_req = NULL;
+	struct ax_cmd_req *last_req = NULL;
+	unsigned long flags;
+
+	/* Need to lock around queue list manipulation */
+	spin_lock_irqsave(&ax_info->ctl_lock, flags);
+
+	if (req == NULL) {
+		last_req =
+		    list_entry(ax_info->ctl_queue.next, struct ax_cmd_req,
+			       list);
+	} else {
+		if (list_empty(&ax_info->ctl_queue)) {
+			next_req = req;
+		}
+
+		req->status = -EINPROGRESS;
+		list_add_tail(&req->list, &ax_info->ctl_queue);
+	}
+
+	while (1) {
+		if (last_req != NULL) {
+			/* dequeue completed entry */
+			list_del(&last_req->list);
+
+			last_req->status = status;
+			if (last_req->cmd_callback(ax_info, last_req)) {
+				/* requeue if told to do so */
+				last_req->status = -EINPROGRESS;
+				list_add_tail(&last_req->list,
+					      &ax_info->ctl_queue);
+			}
+
+			if (list_empty(&ax_info->ctl_queue)) {
+				next_req = NULL;
+			} else {
+				next_req =
+				    list_entry(ax_info->ctl_queue.next,
+					       struct ax_cmd_req, list);
+			}
+		}
+
+		spin_unlock_irqrestore(&ax_info->ctl_lock, flags);
+
+		if (next_req == NULL) {
+			break;
+		}
+
+		/* XXX: do something with timeout */
+		usb_fill_control_urb(ax_info->ctl_urb, ax_info->usb,
+				     next_req->devreq.
+				     bRequestType & USB_DIR_IN ?
+				     usb_rcvctrlpipe(ax_info->usb,
+						     0) :
+				     usb_sndctrlpipe(ax_info->usb, 0),
+				     (void *) &next_req->devreq,
+				     next_req->data, next_req->data_size,
+				     ax_ctl_callback, ax_info);
+
+		status = usb_submit_urb(ax_info->ctl_urb);
+		if (status >= 0) {
+			break;
+		}
+
+		last_req = next_req;
+
+		spin_lock_irqsave(&ax_info->ctl_lock, flags);
+	}
+}
+
+static int ax_sync_cmd_callback(struct ax8817x_info *unused,
+				struct ax_cmd_req *req)
+{
+	wait_queue_head_t *wq = (wait_queue_head_t *) req->priv;
+
+	wake_up(wq);
+
+	return 0;
+}
+
+static int ax_async_cmd_callback(struct ax8817x_info *unused,
+				 struct ax_cmd_req *req)
+{
+	if (req->status < 0) {
+		err("%s: Async command %d failed: %d\n", __FUNCTION__,
+		    req->devreq.bRequest, req->status);
+	}
+
+	/* Nothing else to do here, just need to free the request (and its
+	   allocated data) */
+	if (req->data != NULL) {
+		kfree(req->data);
+	}
+	kfree(req);
+
+	return 0;
+}
+
+/*
+ * This is mostly the same as usb_control_msg(), except that it is able
+ * to queue control messages
+*/
+static int ax_control_msg(struct ax8817x_info *ax_info, u8 requesttype,
+			  u8 request, u16 value, u16 index, void *data,
+			  u16 size, int timeout)
+{
+	struct ax_cmd_req *req;
+	DECLARE_WAIT_QUEUE_HEAD(wq);
+	DECLARE_WAITQUEUE(wait, current);
+	int ret;
+
+	req = kmalloc(sizeof(struct ax_cmd_req), GFP_KERNEL);
+	if (req == NULL) {
+		return -ENOMEM;
+	}
+
+	req->devreq.bRequestType = requesttype;
+	req->devreq.bRequest = request;
+	req->devreq.wValue = cpu_to_le16(value);
+	req->devreq.wIndex = cpu_to_le16(index);
+	req->devreq.wLength = cpu_to_le16(size);
+	req->data = data;
+	req->data_size = size;
+	req->timeout = timeout;
+
+	req->priv = &wq;
+	set_current_state(TASK_UNINTERRUPTIBLE);
+	add_wait_queue(&wq, &wait);
+
+	req->cmd_callback = ax_sync_cmd_callback;
+
+	ax_run_ctl_queue(ax_info, req, 0);
+	schedule();
+
+	ret = req->status;
+
+	kfree(req);
+
+	return ret;
+}
+
+/*
+ * Same, but can be used asynchronously, may fail, and returns no exit
+ * status
+*/
+static void ax_control_msg_async(struct ax8817x_info *ax_info,
+				 u8 requesttype, u8 request, u16 value,
+				 u16 index, void *data, u16 size,
+				 int timeout)
+{
+	struct ax_cmd_req *req;
+
+	req = kmalloc(sizeof(struct ax_cmd_req), GFP_ATOMIC);
+	if (req == NULL) {
+		/* There's not much else we can do here... */
+		err("%s: Failed alloc\n", __FUNCTION__);
+		return;
+	}
+
+	req->devreq.bRequestType = requesttype;
+	req->devreq.bRequest = request;
+	req->devreq.wValue = cpu_to_le16(value);
+	req->devreq.wIndex = cpu_to_le16(index);
+	req->devreq.wLength = cpu_to_le16(size);
+	req->data = data;
+	req->data_size = size;
+	req->timeout = timeout;
+
+	req->cmd_callback = ax_async_cmd_callback;
+
+	ax_run_ctl_queue(ax_info, req, 0);
+}
+
+static inline int ax_read_cmd(struct ax8817x_info *ax_info, u8 cmd,
+			      u16 value, u16 index, u16 size, void *data)
+{
+	return ax_control_msg(ax_info, AX_REQ_READ, cmd, value, index,
+			      data, size, AX_TIMEOUT_CMD);
+}
+
+static inline int ax_write_cmd(struct ax8817x_info *ax_info, u8 cmd,
+			       u16 value, u16 index, u16 size, void *data)
+{
+	return ax_control_msg(ax_info, AX_REQ_WRITE, cmd, value, index,
+			      data, size, AX_TIMEOUT_CMD);
+}
+
+static inline void ax_write_cmd_async(struct ax8817x_info *ax_info, u8 cmd,
+				      u16 value, u16 index, u16 size,
+				      void *data)
+{
+	ax_control_msg_async(ax_info, AX_REQ_WRITE, cmd, value, index,
+			     data, size, AX_TIMEOUT_CMD);
+}
+
+static int ax_refill_rx_urb(struct ax8817x_info *ax_info, struct urb *urb)
+{
+	struct sk_buff *skb;
+	int ret;
+
+	skb = dev_alloc_skb(AX_RX_MAX + 2);
+	if (skb != NULL) {
+		skb_reserve(skb, 2);	/* for IP header alignment */
+		skb->dev = ax_info->net;
+
+		usb_fill_bulk_urb(urb, ax_info->usb,
+				  usb_rcvbulkpipe(ax_info->usb, 3),
+				  skb->data, AX_RX_MAX, ax_rx_callback,
+				  skb);
+
+		ret = usb_submit_urb(urb);
+		if (ret < 0) {
+			err("Failed submit rx URB (%d)\n", ret);
+			dev_kfree_skb_irq(skb);
+			urb->context = NULL;
+		} else {
+			ret = 0;
+		}
+	} else {
+		/* this just means we're low on memory at the moment. Try to
+		   handle it gracefully. */
+		urb->context = NULL;
+		ret = 1;
+	}
+
+	return ret;
+}
+
+static int ax_phy_cmd_callback(struct ax8817x_info *ax_info,
+			       struct ax_cmd_req *req)
+{
+	int full_duplex;
+	int flow_control;
+	u16 mii_data_le;
+
+	if (req->status < 0) {
+		err("%s: Failed at state %d: %d\n", __FUNCTION__,
+		    ax_info->phy_state, req->status);
+		/* Not sure what else we can do, so just bail */
+		ax_info->phy_state = AX_PHY_STATE_ABORTING;
+	}
+
+	switch (ax_info->phy_state) {
+		/* Now that we're in software MII mode, read the BMSR */
+	case AX_PHY_STATE_POLLING_1:
+		ax_info->phy_state = AX_PHY_STATE_POLLING_2;
+		req->devreq.bRequestType = AX_REQ_READ;
+		req->devreq.bRequest = AX_CMD_READ_MII_REG;
+		req->devreq.wValue = cpu_to_le16(ax_info->phy_id);
+		req->devreq.wIndex = cpu_to_le16(MII_BMSR);
+		req->devreq.wLength = cpu_to_le16(2);
+		req->data_size = 2;
+		(long) req->priv = 0;	/* This is the retry count */
+		return 1;
+
+		/* Done reading BMSR */
+	case AX_PHY_STATE_POLLING_2:
+		mii_data_le = *(u16 *) req->data;
+		if ((mii_data_le &
+		     cpu_to_le16(BMSR_LSTATUS | BMSR_ANEGCAPABLE))
+		    == cpu_to_le16(BMSR_LSTATUS | BMSR_ANEGCAPABLE)) {
+			if (mii_data_le & cpu_to_le16(BMSR_ANEGCOMPLETE)) {
+				/* Autonegotiation done, go on to read LPA */
+				ax_info->phy_state =
+				    AX_PHY_STATE_POLLING_3;
+				req->devreq.wIndex = cpu_to_le16(MII_LPA);
+				return 1;
+			} else if ((long) req->priv++ < AX_MAX_PHY_RETRY) {
+				/* Reread BMSR if it's still autonegotiating. This is
+				   probably unnecessary logic, I've never seen it take
+				   more than 1 try... */
+				return 1;
+			}
+			/* else fall through to abort */
+		}
+		/* XXX: should probably handle auto-neg failure better,
+		   by reverting to manual setting of something safe. (?) */
+
+		ax_info->phy_state = AX_PHY_STATE_ABORT_POLL;
+		/* and then fall through to set hw MII */
+
+		/* Got what we needed from PHY, set back to hardware MII mode
+		   (Do same for abort in mid-poll) */
+	case AX_PHY_STATE_POLLING_3:
+	case AX_PHY_STATE_ABORT_POLL:
+		ax_info->phy_state += 1;
+		req->devreq.bRequestType = AX_REQ_WRITE;
+		req->devreq.bRequest = AX_CMD_SET_HW_MII;
+		req->devreq.wValue = cpu_to_le16(0);
+		req->devreq.wIndex = cpu_to_le16(0);
+		req->devreq.wLength = cpu_to_le16(0);
+		req->data_size = 0;
+		return 1;
+
+		/* The end result, set the right duplex and flow control mode in the
+		   MAC (based on the PHY's LPA reg, which should still be in the data
+		   buffer) */
+	case AX_PHY_STATE_POLLING_4:
+		mii_data_le = *(u16 *) req->data;
+		ax_info->phy_state = AX_PHY_STATE_SETTING_MAC;
+		req->devreq.bRequest = AX_CMD_WRITE_MEDIUM_MODE;
+		full_duplex = mii_data_le & cpu_to_le16(LPA_DUPLEX);
+		flow_control = full_duplex &&
+		    (mii_data_le & cpu_to_le16(0x0400));
+		req->devreq.wValue = cpu_to_le16(0x04) |
+		    (full_duplex ? cpu_to_le16(0x02) : 0) |
+		    (flow_control ? cpu_to_le16(0x10) : 0);
+		info("%s: Link established, %s duplex, flow control %sabled\n", ax_info->net->name, full_duplex ? "full" : "half", flow_control ? "en" : "dis");
+		return 1;
+
+		/* All done */
+	case AX_PHY_STATE_SETTING_MAC:
+		ax_info->phy_state = AX_PHY_STATE_LINK;
+		netif_carrier_on(ax_info->net);
+		return 0;
+
+	default:
+		err("%s: Unknown state %d\n", __FUNCTION__,
+		    ax_info->phy_state);
+		/* fall through */
+	case AX_PHY_STATE_ABORTING:
+		ax_info->phy_state = AX_PHY_STATE_NO_LINK;
+		return 0;
+	}
+}
+
+static void ax_int_callback(struct urb *urb)
+{
+	struct ax8817x_info *ax_info =
+	    (struct ax8817x_info *) urb->context;
+	u8 phy_link;
+
+	if (ax_info->drv_state == AX_DRV_STATE_EXITING ||
+	    urb->actual_length < 3) {
+		return;
+	}
+
+	/* Ignore the first PHY link report, it will sometimes be reported as
+	   link active, even though we just told the PHY to reset. If it
+	   really has link, we'll pick it up next int callback.
+	 */
+	if (ax_info->phy_state == AX_PHY_STATE_INITIALIZING) {
+		netif_carrier_off(ax_info->net);
+		ax_info->phy_state = AX_PHY_STATE_NO_LINK;
+		return;
+	}
+
+	/* Assume we're only interested in the primary PHY for now. */
+	phy_link = ax_info->int_buf[2] & 1;
+
+	if (phy_link ==
+	    (ax_info->phy_state == AX_PHY_STATE_NO_LINK) ? 0 : 1) {
+		/* Common case, no change */
+		return;
+	}
+
+	if (phy_link == 0) {
+		netif_carrier_off(ax_info->net);
+		/* Abort an in-progress poll of the PHY if necessary */
+		switch (ax_info->phy_state) {
+		case AX_PHY_STATE_POLLING_1:
+		case AX_PHY_STATE_POLLING_2:
+		case AX_PHY_STATE_POLLING_3:
+			ax_info->phy_state = AX_PHY_STATE_ABORT_POLL;
+			break;
+
+		case AX_PHY_STATE_POLLING_4:
+		case AX_PHY_STATE_SETTING_MAC:
+			ax_info->phy_state = AX_PHY_STATE_ABORTING;
+			break;
+
+		case AX_PHY_STATE_LINK:
+			ax_info->phy_state = AX_PHY_STATE_NO_LINK;
+			break;
+
+		default:
+			/* If we're already aborting, continue aborting */
+			break;
+		}
+	} else {
+		/* Note that we only fall into this case if previous phy_state was
+		   AX_PHY_STATE_NO_LINK. When the link is reported active while
+		   we're still polling, or when we're aborting, the logic above
+		   will just return, and we'll check again next int callback. */
+
+		ax_info->phy_state = AX_PHY_STATE_POLLING_1;
+		ax_info->phy_req.devreq.bRequestType = AX_REQ_WRITE;
+		ax_info->phy_req.devreq.bRequest = AX_CMD_SET_SW_MII;
+		ax_info->phy_req.devreq.wValue = cpu_to_le16(0);
+		ax_info->phy_req.devreq.wIndex = cpu_to_le16(0);
+		ax_info->phy_req.devreq.wLength = cpu_to_le16(0);
+		ax_info->phy_req.data_size = 0;
+		ax_info->phy_req.timeout = AX_TIMEOUT_CMD;
+		ax_info->phy_req.cmd_callback = ax_phy_cmd_callback;
+
+		ax_run_ctl_queue(ax_info, &ax_info->phy_req, 0);
+	}
+}
+
+static void ax_rx_callback(struct urb *urb)
+{
+	struct sk_buff *skb = (struct sk_buff *) urb->context;
+	struct net_device *net = skb->dev;
+	struct ax8817x_info *ax_info = (struct ax8817x_info *) net->priv;
+	int ret, len, refill;
+
+	switch (urb->status) {
+	case 0:
+		break;
+
+	default:
+		err("%s: URB status %d\n", __FUNCTION__, urb->status);
+		/* It's not clear that we can do much in this case, the rx pipe
+		   doesn't ever seem to stall, so if we got -ETIMEDOUT, that
+		   usually means the device was unplugged, and we just haven't
+		   noticed yet.
+		   Just fall through and free skb without resubmitting urb. */
+	case -ENOENT:		/* */
+	case -ECONNRESET:	/* Async unlink */
+	case -ESHUTDOWN:	/* Hardware gone */
+	case -EILSEQ:		/* Get this when you yank it out on UHCI */
+	case -ETIMEDOUT:	/* OHCI */
+	case -EPROTO:		/* EHCI */
+	case -EPIPE:		
+		dev_kfree_skb_any(skb);
+		urb->context = NULL;
+		return;
+	}
+
+	if (ax_info->drv_state == AX_DRV_STATE_INITIALIZING) {
+		/* Not really expecting this to ever happen, since we haven't yet
+		   enabled receive in the rx_ctl register, but ya never know... */
+		goto refill_same;
+	} else if (ax_info->drv_state == AX_DRV_STATE_EXITING) {
+		dev_kfree_skb_any(skb);
+		urb->context = NULL;
+		return;
+	}
+
+	len = urb->actual_length;
+	if (len == 0) {
+		/* this shouldn't happen... */
+		goto refill_same;
+	}
+
+	refill = ax_refill_rx_urb(ax_info, urb);
+
+	if (refill == 0
+	    || atomic_read(&ax_info->rx_refill_cnt) < n_rx_urbs) {
+		/* Send the receive buffer up the network stack */
+		skb_put(skb, len);
+		skb->protocol = eth_type_trans(skb, net);
+		net->last_rx = jiffies;
+		ax_info->stats.rx_packets++;
+		ax_info->stats.rx_bytes += len;
+
+		netif_rx(skb);
+
+		if (refill == 0) {
+			int i;
+
+			/* This is the common case. This URB got refilled OK, and
+			   no other URBs need to be refilled. */
+			if (atomic_read(&ax_info->rx_refill_cnt) == 0) {
+				return;
+			}
+
+			for (i = 0; i < n_rx_urbs; i++) {
+				struct urb *urb = ax_info->rx_urbs[i];
+
+				if (urb->context == NULL) {
+					if (ax_refill_rx_urb(ax_info, urb)
+					    == 0) {
+						atomic_dec(&ax_info->
+							   rx_refill_cnt);
+					} else {
+						break;
+					}
+				}
+			}
+		} else {
+			/* remember to refill this one later */
+			atomic_inc(&ax_info->rx_refill_cnt);
+		}
+
+		return;
+	} else {
+		ax_info->stats.rx_dropped++;
+		if (refill < 0) {
+			/* the error code was already printk'ed in ax_refill_rx_urb()
+			   so just note the consequences here: */
+			warn("Halting rx due to error\n");
+			return;
+		}
+
+		/* fall through to resubmit this URB with the existing skb
+		   will try to reallocate skb's on next rx callback */
+	}
+
+      refill_same:
+	usb_fill_bulk_urb(urb, ax_info->usb,
+			  usb_rcvbulkpipe(ax_info->usb, 3), skb->data,
+			  AX_RX_MAX, ax_rx_callback, skb);
+
+	ret = usb_submit_urb(urb);
+	if (ret < 0) {
+		err("Failed submit rx URB (%d)\n", ret);
+	}
+}
+
+static int ax8817x_open(struct net_device *net)
+{
+	struct ax8817x_info *ax_info = (struct ax8817x_info *) net->priv;
+	u8 buf[4];
+	int i, ret;
+
+	ret = ax_write_cmd(ax_info, AX_CMD_WRITE_RX_CTL, 0x80, 0, 0, buf);
+	if (ret < 0) {
+		return ret;
+	}
+
+	ret = 0;
+
+	ax_info->tx_urb = usb_alloc_urb(0);
+	if (ax_info->tx_urb == NULL)
+		ret = -ENOMEM;
+
+	atomic_set(&ax_info->rx_refill_cnt, 0);
+
+	for (i = 0; i < n_rx_urbs && ret == 0; i++) {
+		struct urb *urb = ax_info->rx_urbs[i];
+
+		if (urb == NULL) {
+			urb = ax_info->rx_urbs[i] = usb_alloc_urb(0);
+			if (urb == NULL) {
+				ret = -ENOMEM;
+				break;
+			}
+			if (n_rx_urbs > 1) {
+				urb->transfer_flags |= USB_QUEUE_BULK;
+			}
+		}
+		ret = ax_refill_rx_urb(ax_info, urb);
+		if (ret == 1) {
+			atomic_inc(&ax_info->rx_refill_cnt);
+			ret = 0;
+		}
+	}
+
+	/* XXX: should handle the case where we couldn't allocate any skb's
+	   better. They get allocated with GFP_ATOMIC, so they may all fail... */
+	if (ret == 0 && atomic_read(&ax_info->rx_refill_cnt) < n_rx_urbs) {
+		netif_start_queue(net);
+	} else {
+		/* Error: clean up anything we allocated and bail. */
+		usb_free_urb(ax_info->tx_urb);
+
+		for (i = 0; i < n_rx_urbs; i++) {
+			struct urb *urb = ax_info->rx_urbs[i];
+
+			if (urb != NULL) {
+				/* skb gets freed in the URB callback */
+				usb_unlink_urb(urb);
+				usb_free_urb(urb);
+			}
+		}
+
+		err("%s: Failed start rx queue (%d)\n", __FUNCTION__, ret);
+	}
+	return ret;
+}
+
+static int ax8817x_stop(struct net_device *net)
+{
+	struct ax8817x_info *ax_info = (struct ax8817x_info *) net->priv;
+	u8 buf[4];
+	int i, ret;
+
+	netif_stop_queue(net);
+
+	ret = ax_write_cmd(ax_info, AX_CMD_WRITE_RX_CTL, 0x80, 0, 0, buf);
+	if (ret < 0 && ax_info->drv_state != AX_DRV_STATE_EXITING) {
+		err("%s: Failed cmd (%d)\n", __FUNCTION__, ret);
+	}
+
+	if (ax_info->tx_urb != NULL) {
+		usb_unlink_urb(ax_info->tx_urb);
+		usb_free_urb(ax_info->tx_urb);
+		ax_info->tx_urb = NULL;
+	}
+
+	for (i = 0; i < n_rx_urbs; i++) {
+		struct urb *urb = ax_info->rx_urbs[i];
+		if (urb != NULL) {
+			/* skb gets freed in the URB callback */
+			usb_unlink_urb(urb);
+			usb_free_urb(urb);
+			ax_info->rx_urbs[i] = NULL;
+		}
+	}
+
+	return 0;
+}
+
+static void write_bulk_callback(struct urb *urb)
+{
+	struct ax8817x_info *ax_info = urb->context;
+
+	if (!ax_info || (ax_info->drv_state == AX_DRV_STATE_EXITING))
+		return;
+
+	if (!netif_device_present(ax_info->net))
+		return;
+
+	if (urb->status)
+		info("%s: TX status %d", ax_info->net->name, urb->status);
+
+	ax_info->net->trans_start = jiffies;
+	netif_wake_queue(ax_info->net);
+}
+
+static int ax8817x_start_xmit(struct sk_buff *skb, struct net_device *net)
+{
+	struct ax8817x_info *ax_info = net->priv;
+	int res;
+
+	netif_stop_queue(net);
+
+	ax_info->tx_urb->transfer_flags |= USB_ZERO_PACKET;
+	usb_fill_bulk_urb(ax_info->tx_urb, ax_info->usb,
+			  usb_sndbulkpipe(ax_info->usb, 2),
+			  skb->data, skb->len, write_bulk_callback,
+			  ax_info);
+	if ((res = usb_submit_urb(ax_info->tx_urb))) {
+		warn("Failed tx_urb %d", res);
+		ax_info->stats.tx_errors++;
+		netif_start_queue(net);
+	} else {
+		ax_info->stats.tx_packets++;
+		ax_info->stats.tx_bytes += skb->len;
+		net->trans_start = jiffies;
+	}
+	dev_kfree_skb(skb);
+
+	return 0;
+}
+
+static void ax8817x_tx_timeout(struct net_device *net)
+{
+	struct ax8817x_info *ax_info = net->priv;
+
+	if (!ax_info)
+		return;
+
+	warn("%s: Tx timed out.", net->name);
+	ax_info->tx_urb->transfer_flags |= USB_ASYNC_UNLINK;
+	usb_unlink_urb(ax_info->tx_urb);
+	ax_info->stats.tx_errors++;
+}
+
+static struct net_device_stats *ax8817x_stats(struct net_device *net)
+{
+	struct ax8817x_info *ax_info = (struct ax8817x_info *) net->priv;
+
+	return &ax_info->stats;
+}
+
+static void ax8817x_set_multicast(struct net_device *net)
+{
+	struct ax8817x_info *ax_info = (struct ax8817x_info *) net->priv;
+	u8 rx_ctl = 0x8c;
+
+	if (net->flags & IFF_PROMISC) {
+		rx_ctl |= 0x01;
+	} else if (net->flags & IFF_ALLMULTI
+		   || net->mc_count > AX_MAX_MCAST) {
+		rx_ctl |= 0x02;
+	} else if (net->mc_count == 0) {
+		/* just broadcast and directed */
+	} else {
+		struct dev_mc_list *mc_list = net->mc_list;
+		u8 *multi_filter;
+		u32 crc_bits;
+		int i;
+
+		multi_filter = kmalloc(8, GFP_ATOMIC);
+		if (multi_filter == NULL) {
+			/* Oops, couldn't allocate a DMA buffer for setting the multicast
+			   filter. Try all multi mode, although the ax_write_cmd_async
+			   will almost certainly fail, too... (but it will printk). */
+			rx_ctl |= 0x02;
+		} else {
+			memset(multi_filter, 0, 8);
+
+			/* Build the multicast hash filter. */
+			for (i = 0; i < net->mc_count; i++) {
+				crc_bits =
+				    ether_crc(ETH_ALEN,
+					      mc_list->dmi_addr) >> 26;
+				multi_filter[crc_bits >> 3] |=
+				    1 << (crc_bits & 7);
+				mc_list = mc_list->next;
+			}
+
+			ax_write_cmd_async(ax_info,
+					   AX_CMD_WRITE_MULTI_FILTER, 0, 0,
+					   8, multi_filter);
+
+			rx_ctl |= 0x10;
+		}
+	}
+
+	ax_write_cmd_async(ax_info, AX_CMD_WRITE_RX_CTL, rx_ctl, 0, 0,
+			   NULL);
+}
+
+static int ax8817x_ethtool_ioctl(struct net_device *net, void *uaddr)
+{
+	struct ax8817x_info *ax_info;
+	int cmd;
+
+	ax_info = net->priv;
+	if (get_user(cmd, (int *) uaddr))
+		return -EFAULT;
+
+	switch (cmd) {
+	case ETHTOOL_GDRVINFO:{
+			struct ethtool_drvinfo info = { ETHTOOL_GDRVINFO };
+
+			strncpy(info.driver, driver_name,
+				ETHTOOL_BUSINFO_LEN);
+			strncpy(info.version, DRIVER_VERSION,
+				ETHTOOL_BUSINFO_LEN);
+			usb_make_path(ax_info->usb, info.bus_info,
+				      sizeof info.bus_info);
+			if (copy_to_user(uaddr, &info, sizeof(info)))
+				return -EFAULT;
+			return 0;
+		}
+	case ETHTOOL_GSET:{
+			struct ethtool_cmd ecmd;
+
+			if (copy_from_user(&ecmd, uaddr, sizeof(ecmd)))
+				return -EFAULT;
+			ecmd.supported = (SUPPORTED_10baseT_Half |
+					  SUPPORTED_10baseT_Full |
+					  SUPPORTED_100baseT_Half |
+					  SUPPORTED_100baseT_Full |
+					  SUPPORTED_Autoneg |
+					  SUPPORTED_TP | SUPPORTED_MII);
+			ecmd.port = PORT_TP;
+			ecmd.transceiver = XCVR_INTERNAL;
+			ecmd.phy_address = 0;	/* FIXME */
+
+			if (copy_to_user(uaddr, &ecmd, sizeof(ecmd)))
+				return -EFAULT;
+			return 0;
+		}
+	case ETHTOOL_SSET:
+		return -ENOTSUPP;
+	case ETHTOOL_GLINK:{
+			struct ethtool_value edata = { ETHTOOL_GLINK };
+
+			edata.data = netif_carrier_ok(net);
+			if (copy_to_user(uaddr, &edata, sizeof(edata)))
+				return -EFAULT;
+			return 0;
+		}
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+static int ax8817x_mii_ioctl(struct net_device *net, struct ifreq *ifr,
+			     int cmd)
+{
+	struct ax8817x_info *ax_info;
+	struct mii_ioctl_data *data_ptr =
+	    (struct mii_ioctl_data *) &(ifr->ifr_data);
+
+	ax_info = net->priv;
+
+	switch (cmd) {
+	case SIOCGMIIPHY:
+		data_ptr->phy_id = ax_info->phy_id;
+		break;
+	case SIOCGMIIREG:
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+
+		ax_read_cmd(ax_info, AX_CMD_READ_MII_REG, 0,
+			    data_ptr->reg_num & 0x1f, 2,
+			    &(data_ptr->val_out));
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+	return 0;
+}
+
+static int ax8817x_ioctl(struct net_device *net, struct ifreq *ifr,
+			 int cmd)
+{
+	struct ax8817x_info *ax_info;
+	int res;
+
+	ax_info = net->priv;
+	res = 0;
+
+	switch (cmd) {
+	case SIOCETHTOOL:
+		res = ax8817x_ethtool_ioctl(net, ifr->ifr_data);
+		break;
+	case SIOCGMIIPHY:	/* Get address of PHY in use */
+	case SIOCGMIIREG:	/* Read from MII PHY register */
+	case SIOCSMIIREG:	/* Write to MII PHY register */
+		return ax8817x_mii_ioctl(net, ifr, cmd);
+	default:
+		res = -EOPNOTSUPP;
+	}
+
+	return res;
+}
+
+static int ax8817x_net_init(struct net_device *net)
+{
+	struct ax8817x_info *ax_info = (struct ax8817x_info *) net->priv;
+	u8 buf[6];
+	u16 *buf16 = (u16 *) buf;
+	int ret;
+
+	spin_lock_init(&ax_info->tx_lock);
+
+	ret = ax_write_cmd(ax_info, AX_CMD_WRITE_RX_CTL, 0x80, 0, 0, buf);
+	if (ret < 0) {
+		return ret;
+	}
+
+	memset(buf, 0, 6);
+
+	/* Get the MAC address */
+	ret = ax_read_cmd(ax_info, AX_CMD_READ_NODE_ID, 0, 0, 6, buf);
+	if (ret < 0) {
+		return ret;
+	}
+
+	memcpy(net->dev_addr, buf, 6);
+
+	/* Get the PHY id */
+	ret = ax_read_cmd(ax_info, AX_CMD_READ_PHY_ID, 0, 0, 2, buf);
+	if (ret < 0) {
+		return ret;
+	} else if (ret < 2) {
+		/* this should always return 2 bytes */
+		return -EIO;
+	}
+
+	/* Reset the PHY, and drop it into auto-negotiation mode */
+	ax_info->phy_id = buf[1];
+	ax_info->phy_state = AX_PHY_STATE_INITIALIZING;
+
+	ret = ax_write_cmd(ax_info, AX_CMD_SET_SW_MII, 0, 0, 0, &buf);
+	if (ret < 0) {
+		return ret;
+	}
+
+	*buf16 = cpu_to_le16(BMCR_RESET);
+	ret = ax_write_cmd(ax_info, AX_CMD_WRITE_MII_REG,
+			   ax_info->phy_id, MII_BMCR, 2, buf16);
+	if (ret < 0) {
+		return ret;
+	}
+
+	/* Advertise that we can do full-duplex pause */
+	*buf16 = cpu_to_le16(ADVERTISE_ALL | ADVERTISE_CSMA | 0x0400);
+	ret = ax_write_cmd(ax_info, AX_CMD_WRITE_MII_REG,
+			   ax_info->phy_id, MII_ADVERTISE, 2, buf16);
+	if (ret < 0) {
+		return ret;
+	}
+
+	*buf16 = cpu_to_le16(BMCR_ANENABLE | BMCR_ANRESTART);
+	ret = ax_write_cmd(ax_info, AX_CMD_WRITE_MII_REG,
+			   ax_info->phy_id, MII_BMCR, 2, buf16);
+	if (ret < 0) {
+		return ret;
+	}
+
+	ret = ax_write_cmd(ax_info, AX_CMD_SET_HW_MII, 0, 0, 0, &buf);
+	if (ret < 0) {
+		return ret;
+	}
+
+	net->open = ax8817x_open;
+	net->stop = ax8817x_stop;
+	net->hard_start_xmit = ax8817x_start_xmit;
+	net->tx_timeout = ax8817x_tx_timeout;
+	net->watchdog_timeo = AX_TIMEOUT_TX;
+	net->get_stats = ax8817x_stats;
+	net->do_ioctl = ax8817x_ioctl;
+	net->set_multicast_list = ax8817x_set_multicast;
+
+	return 0;
+}
+
+static void *ax8817x_bind(struct usb_device *usb, unsigned intf,
+			  const struct usb_device_id *id)
+{
+	struct ax8817x_info *ax_info;
+	struct net_device *net;
+	int i, ret;
+	unsigned long gpio_bits = id->driver_info;
+	u8 buf[2];
+
+	/* Allocate the URB lists along with the device info struct */
+	ax_info = kmalloc(sizeof(struct ax8817x_info) +
+			  n_rx_urbs * sizeof(struct urb *), GFP_KERNEL);
+	if (ax_info == NULL) {
+		err("%s: Failed ax alloc\n", __FUNCTION__);
+		goto exit_err;
+	}
+
+	memset(ax_info, 0, sizeof(struct ax8817x_info) +
+	       n_rx_urbs * sizeof(struct urb *));
+
+	ax_info->drv_state = AX_DRV_STATE_INITIALIZING;
+	ax_info->rx_urbs = (struct urb **) (ax_info + 1);
+	ax_info->usb = usb;
+
+	/* Set up the control URB queue */
+
+	INIT_LIST_HEAD(&ax_info->ctl_queue);
+	spin_lock_init(&ax_info->ctl_lock);
+	ax_info->ctl_urb = usb_alloc_urb(0);
+	if (ax_info->ctl_urb == NULL) {
+		goto exit_err_free_ax;
+	}
+
+	/* Toggle the GPIOs in a manufacturer/model specific way */
+
+	for (i = 2; i >= 0; i--) {
+		ret = ax_write_cmd(ax_info, AX_CMD_WRITE_GPIOS,
+				   (gpio_bits >> (i * 8)) & 0xff, 0, 0,
+				   buf);
+		if (ret < 0) {
+			goto exit_err_free_ax;
+		}
+		wait_ms(5);
+	}
+
+	/* Set up the net device */
+
+	net = alloc_etherdev(0);
+	if (net == NULL) {
+		err("%s: Failed net alloc\n", __FUNCTION__);
+		goto exit_err_free_ax;
+	}
+
+	ax_info->net = net;
+
+	SET_MODULE_OWNER(net);
+	net->init = ax8817x_net_init;
+	net->priv = ax_info;
+
+	ret = register_netdev(net);
+	if (ret < 0) {
+		err("%s: Failed net init (%d)\n", __FUNCTION__, ret);
+		goto exit_err_free_net;
+	}
+
+	/* Set up the interrupt URB, and start PHY state monitoring */
+
+	ax_info->int_urb = usb_alloc_urb(0);
+	if (ax_info->int_urb == NULL) {
+		goto exit_err_unregister_net;
+	}
+	ax_info->int_buf = kmalloc(8, GFP_KERNEL);
+	if (ax_info->int_buf == NULL) {
+		goto exit_err_free_int_urb;
+	}
+	ax_info->phy_req.data = kmalloc(2, GFP_KERNEL);
+	if (ax_info->phy_req.data == NULL) {
+		goto exit_err_free_int_buf;
+	}
+
+	usb_fill_int_urb(ax_info->int_urb, usb, usb_rcvintpipe(usb, 1),
+			 ax_info->int_buf, 8, ax_int_callback, ax_info,
+			 100);
+
+	ret = usb_submit_urb(ax_info->int_urb);
+	if (ret < 0) {
+		err("%s: Failed int URB submit (%d)\n", __FUNCTION__, ret);
+		goto exit_err_free_phy_buf;
+	}
+
+	ax_info->drv_state = AX_DRV_STATE_RUNNING;
+	return ax_info;
+
+      exit_err_free_phy_buf:
+	kfree(ax_info->phy_req.data);
+
+      exit_err_free_int_buf:
+	kfree(ax_info->int_buf);
+
+      exit_err_free_int_urb:
+	usb_free_urb(ax_info->int_urb);
+
+      exit_err_unregister_net:
+	ax_info->drv_state = AX_DRV_STATE_EXITING;
+	unregister_netdev(net);
+
+      exit_err_free_net:
+	kfree(net);
+
+      exit_err_free_ax:
+	if (ax_info->ctl_urb != NULL) {
+		/* no need to unlink, since there should not be any ctl URBs
+		   pending at this point */
+		usb_free_urb(ax_info->ctl_urb);
+	}
+
+	kfree(ax_info);
+
+      exit_err:
+	err("%s: Failed to initialize\n", __FUNCTION__);
+	return NULL;
+}
+
+static void ax8817x_disconnect(struct usb_device *usb, void *p)
+{
+	struct ax8817x_info *ax_info = (struct ax8817x_info *) p;
+
+	ax_info->drv_state = AX_DRV_STATE_EXITING;
+
+	if (ax_info->int_urb != NULL) {
+		usb_unlink_urb(ax_info->int_urb);
+		usb_free_urb(ax_info->int_urb);
+		kfree(ax_info->int_buf);
+	}
+
+	unregister_netdev(ax_info->net);
+
+	/* XXX: hmmm... need to go through and clear out the ctl queue, too... */
+	if (ax_info->ctl_urb != NULL) {
+		usb_unlink_urb(ax_info->ctl_urb);
+		usb_free_urb(ax_info->ctl_urb);
+	}
+
+	kfree(ax_info);
+}
+
+
+static struct usb_driver ax8817x_driver = {
+	.owner = THIS_MODULE,
+	.name = driver_name,
+	.probe = ax8817x_bind,
+	.disconnect = ax8817x_disconnect,
+	.id_table = ax8817x_id_table,
+};
+
+static int __init ax8817x_init(void)
+{
+	int ret;
+
+	if (n_rx_urbs < 1) 
+		n_rx_urbs = AX_RX_URBS_DEFAULT;
+
+	ret = usb_register(&ax8817x_driver);
+	if (ret < 0) {
+		err("%s: Failed to register\n", __FUNCTION__);
+	} else {
+		info(DRIVER_DESC " " DRIVER_VERSION);
+	}
+
+	return ret;
+}
+
+static void __exit ax8817x_exit(void)
+{
+	usb_deregister(&ax8817x_driver);
+}
+
+module_init(ax8817x_init);
+module_exit(ax8817x_exit);
+
+EXPORT_NO_SYMBOLS;

FUNET's LINUX-ADM group, linux-adm@nic.funet.fi
TCL-scripts by Sam Shen (who was at: slshen@lbl.gov)