Move USB network drivers to drivers/net/usb.
[powerpc.git] / drivers / net / usb / usbnet.c
diff --git a/drivers/net/usb/usbnet.c b/drivers/net/usb/usbnet.c
new file mode 100644 (file)
index 0000000..f9cd42d
--- /dev/null
@@ -0,0 +1,1304 @@
+/*
+ * USB Network driver infrastructure
+ * Copyright (C) 2000-2005 by David Brownell
+ * Copyright (C) 2003-2005 David Hollis <dhollis@davehollis.com>
+ *
+ * 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, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/*
+ * This is a generic "USB networking" framework that works with several
+ * kinds of full and high speed networking devices:  host-to-host cables,
+ * smart usb peripherals, and actual Ethernet adapters.
+ *
+ * These devices usually differ in terms of control protocols (if they
+ * even have one!) and sometimes they define new framing to wrap or batch
+ * Ethernet packets.  Otherwise, they talk to USB pretty much the same,
+ * so interface (un)binding, endpoint I/O queues, fault handling, and other
+ * issues can usefully be addressed by this framework.
+ */
+
+// #define     DEBUG                   // error path messages, extra info
+// #define     VERBOSE                 // more; success messages
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/ethtool.h>
+#include <linux/workqueue.h>
+#include <linux/mii.h>
+#include <linux/usb.h>
+
+#include "usbnet.h"
+
+#define DRIVER_VERSION         "22-Aug-2005"
+
+
+/*-------------------------------------------------------------------------*/
+
+/*
+ * Nineteen USB 1.1 max size bulk transactions per frame (ms), max.
+ * Several dozen bytes of IPv4 data can fit in two such transactions.
+ * One maximum size Ethernet packet takes twenty four of them.
+ * For high speed, each frame comfortably fits almost 36 max size
+ * Ethernet packets (so queues should be bigger).
+ *
+ * REVISIT qlens should be members of 'struct usbnet'; the goal is to
+ * let the USB host controller be busy for 5msec or more before an irq
+ * is required, under load.  Jumbograms change the equation.
+ */
+#define RX_MAX_QUEUE_MEMORY (60 * 1518)
+#define        RX_QLEN(dev) (((dev)->udev->speed == USB_SPEED_HIGH) ? \
+                       (RX_MAX_QUEUE_MEMORY/(dev)->rx_urb_size) : 4)
+#define        TX_QLEN(dev) (((dev)->udev->speed == USB_SPEED_HIGH) ? \
+                       (RX_MAX_QUEUE_MEMORY/(dev)->hard_mtu) : 4)
+
+// reawaken network queue this soon after stopping; else watchdog barks
+#define TX_TIMEOUT_JIFFIES     (5*HZ)
+
+// throttle rx/tx briefly after some faults, so khubd might disconnect()
+// us (it polls at HZ/4 usually) before we report too many false errors.
+#define THROTTLE_JIFFIES       (HZ/8)
+
+// between wakeups
+#define UNLINK_TIMEOUT_MS      3
+
+/*-------------------------------------------------------------------------*/
+
+// randomly generated ethernet address
+static u8      node_id [ETH_ALEN];
+
+static const char driver_name [] = "usbnet";
+
+/* use ethtool to change the level for any given device */
+static int msg_level = -1;
+module_param (msg_level, int, 0);
+MODULE_PARM_DESC (msg_level, "Override default message level");
+
+/*-------------------------------------------------------------------------*/
+
+/* handles CDC Ethernet and many other network "bulk data" interfaces */
+int usbnet_get_endpoints(struct usbnet *dev, struct usb_interface *intf)
+{
+       int                             tmp;
+       struct usb_host_interface       *alt = NULL;
+       struct usb_host_endpoint        *in = NULL, *out = NULL;
+       struct usb_host_endpoint        *status = NULL;
+
+       for (tmp = 0; tmp < intf->num_altsetting; tmp++) {
+               unsigned        ep;
+
+               in = out = status = NULL;
+               alt = intf->altsetting + tmp;
+
+               /* take the first altsetting with in-bulk + out-bulk;
+                * remember any status endpoint, just in case;
+                * ignore other endpoints and altsetttings.
+                */
+               for (ep = 0; ep < alt->desc.bNumEndpoints; ep++) {
+                       struct usb_host_endpoint        *e;
+                       int                             intr = 0;
+
+                       e = alt->endpoint + ep;
+                       switch (e->desc.bmAttributes) {
+                       case USB_ENDPOINT_XFER_INT:
+                               if (!usb_endpoint_dir_in(&e->desc))
+                                       continue;
+                               intr = 1;
+                               /* FALLTHROUGH */
+                       case USB_ENDPOINT_XFER_BULK:
+                               break;
+                       default:
+                               continue;
+                       }
+                       if (usb_endpoint_dir_in(&e->desc)) {
+                               if (!intr && !in)
+                                       in = e;
+                               else if (intr && !status)
+                                       status = e;
+                       } else {
+                               if (!out)
+                                       out = e;
+                       }
+               }
+               if (in && out)
+                       break;
+       }
+       if (!alt || !in || !out)
+               return -EINVAL;
+
+       if (alt->desc.bAlternateSetting != 0
+                       || !(dev->driver_info->flags & FLAG_NO_SETINT)) {
+               tmp = usb_set_interface (dev->udev, alt->desc.bInterfaceNumber,
+                               alt->desc.bAlternateSetting);
+               if (tmp < 0)
+                       return tmp;
+       }
+
+       dev->in = usb_rcvbulkpipe (dev->udev,
+                       in->desc.bEndpointAddress & USB_ENDPOINT_NUMBER_MASK);
+       dev->out = usb_sndbulkpipe (dev->udev,
+                       out->desc.bEndpointAddress & USB_ENDPOINT_NUMBER_MASK);
+       dev->status = status;
+       return 0;
+}
+EXPORT_SYMBOL_GPL(usbnet_get_endpoints);
+
+static void intr_complete (struct urb *urb);
+
+static int init_status (struct usbnet *dev, struct usb_interface *intf)
+{
+       char            *buf = NULL;
+       unsigned        pipe = 0;
+       unsigned        maxp;
+       unsigned        period;
+
+       if (!dev->driver_info->status)
+               return 0;
+
+       pipe = usb_rcvintpipe (dev->udev,
+                       dev->status->desc.bEndpointAddress
+                               & USB_ENDPOINT_NUMBER_MASK);
+       maxp = usb_maxpacket (dev->udev, pipe, 0);
+
+       /* avoid 1 msec chatter:  min 8 msec poll rate */
+       period = max ((int) dev->status->desc.bInterval,
+               (dev->udev->speed == USB_SPEED_HIGH) ? 7 : 3);
+
+       buf = kmalloc (maxp, GFP_KERNEL);
+       if (buf) {
+               dev->interrupt = usb_alloc_urb (0, GFP_KERNEL);
+               if (!dev->interrupt) {
+                       kfree (buf);
+                       return -ENOMEM;
+               } else {
+                       usb_fill_int_urb(dev->interrupt, dev->udev, pipe,
+                               buf, maxp, intr_complete, dev, period);
+                       dev_dbg(&intf->dev,
+                               "status ep%din, %d bytes period %d\n",
+                               usb_pipeendpoint(pipe), maxp, period);
+               }
+       }
+       return  0;
+}
+
+/* Passes this packet up the stack, updating its accounting.
+ * Some link protocols batch packets, so their rx_fixup paths
+ * can return clones as well as just modify the original skb.
+ */
+void usbnet_skb_return (struct usbnet *dev, struct sk_buff *skb)
+{
+       int     status;
+
+       skb->protocol = eth_type_trans (skb, dev->net);
+       dev->stats.rx_packets++;
+       dev->stats.rx_bytes += skb->len;
+
+       if (netif_msg_rx_status (dev))
+               devdbg (dev, "< rx, len %zu, type 0x%x",
+                       skb->len + sizeof (struct ethhdr), skb->protocol);
+       memset (skb->cb, 0, sizeof (struct skb_data));
+       status = netif_rx (skb);
+       if (status != NET_RX_SUCCESS && netif_msg_rx_err (dev))
+               devdbg (dev, "netif_rx status %d", status);
+}
+EXPORT_SYMBOL_GPL(usbnet_skb_return);
+
+\f
+/*-------------------------------------------------------------------------
+ *
+ * Network Device Driver (peer link to "Host Device", from USB host)
+ *
+ *-------------------------------------------------------------------------*/
+
+static int usbnet_change_mtu (struct net_device *net, int new_mtu)
+{
+       struct usbnet   *dev = netdev_priv(net);
+       int             ll_mtu = new_mtu + net->hard_header_len;
+       int             old_hard_mtu = dev->hard_mtu;
+       int             old_rx_urb_size = dev->rx_urb_size;
+
+       if (new_mtu <= 0)
+               return -EINVAL;
+       // no second zero-length packet read wanted after mtu-sized packets
+       if ((ll_mtu % dev->maxpacket) == 0)
+               return -EDOM;
+       net->mtu = new_mtu;
+
+       dev->hard_mtu = net->mtu + net->hard_header_len;
+       if (dev->rx_urb_size == old_hard_mtu) {
+               dev->rx_urb_size = dev->hard_mtu;
+               if (dev->rx_urb_size > old_rx_urb_size)
+                       usbnet_unlink_rx_urbs(dev);
+       }
+
+       return 0;
+}
+
+/*-------------------------------------------------------------------------*/
+
+static struct net_device_stats *usbnet_get_stats (struct net_device *net)
+{
+       struct usbnet   *dev = netdev_priv(net);
+       return &dev->stats;
+}
+
+/*-------------------------------------------------------------------------*/
+
+/* some LK 2.4 HCDs oopsed if we freed or resubmitted urbs from
+ * completion callbacks.  2.5 should have fixed those bugs...
+ */
+
+static void defer_bh(struct usbnet *dev, struct sk_buff *skb, struct sk_buff_head *list)
+{
+       unsigned long           flags;
+
+       spin_lock_irqsave(&list->lock, flags);
+       __skb_unlink(skb, list);
+       spin_unlock(&list->lock);
+       spin_lock(&dev->done.lock);
+       __skb_queue_tail(&dev->done, skb);
+       if (dev->done.qlen == 1)
+               tasklet_schedule(&dev->bh);
+       spin_unlock_irqrestore(&dev->done.lock, flags);
+}
+
+/* some work can't be done in tasklets, so we use keventd
+ *
+ * NOTE:  annoying asymmetry:  if it's active, schedule_work() fails,
+ * but tasklet_schedule() doesn't.  hope the failure is rare.
+ */
+void usbnet_defer_kevent (struct usbnet *dev, int work)
+{
+       set_bit (work, &dev->flags);
+       if (!schedule_work (&dev->kevent))
+               deverr (dev, "kevent %d may have been dropped", work);
+       else
+               devdbg (dev, "kevent %d scheduled", work);
+}
+EXPORT_SYMBOL_GPL(usbnet_defer_kevent);
+
+/*-------------------------------------------------------------------------*/
+
+static void rx_complete (struct urb *urb);
+
+static void rx_submit (struct usbnet *dev, struct urb *urb, gfp_t flags)
+{
+       struct sk_buff          *skb;
+       struct skb_data         *entry;
+       int                     retval = 0;
+       unsigned long           lockflags;
+       size_t                  size = dev->rx_urb_size;
+
+       if ((skb = alloc_skb (size + NET_IP_ALIGN, flags)) == NULL) {
+               if (netif_msg_rx_err (dev))
+                       devdbg (dev, "no rx skb");
+               usbnet_defer_kevent (dev, EVENT_RX_MEMORY);
+               usb_free_urb (urb);
+               return;
+       }
+       skb_reserve (skb, NET_IP_ALIGN);
+
+       entry = (struct skb_data *) skb->cb;
+       entry->urb = urb;
+       entry->dev = dev;
+       entry->state = rx_start;
+       entry->length = 0;
+
+       usb_fill_bulk_urb (urb, dev->udev, dev->in,
+               skb->data, size, rx_complete, skb);
+
+       spin_lock_irqsave (&dev->rxq.lock, lockflags);
+
+       if (netif_running (dev->net)
+                       && netif_device_present (dev->net)
+                       && !test_bit (EVENT_RX_HALT, &dev->flags)) {
+               switch (retval = usb_submit_urb (urb, GFP_ATOMIC)){
+               case -EPIPE:
+                       usbnet_defer_kevent (dev, EVENT_RX_HALT);
+                       break;
+               case -ENOMEM:
+                       usbnet_defer_kevent (dev, EVENT_RX_MEMORY);
+                       break;
+               case -ENODEV:
+                       if (netif_msg_ifdown (dev))
+                               devdbg (dev, "device gone");
+                       netif_device_detach (dev->net);
+                       break;
+               default:
+                       if (netif_msg_rx_err (dev))
+                               devdbg (dev, "rx submit, %d", retval);
+                       tasklet_schedule (&dev->bh);
+                       break;
+               case 0:
+                       __skb_queue_tail (&dev->rxq, skb);
+               }
+       } else {
+               if (netif_msg_ifdown (dev))
+                       devdbg (dev, "rx: stopped");
+               retval = -ENOLINK;
+       }
+       spin_unlock_irqrestore (&dev->rxq.lock, lockflags);
+       if (retval) {
+               dev_kfree_skb_any (skb);
+               usb_free_urb (urb);
+       }
+}
+
+
+/*-------------------------------------------------------------------------*/
+
+static inline void rx_process (struct usbnet *dev, struct sk_buff *skb)
+{
+       if (dev->driver_info->rx_fixup
+                       && !dev->driver_info->rx_fixup (dev, skb))
+               goto error;
+       // else network stack removes extra byte if we forced a short packet
+
+       if (skb->len)
+               usbnet_skb_return (dev, skb);
+       else {
+               if (netif_msg_rx_err (dev))
+                       devdbg (dev, "drop");
+error:
+               dev->stats.rx_errors++;
+               skb_queue_tail (&dev->done, skb);
+       }
+}
+
+/*-------------------------------------------------------------------------*/
+
+static void rx_complete (struct urb *urb)
+{
+       struct sk_buff          *skb = (struct sk_buff *) urb->context;
+       struct skb_data         *entry = (struct skb_data *) skb->cb;
+       struct usbnet           *dev = entry->dev;
+       int                     urb_status = urb->status;
+
+       skb_put (skb, urb->actual_length);
+       entry->state = rx_done;
+       entry->urb = NULL;
+
+       switch (urb_status) {
+           // success
+           case 0:
+               if (skb->len < dev->net->hard_header_len) {
+                       entry->state = rx_cleanup;
+                       dev->stats.rx_errors++;
+                       dev->stats.rx_length_errors++;
+                       if (netif_msg_rx_err (dev))
+                               devdbg (dev, "rx length %d", skb->len);
+               }
+               break;
+
+           // stalls need manual reset. this is rare ... except that
+           // when going through USB 2.0 TTs, unplug appears this way.
+           // we avoid the highspeed version of the ETIMEOUT/EILSEQ
+           // storm, recovering as needed.
+           case -EPIPE:
+               dev->stats.rx_errors++;
+               usbnet_defer_kevent (dev, EVENT_RX_HALT);
+               // FALLTHROUGH
+
+           // software-driven interface shutdown
+           case -ECONNRESET:           // async unlink
+           case -ESHUTDOWN:            // hardware gone
+               if (netif_msg_ifdown (dev))
+                       devdbg (dev, "rx shutdown, code %d", urb_status);
+               goto block;
+
+           // we get controller i/o faults during khubd disconnect() delays.
+           // throttle down resubmits, to avoid log floods; just temporarily,
+           // so we still recover when the fault isn't a khubd delay.
+           case -EPROTO:
+           case -ETIME:
+           case -EILSEQ:
+               dev->stats.rx_errors++;
+               if (!timer_pending (&dev->delay)) {
+                       mod_timer (&dev->delay, jiffies + THROTTLE_JIFFIES);
+                       if (netif_msg_link (dev))
+                               devdbg (dev, "rx throttle %d", urb_status);
+               }
+block:
+               entry->state = rx_cleanup;
+               entry->urb = urb;
+               urb = NULL;
+               break;
+
+           // data overrun ... flush fifo?
+           case -EOVERFLOW:
+               dev->stats.rx_over_errors++;
+               // FALLTHROUGH
+
+           default:
+               entry->state = rx_cleanup;
+               dev->stats.rx_errors++;
+               if (netif_msg_rx_err (dev))
+                       devdbg (dev, "rx status %d", urb_status);
+               break;
+       }
+
+       defer_bh(dev, skb, &dev->rxq);
+
+       if (urb) {
+               if (netif_running (dev->net)
+                               && !test_bit (EVENT_RX_HALT, &dev->flags)) {
+                       rx_submit (dev, urb, GFP_ATOMIC);
+                       return;
+               }
+               usb_free_urb (urb);
+       }
+       if (netif_msg_rx_err (dev))
+               devdbg (dev, "no read resubmitted");
+}
+
+static void intr_complete (struct urb *urb)
+{
+       struct usbnet   *dev = urb->context;
+       int             status = urb->status;
+
+       switch (status) {
+           /* success */
+           case 0:
+               dev->driver_info->status(dev, urb);
+               break;
+
+           /* software-driven interface shutdown */
+           case -ENOENT:               // urb killed
+           case -ESHUTDOWN:            // hardware gone
+               if (netif_msg_ifdown (dev))
+                       devdbg (dev, "intr shutdown, code %d", status);
+               return;
+
+           /* NOTE:  not throttling like RX/TX, since this endpoint
+            * already polls infrequently
+            */
+           default:
+               devdbg (dev, "intr status %d", status);
+               break;
+       }
+
+       if (!netif_running (dev->net))
+               return;
+
+       memset(urb->transfer_buffer, 0, urb->transfer_buffer_length);
+       status = usb_submit_urb (urb, GFP_ATOMIC);
+       if (status != 0 && netif_msg_timer (dev))
+               deverr(dev, "intr resubmit --> %d", status);
+}
+
+/*-------------------------------------------------------------------------*/
+
+// unlink pending rx/tx; completion handlers do all other cleanup
+
+static int unlink_urbs (struct usbnet *dev, struct sk_buff_head *q)
+{
+       unsigned long           flags;
+       struct sk_buff          *skb, *skbnext;
+       int                     count = 0;
+
+       spin_lock_irqsave (&q->lock, flags);
+       for (skb = q->next; skb != (struct sk_buff *) q; skb = skbnext) {
+               struct skb_data         *entry;
+               struct urb              *urb;
+               int                     retval;
+
+               entry = (struct skb_data *) skb->cb;
+               urb = entry->urb;
+               skbnext = skb->next;
+
+               // during some PM-driven resume scenarios,
+               // these (async) unlinks complete immediately
+               retval = usb_unlink_urb (urb);
+               if (retval != -EINPROGRESS && retval != 0)
+                       devdbg (dev, "unlink urb err, %d", retval);
+               else
+                       count++;
+       }
+       spin_unlock_irqrestore (&q->lock, flags);
+       return count;
+}
+
+// Flush all pending rx urbs
+// minidrivers may need to do this when the MTU changes
+
+void usbnet_unlink_rx_urbs(struct usbnet *dev)
+{
+       if (netif_running(dev->net)) {
+               (void) unlink_urbs (dev, &dev->rxq);
+               tasklet_schedule(&dev->bh);
+       }
+}
+EXPORT_SYMBOL_GPL(usbnet_unlink_rx_urbs);
+
+/*-------------------------------------------------------------------------*/
+
+// precondition: never called in_interrupt
+
+static int usbnet_stop (struct net_device *net)
+{
+       struct usbnet           *dev = netdev_priv(net);
+       int                     temp;
+       DECLARE_WAIT_QUEUE_HEAD_ONSTACK (unlink_wakeup);
+       DECLARE_WAITQUEUE (wait, current);
+
+       netif_stop_queue (net);
+
+       if (netif_msg_ifdown (dev))
+               devinfo (dev, "stop stats: rx/tx %ld/%ld, errs %ld/%ld",
+                       dev->stats.rx_packets, dev->stats.tx_packets,
+                       dev->stats.rx_errors, dev->stats.tx_errors
+                       );
+
+       // ensure there are no more active urbs
+       add_wait_queue (&unlink_wakeup, &wait);
+       dev->wait = &unlink_wakeup;
+       temp = unlink_urbs (dev, &dev->txq) + unlink_urbs (dev, &dev->rxq);
+
+       // maybe wait for deletions to finish.
+       while (!skb_queue_empty(&dev->rxq) &&
+              !skb_queue_empty(&dev->txq) &&
+              !skb_queue_empty(&dev->done)) {
+               msleep(UNLINK_TIMEOUT_MS);
+               if (netif_msg_ifdown (dev))
+                       devdbg (dev, "waited for %d urb completions", temp);
+       }
+       dev->wait = NULL;
+       remove_wait_queue (&unlink_wakeup, &wait);
+
+       usb_kill_urb(dev->interrupt);
+
+       /* deferred work (task, timer, softirq) must also stop.
+        * can't flush_scheduled_work() until we drop rtnl (later),
+        * else workers could deadlock; so make workers a NOP.
+        */
+       dev->flags = 0;
+       del_timer_sync (&dev->delay);
+       tasklet_kill (&dev->bh);
+
+       return 0;
+}
+
+/*-------------------------------------------------------------------------*/
+
+// posts reads, and enables write queuing
+
+// precondition: never called in_interrupt
+
+static int usbnet_open (struct net_device *net)
+{
+       struct usbnet           *dev = netdev_priv(net);
+       int                     retval = 0;
+       struct driver_info      *info = dev->driver_info;
+
+       // put into "known safe" state
+       if (info->reset && (retval = info->reset (dev)) < 0) {
+               if (netif_msg_ifup (dev))
+                       devinfo (dev,
+                               "open reset fail (%d) usbnet usb-%s-%s, %s",
+                               retval,
+                               dev->udev->bus->bus_name, dev->udev->devpath,
+                       info->description);
+               goto done;
+       }
+
+       // insist peer be connected
+       if (info->check_connect && (retval = info->check_connect (dev)) < 0) {
+               if (netif_msg_ifup (dev))
+                       devdbg (dev, "can't open; %d", retval);
+               goto done;
+       }
+
+       /* start any status interrupt transfer */
+       if (dev->interrupt) {
+               retval = usb_submit_urb (dev->interrupt, GFP_KERNEL);
+               if (retval < 0) {
+                       if (netif_msg_ifup (dev))
+                               deverr (dev, "intr submit %d", retval);
+                       goto done;
+               }
+       }
+
+       netif_start_queue (net);
+       if (netif_msg_ifup (dev)) {
+               char    *framing;
+
+               if (dev->driver_info->flags & FLAG_FRAMING_NC)
+                       framing = "NetChip";
+               else if (dev->driver_info->flags & FLAG_FRAMING_GL)
+                       framing = "GeneSys";
+               else if (dev->driver_info->flags & FLAG_FRAMING_Z)
+                       framing = "Zaurus";
+               else if (dev->driver_info->flags & FLAG_FRAMING_RN)
+                       framing = "RNDIS";
+               else if (dev->driver_info->flags & FLAG_FRAMING_AX)
+                       framing = "ASIX";
+               else
+                       framing = "simple";
+
+               devinfo (dev, "open: enable queueing "
+                               "(rx %d, tx %d) mtu %d %s framing",
+                       (int)RX_QLEN (dev), (int)TX_QLEN (dev), dev->net->mtu,
+                       framing);
+       }
+
+       // delay posting reads until we're fully open
+       tasklet_schedule (&dev->bh);
+done:
+       return retval;
+}
+
+/*-------------------------------------------------------------------------*/
+
+/* ethtool methods; minidrivers may need to add some more, but
+ * they'll probably want to use this base set.
+ */
+
+#if defined(CONFIG_MII) || defined(CONFIG_MII_MODULE)
+#define HAVE_MII
+
+int usbnet_get_settings (struct net_device *net, struct ethtool_cmd *cmd)
+{
+       struct usbnet *dev = netdev_priv(net);
+
+       if (!dev->mii.mdio_read)
+               return -EOPNOTSUPP;
+
+       return mii_ethtool_gset(&dev->mii, cmd);
+}
+EXPORT_SYMBOL_GPL(usbnet_get_settings);
+
+int usbnet_set_settings (struct net_device *net, struct ethtool_cmd *cmd)
+{
+       struct usbnet *dev = netdev_priv(net);
+       int retval;
+
+       if (!dev->mii.mdio_write)
+               return -EOPNOTSUPP;
+
+       retval = mii_ethtool_sset(&dev->mii, cmd);
+
+       /* link speed/duplex might have changed */
+       if (dev->driver_info->link_reset)
+               dev->driver_info->link_reset(dev);
+
+       return retval;
+
+}
+EXPORT_SYMBOL_GPL(usbnet_set_settings);
+
+u32 usbnet_get_link (struct net_device *net)
+{
+       struct usbnet *dev = netdev_priv(net);
+
+       /* If a check_connect is defined, return its result */
+       if (dev->driver_info->check_connect)
+               return dev->driver_info->check_connect (dev) == 0;
+
+       /* if the device has mii operations, use those */
+       if (dev->mii.mdio_read)
+               return mii_link_ok(&dev->mii);
+
+       /* Otherwise, say we're up (to avoid breaking scripts) */
+       return 1;
+}
+EXPORT_SYMBOL_GPL(usbnet_get_link);
+
+int usbnet_nway_reset(struct net_device *net)
+{
+       struct usbnet *dev = netdev_priv(net);
+
+       if (!dev->mii.mdio_write)
+               return -EOPNOTSUPP;
+
+       return mii_nway_restart(&dev->mii);
+}
+EXPORT_SYMBOL_GPL(usbnet_nway_reset);
+
+#endif /* HAVE_MII */
+
+void usbnet_get_drvinfo (struct net_device *net, struct ethtool_drvinfo *info)
+{
+       struct usbnet *dev = netdev_priv(net);
+
+       strncpy (info->driver, dev->driver_name, sizeof info->driver);
+       strncpy (info->version, DRIVER_VERSION, sizeof info->version);
+       strncpy (info->fw_version, dev->driver_info->description,
+               sizeof info->fw_version);
+       usb_make_path (dev->udev, info->bus_info, sizeof info->bus_info);
+}
+EXPORT_SYMBOL_GPL(usbnet_get_drvinfo);
+
+u32 usbnet_get_msglevel (struct net_device *net)
+{
+       struct usbnet *dev = netdev_priv(net);
+
+       return dev->msg_enable;
+}
+EXPORT_SYMBOL_GPL(usbnet_get_msglevel);
+
+void usbnet_set_msglevel (struct net_device *net, u32 level)
+{
+       struct usbnet *dev = netdev_priv(net);
+
+       dev->msg_enable = level;
+}
+EXPORT_SYMBOL_GPL(usbnet_set_msglevel);
+
+/* drivers may override default ethtool_ops in their bind() routine */
+static struct ethtool_ops usbnet_ethtool_ops = {
+#ifdef HAVE_MII
+       .get_settings           = usbnet_get_settings,
+       .set_settings           = usbnet_set_settings,
+       .get_link               = usbnet_get_link,
+       .nway_reset             = usbnet_nway_reset,
+#endif
+       .get_drvinfo            = usbnet_get_drvinfo,
+       .get_msglevel           = usbnet_get_msglevel,
+       .set_msglevel           = usbnet_set_msglevel,
+};
+
+/*-------------------------------------------------------------------------*/
+
+/* work that cannot be done in interrupt context uses keventd.
+ *
+ * NOTE:  with 2.5 we could do more of this using completion callbacks,
+ * especially now that control transfers can be queued.
+ */
+static void
+kevent (struct work_struct *work)
+{
+       struct usbnet           *dev =
+               container_of(work, struct usbnet, kevent);
+       int                     status;
+
+       /* usb_clear_halt() needs a thread context */
+       if (test_bit (EVENT_TX_HALT, &dev->flags)) {
+               unlink_urbs (dev, &dev->txq);
+               status = usb_clear_halt (dev->udev, dev->out);
+               if (status < 0
+                               && status != -EPIPE
+                               && status != -ESHUTDOWN) {
+                       if (netif_msg_tx_err (dev))
+                               deverr (dev, "can't clear tx halt, status %d",
+                                       status);
+               } else {
+                       clear_bit (EVENT_TX_HALT, &dev->flags);
+                       if (status != -ESHUTDOWN)
+                               netif_wake_queue (dev->net);
+               }
+       }
+       if (test_bit (EVENT_RX_HALT, &dev->flags)) {
+               unlink_urbs (dev, &dev->rxq);
+               status = usb_clear_halt (dev->udev, dev->in);
+               if (status < 0
+                               && status != -EPIPE
+                               && status != -ESHUTDOWN) {
+                       if (netif_msg_rx_err (dev))
+                               deverr (dev, "can't clear rx halt, status %d",
+                                       status);
+               } else {
+                       clear_bit (EVENT_RX_HALT, &dev->flags);
+                       tasklet_schedule (&dev->bh);
+               }
+       }
+
+       /* tasklet could resubmit itself forever if memory is tight */
+       if (test_bit (EVENT_RX_MEMORY, &dev->flags)) {
+               struct urb      *urb = NULL;
+
+               if (netif_running (dev->net))
+                       urb = usb_alloc_urb (0, GFP_KERNEL);
+               else
+                       clear_bit (EVENT_RX_MEMORY, &dev->flags);
+               if (urb != NULL) {
+                       clear_bit (EVENT_RX_MEMORY, &dev->flags);
+                       rx_submit (dev, urb, GFP_KERNEL);
+                       tasklet_schedule (&dev->bh);
+               }
+       }
+
+       if (test_bit (EVENT_LINK_RESET, &dev->flags)) {
+               struct driver_info      *info = dev->driver_info;
+               int                     retval = 0;
+
+               clear_bit (EVENT_LINK_RESET, &dev->flags);
+               if(info->link_reset && (retval = info->link_reset(dev)) < 0) {
+                       devinfo(dev, "link reset failed (%d) usbnet usb-%s-%s, %s",
+                               retval,
+                               dev->udev->bus->bus_name, dev->udev->devpath,
+                               info->description);
+               }
+       }
+
+       if (dev->flags)
+               devdbg (dev, "kevent done, flags = 0x%lx",
+                       dev->flags);
+}
+
+/*-------------------------------------------------------------------------*/
+
+static void tx_complete (struct urb *urb)
+{
+       struct sk_buff          *skb = (struct sk_buff *) urb->context;
+       struct skb_data         *entry = (struct skb_data *) skb->cb;
+       struct usbnet           *dev = entry->dev;
+
+       if (urb->status == 0) {
+               dev->stats.tx_packets++;
+               dev->stats.tx_bytes += entry->length;
+       } else {
+               dev->stats.tx_errors++;
+
+               switch (urb->status) {
+               case -EPIPE:
+                       usbnet_defer_kevent (dev, EVENT_TX_HALT);
+                       break;
+
+               /* software-driven interface shutdown */
+               case -ECONNRESET:               // async unlink
+               case -ESHUTDOWN:                // hardware gone
+                       break;
+
+               // like rx, tx gets controller i/o faults during khubd delays
+               // and so it uses the same throttling mechanism.
+               case -EPROTO:
+               case -ETIME:
+               case -EILSEQ:
+                       if (!timer_pending (&dev->delay)) {
+                               mod_timer (&dev->delay,
+                                       jiffies + THROTTLE_JIFFIES);
+                               if (netif_msg_link (dev))
+                                       devdbg (dev, "tx throttle %d",
+                                                       urb->status);
+                       }
+                       netif_stop_queue (dev->net);
+                       break;
+               default:
+                       if (netif_msg_tx_err (dev))
+                               devdbg (dev, "tx err %d", entry->urb->status);
+                       break;
+               }
+       }
+
+       urb->dev = NULL;
+       entry->state = tx_done;
+       defer_bh(dev, skb, &dev->txq);
+}
+
+/*-------------------------------------------------------------------------*/
+
+static void usbnet_tx_timeout (struct net_device *net)
+{
+       struct usbnet           *dev = netdev_priv(net);
+
+       unlink_urbs (dev, &dev->txq);
+       tasklet_schedule (&dev->bh);
+
+       // FIXME: device recovery -- reset?
+}
+
+/*-------------------------------------------------------------------------*/
+
+static int usbnet_start_xmit (struct sk_buff *skb, struct net_device *net)
+{
+       struct usbnet           *dev = netdev_priv(net);
+       int                     length;
+       int                     retval = NET_XMIT_SUCCESS;
+       struct urb              *urb = NULL;
+       struct skb_data         *entry;
+       struct driver_info      *info = dev->driver_info;
+       unsigned long           flags;
+
+       // some devices want funky USB-level framing, for
+       // win32 driver (usually) and/or hardware quirks
+       if (info->tx_fixup) {
+               skb = info->tx_fixup (dev, skb, GFP_ATOMIC);
+               if (!skb) {
+                       if (netif_msg_tx_err (dev))
+                               devdbg (dev, "can't tx_fixup skb");
+                       goto drop;
+               }
+       }
+       length = skb->len;
+
+       if (!(urb = usb_alloc_urb (0, GFP_ATOMIC))) {
+               if (netif_msg_tx_err (dev))
+                       devdbg (dev, "no urb");
+               goto drop;
+       }
+
+       entry = (struct skb_data *) skb->cb;
+       entry->urb = urb;
+       entry->dev = dev;
+       entry->state = tx_start;
+       entry->length = length;
+
+       usb_fill_bulk_urb (urb, dev->udev, dev->out,
+                       skb->data, skb->len, tx_complete, skb);
+
+       /* don't assume the hardware handles USB_ZERO_PACKET
+        * NOTE:  strictly conforming cdc-ether devices should expect
+        * the ZLP here, but ignore the one-byte packet.
+        *
+        * FIXME zero that byte, if it doesn't require a new skb.
+        */
+       if ((length % dev->maxpacket) == 0)
+               urb->transfer_buffer_length++;
+
+       spin_lock_irqsave (&dev->txq.lock, flags);
+
+       switch ((retval = usb_submit_urb (urb, GFP_ATOMIC))) {
+       case -EPIPE:
+               netif_stop_queue (net);
+               usbnet_defer_kevent (dev, EVENT_TX_HALT);
+               break;
+       default:
+               if (netif_msg_tx_err (dev))
+                       devdbg (dev, "tx: submit urb err %d", retval);
+               break;
+       case 0:
+               net->trans_start = jiffies;
+               __skb_queue_tail (&dev->txq, skb);
+               if (dev->txq.qlen >= TX_QLEN (dev))
+                       netif_stop_queue (net);
+       }
+       spin_unlock_irqrestore (&dev->txq.lock, flags);
+
+       if (retval) {
+               if (netif_msg_tx_err (dev))
+                       devdbg (dev, "drop, code %d", retval);
+drop:
+               retval = NET_XMIT_SUCCESS;
+               dev->stats.tx_dropped++;
+               if (skb)
+                       dev_kfree_skb_any (skb);
+               usb_free_urb (urb);
+       } else if (netif_msg_tx_queued (dev)) {
+               devdbg (dev, "> tx, len %d, type 0x%x",
+                       length, skb->protocol);
+       }
+       return retval;
+}
+
+
+/*-------------------------------------------------------------------------*/
+
+// tasklet (work deferred from completions, in_irq) or timer
+
+static void usbnet_bh (unsigned long param)
+{
+       struct usbnet           *dev = (struct usbnet *) param;
+       struct sk_buff          *skb;
+       struct skb_data         *entry;
+
+       while ((skb = skb_dequeue (&dev->done))) {
+               entry = (struct skb_data *) skb->cb;
+               switch (entry->state) {
+                   case rx_done:
+                       entry->state = rx_cleanup;
+                       rx_process (dev, skb);
+                       continue;
+                   case tx_done:
+                   case rx_cleanup:
+                       usb_free_urb (entry->urb);
+                       dev_kfree_skb (skb);
+                       continue;
+                   default:
+                       devdbg (dev, "bogus skb state %d", entry->state);
+               }
+       }
+
+       // waiting for all pending urbs to complete?
+       if (dev->wait) {
+               if ((dev->txq.qlen + dev->rxq.qlen + dev->done.qlen) == 0) {
+                       wake_up (dev->wait);
+               }
+
+       // or are we maybe short a few urbs?
+       } else if (netif_running (dev->net)
+                       && netif_device_present (dev->net)
+                       && !timer_pending (&dev->delay)
+                       && !test_bit (EVENT_RX_HALT, &dev->flags)) {
+               int     temp = dev->rxq.qlen;
+               int     qlen = RX_QLEN (dev);
+
+               if (temp < qlen) {
+                       struct urb      *urb;
+                       int             i;
+
+                       // don't refill the queue all at once
+                       for (i = 0; i < 10 && dev->rxq.qlen < qlen; i++) {
+                               urb = usb_alloc_urb (0, GFP_ATOMIC);
+                               if (urb != NULL)
+                                       rx_submit (dev, urb, GFP_ATOMIC);
+                       }
+                       if (temp != dev->rxq.qlen && netif_msg_link (dev))
+                               devdbg (dev, "rxqlen %d --> %d",
+                                               temp, dev->rxq.qlen);
+                       if (dev->rxq.qlen < qlen)
+                               tasklet_schedule (&dev->bh);
+               }
+               if (dev->txq.qlen < TX_QLEN (dev))
+                       netif_wake_queue (dev->net);
+       }
+}
+
+
+\f
+/*-------------------------------------------------------------------------
+ *
+ * USB Device Driver support
+ *
+ *-------------------------------------------------------------------------*/
+
+// precondition: never called in_interrupt
+
+void usbnet_disconnect (struct usb_interface *intf)
+{
+       struct usbnet           *dev;
+       struct usb_device       *xdev;
+       struct net_device       *net;
+
+       dev = usb_get_intfdata(intf);
+       usb_set_intfdata(intf, NULL);
+       if (!dev)
+               return;
+
+       xdev = interface_to_usbdev (intf);
+
+       if (netif_msg_probe (dev))
+               devinfo (dev, "unregister '%s' usb-%s-%s, %s",
+                       intf->dev.driver->name,
+                       xdev->bus->bus_name, xdev->devpath,
+                       dev->driver_info->description);
+
+       net = dev->net;
+       unregister_netdev (net);
+
+       /* we don't hold rtnl here ... */
+       flush_scheduled_work ();
+
+       if (dev->driver_info->unbind)
+               dev->driver_info->unbind (dev, intf);
+
+       free_netdev(net);
+       usb_put_dev (xdev);
+}
+EXPORT_SYMBOL_GPL(usbnet_disconnect);
+
+
+/*-------------------------------------------------------------------------*/
+
+// precondition: never called in_interrupt
+
+int
+usbnet_probe (struct usb_interface *udev, const struct usb_device_id *prod)
+{
+       struct usbnet                   *dev;
+       struct net_device               *net;
+       struct usb_host_interface       *interface;
+       struct driver_info              *info;
+       struct usb_device               *xdev;
+       int                             status;
+       const char                      *name;
+
+       name = udev->dev.driver->name;
+       info = (struct driver_info *) prod->driver_info;
+       if (!info) {
+               dev_dbg (&udev->dev, "blacklisted by %s\n", name);
+               return -ENODEV;
+       }
+       xdev = interface_to_usbdev (udev);
+       interface = udev->cur_altsetting;
+
+       usb_get_dev (xdev);
+
+       status = -ENOMEM;
+
+       // set up our own records
+       net = alloc_etherdev(sizeof(*dev));
+       if (!net) {
+               dbg ("can't kmalloc dev");
+               goto out;
+       }
+
+       dev = netdev_priv(net);
+       dev->udev = xdev;
+       dev->driver_info = info;
+       dev->driver_name = name;
+       dev->msg_enable = netif_msg_init (msg_level, NETIF_MSG_DRV
+                               | NETIF_MSG_PROBE | NETIF_MSG_LINK);
+       skb_queue_head_init (&dev->rxq);
+       skb_queue_head_init (&dev->txq);
+       skb_queue_head_init (&dev->done);
+       dev->bh.func = usbnet_bh;
+       dev->bh.data = (unsigned long) dev;
+       INIT_WORK (&dev->kevent, kevent);
+       dev->delay.function = usbnet_bh;
+       dev->delay.data = (unsigned long) dev;
+       init_timer (&dev->delay);
+       mutex_init (&dev->phy_mutex);
+
+       SET_MODULE_OWNER (net);
+       dev->net = net;
+       strcpy (net->name, "usb%d");
+       memcpy (net->dev_addr, node_id, sizeof node_id);
+
+       /* rx and tx sides can use different message sizes;
+        * bind() should set rx_urb_size in that case.
+        */
+       dev->hard_mtu = net->mtu + net->hard_header_len;
+#if 0
+// dma_supported() is deeply broken on almost all architectures
+       // possible with some EHCI controllers
+       if (dma_supported (&udev->dev, DMA_64BIT_MASK))
+               net->features |= NETIF_F_HIGHDMA;
+#endif
+
+       net->change_mtu = usbnet_change_mtu;
+       net->get_stats = usbnet_get_stats;
+       net->hard_start_xmit = usbnet_start_xmit;
+       net->open = usbnet_open;
+       net->stop = usbnet_stop;
+       net->watchdog_timeo = TX_TIMEOUT_JIFFIES;
+       net->tx_timeout = usbnet_tx_timeout;
+       net->ethtool_ops = &usbnet_ethtool_ops;
+
+       // allow device-specific bind/init procedures
+       // NOTE net->name still not usable ...
+       if (info->bind) {
+               status = info->bind (dev, udev);
+               if (status < 0)
+                       goto out1;
+
+               // heuristic:  "usb%d" for links we know are two-host,
+               // else "eth%d" when there's reasonable doubt.  userspace
+               // can rename the link if it knows better.
+               if ((dev->driver_info->flags & FLAG_ETHER) != 0
+                               && (net->dev_addr [0] & 0x02) == 0)
+                       strcpy (net->name, "eth%d");
+
+               /* maybe the remote can't receive an Ethernet MTU */
+               if (net->mtu > (dev->hard_mtu - net->hard_header_len))
+                       net->mtu = dev->hard_mtu - net->hard_header_len;
+       } else if (!info->in || !info->out)
+               status = usbnet_get_endpoints (dev, udev);
+       else {
+               dev->in = usb_rcvbulkpipe (xdev, info->in);
+               dev->out = usb_sndbulkpipe (xdev, info->out);
+               if (!(info->flags & FLAG_NO_SETINT))
+                       status = usb_set_interface (xdev,
+                               interface->desc.bInterfaceNumber,
+                               interface->desc.bAlternateSetting);
+               else
+                       status = 0;
+
+       }
+       if (status == 0 && dev->status)
+               status = init_status (dev, udev);
+       if (status < 0)
+               goto out3;
+
+       if (!dev->rx_urb_size)
+               dev->rx_urb_size = dev->hard_mtu;
+       dev->maxpacket = usb_maxpacket (dev->udev, dev->out, 1);
+
+       SET_NETDEV_DEV(net, &udev->dev);
+       status = register_netdev (net);
+       if (status)
+               goto out3;
+       if (netif_msg_probe (dev))
+               devinfo (dev, "register '%s' at usb-%s-%s, %s, "
+                               "%02x:%02x:%02x:%02x:%02x:%02x",
+                       udev->dev.driver->name,
+                       xdev->bus->bus_name, xdev->devpath,
+                       dev->driver_info->description,
+                       net->dev_addr [0], net->dev_addr [1],
+                       net->dev_addr [2], net->dev_addr [3],
+                       net->dev_addr [4], net->dev_addr [5]);
+
+       // ok, it's ready to go.
+       usb_set_intfdata (udev, dev);
+
+       // start as if the link is up
+       netif_device_attach (net);
+
+       return 0;
+
+out3:
+       if (info->unbind)
+               info->unbind (dev, udev);
+out1:
+       free_netdev(net);
+out:
+       usb_put_dev(xdev);
+       return status;
+}
+EXPORT_SYMBOL_GPL(usbnet_probe);
+
+/*-------------------------------------------------------------------------*/
+
+/* FIXME these suspend/resume methods assume non-CDC style
+ * devices, with only one interface.
+ */
+
+int usbnet_suspend (struct usb_interface *intf, pm_message_t message)
+{
+       struct usbnet           *dev = usb_get_intfdata(intf);
+
+       /* accelerate emptying of the rx and queues, to avoid
+        * having everything error out.
+        */
+       netif_device_detach (dev->net);
+       (void) unlink_urbs (dev, &dev->rxq);
+       (void) unlink_urbs (dev, &dev->txq);
+       return 0;
+}
+EXPORT_SYMBOL_GPL(usbnet_suspend);
+
+int usbnet_resume (struct usb_interface *intf)
+{
+       struct usbnet           *dev = usb_get_intfdata(intf);
+
+       netif_device_attach (dev->net);
+       tasklet_schedule (&dev->bh);
+       return 0;
+}
+EXPORT_SYMBOL_GPL(usbnet_resume);
+
+
+/*-------------------------------------------------------------------------*/
+
+static int __init usbnet_init(void)
+{
+       /* compiler should optimize this out */
+       BUILD_BUG_ON (sizeof (((struct sk_buff *)0)->cb)
+                       < sizeof (struct skb_data));
+
+       random_ether_addr(node_id);
+       return 0;
+}
+module_init(usbnet_init);
+
+static void __exit usbnet_exit(void)
+{
+}
+module_exit(usbnet_exit);
+
+MODULE_AUTHOR("David Brownell");
+MODULE_DESCRIPTION("USB network driver framework");
+MODULE_LICENSE("GPL");