Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / usb / gadget / function / u_ether.c
diff --git a/kernel/drivers/usb/gadget/function/u_ether.c b/kernel/drivers/usb/gadget/function/u_ether.c
new file mode 100644 (file)
index 0000000..f1fd777
--- /dev/null
@@ -0,0 +1,1177 @@
+/*
+ * u_ether.c -- Ethernet-over-USB link layer utilities for Gadget stack
+ *
+ * Copyright (C) 2003-2005,2008 David Brownell
+ * Copyright (C) 2003-2004 Robert Schwebel, Benedikt Spranger
+ * Copyright (C) 2008 Nokia Corporation
+ *
+ * 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.
+ */
+
+/* #define VERBOSE_DEBUG */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/gfp.h>
+#include <linux/device.h>
+#include <linux/ctype.h>
+#include <linux/etherdevice.h>
+#include <linux/ethtool.h>
+#include <linux/if_vlan.h>
+
+#include "u_ether.h"
+
+
+/*
+ * This component encapsulates the Ethernet link glue needed to provide
+ * one (!) network link through the USB gadget stack, normally "usb0".
+ *
+ * The control and data models are handled by the function driver which
+ * connects to this code; such as CDC Ethernet (ECM or EEM),
+ * "CDC Subset", or RNDIS.  That includes all descriptor and endpoint
+ * management.
+ *
+ * Link level addressing is handled by this component using module
+ * parameters; if no such parameters are provided, random link level
+ * addresses are used.  Each end of the link uses one address.  The
+ * host end address is exported in various ways, and is often recorded
+ * in configuration databases.
+ *
+ * The driver which assembles each configuration using such a link is
+ * responsible for ensuring that each configuration includes at most one
+ * instance of is network link.  (The network layer provides ways for
+ * this single "physical" link to be used by multiple virtual links.)
+ */
+
+#define UETH__VERSION  "29-May-2008"
+
+struct eth_dev {
+       /* lock is held while accessing port_usb
+        */
+       spinlock_t              lock;
+       struct gether           *port_usb;
+
+       struct net_device       *net;
+       struct usb_gadget       *gadget;
+
+       spinlock_t              req_lock;       /* guard {rx,tx}_reqs */
+       struct list_head        tx_reqs, rx_reqs;
+       atomic_t                tx_qlen;
+
+       struct sk_buff_head     rx_frames;
+
+       unsigned                qmult;
+
+       unsigned                header_len;
+       struct sk_buff          *(*wrap)(struct gether *, struct sk_buff *skb);
+       int                     (*unwrap)(struct gether *,
+                                               struct sk_buff *skb,
+                                               struct sk_buff_head *list);
+
+       struct work_struct      work;
+
+       unsigned long           todo;
+#define        WORK_RX_MEMORY          0
+
+       bool                    zlp;
+       u8                      host_mac[ETH_ALEN];
+       u8                      dev_mac[ETH_ALEN];
+};
+
+/*-------------------------------------------------------------------------*/
+
+#define RX_EXTRA       20      /* bytes guarding against rx overflows */
+
+#define DEFAULT_QLEN   2       /* double buffering by default */
+
+/* for dual-speed hardware, use deeper queues at high/super speed */
+static inline int qlen(struct usb_gadget *gadget, unsigned qmult)
+{
+       if (gadget_is_dualspeed(gadget) && (gadget->speed == USB_SPEED_HIGH ||
+                                           gadget->speed == USB_SPEED_SUPER))
+               return qmult * DEFAULT_QLEN;
+       else
+               return DEFAULT_QLEN;
+}
+
+/*-------------------------------------------------------------------------*/
+
+/* REVISIT there must be a better way than having two sets
+ * of debug calls ...
+ */
+
+#undef DBG
+#undef VDBG
+#undef ERROR
+#undef INFO
+
+#define xprintk(d, level, fmt, args...) \
+       printk(level "%s: " fmt , (d)->net->name , ## args)
+
+#ifdef DEBUG
+#undef DEBUG
+#define DBG(dev, fmt, args...) \
+       xprintk(dev , KERN_DEBUG , fmt , ## args)
+#else
+#define DBG(dev, fmt, args...) \
+       do { } while (0)
+#endif /* DEBUG */
+
+#ifdef VERBOSE_DEBUG
+#define VDBG   DBG
+#else
+#define VDBG(dev, fmt, args...) \
+       do { } while (0)
+#endif /* DEBUG */
+
+#define ERROR(dev, fmt, args...) \
+       xprintk(dev , KERN_ERR , fmt , ## args)
+#define INFO(dev, fmt, args...) \
+       xprintk(dev , KERN_INFO , fmt , ## args)
+
+/*-------------------------------------------------------------------------*/
+
+/* NETWORK DRIVER HOOKUP (to the layer above this driver) */
+
+static int ueth_change_mtu(struct net_device *net, int new_mtu)
+{
+       struct eth_dev  *dev = netdev_priv(net);
+       unsigned long   flags;
+       int             status = 0;
+
+       /* don't change MTU on "live" link (peer won't know) */
+       spin_lock_irqsave(&dev->lock, flags);
+       if (dev->port_usb)
+               status = -EBUSY;
+       else if (new_mtu <= ETH_HLEN || new_mtu > ETH_FRAME_LEN)
+               status = -ERANGE;
+       else
+               net->mtu = new_mtu;
+       spin_unlock_irqrestore(&dev->lock, flags);
+
+       return status;
+}
+
+static void eth_get_drvinfo(struct net_device *net, struct ethtool_drvinfo *p)
+{
+       struct eth_dev *dev = netdev_priv(net);
+
+       strlcpy(p->driver, "g_ether", sizeof(p->driver));
+       strlcpy(p->version, UETH__VERSION, sizeof(p->version));
+       strlcpy(p->fw_version, dev->gadget->name, sizeof(p->fw_version));
+       strlcpy(p->bus_info, dev_name(&dev->gadget->dev), sizeof(p->bus_info));
+}
+
+/* REVISIT can also support:
+ *   - WOL (by tracking suspends and issuing remote wakeup)
+ *   - msglevel (implies updated messaging)
+ *   - ... probably more ethtool ops
+ */
+
+static const struct ethtool_ops ops = {
+       .get_drvinfo = eth_get_drvinfo,
+       .get_link = ethtool_op_get_link,
+};
+
+static void defer_kevent(struct eth_dev *dev, int flag)
+{
+       if (test_and_set_bit(flag, &dev->todo))
+               return;
+       if (!schedule_work(&dev->work))
+               ERROR(dev, "kevent %d may have been dropped\n", flag);
+       else
+               DBG(dev, "kevent %d scheduled\n", flag);
+}
+
+static void rx_complete(struct usb_ep *ep, struct usb_request *req);
+
+static int
+rx_submit(struct eth_dev *dev, struct usb_request *req, gfp_t gfp_flags)
+{
+       struct sk_buff  *skb;
+       int             retval = -ENOMEM;
+       size_t          size = 0;
+       struct usb_ep   *out;
+       unsigned long   flags;
+
+       spin_lock_irqsave(&dev->lock, flags);
+       if (dev->port_usb)
+               out = dev->port_usb->out_ep;
+       else
+               out = NULL;
+       spin_unlock_irqrestore(&dev->lock, flags);
+
+       if (!out)
+               return -ENOTCONN;
+
+
+       /* Padding up to RX_EXTRA handles minor disagreements with host.
+        * Normally we use the USB "terminate on short read" convention;
+        * so allow up to (N*maxpacket), since that memory is normally
+        * already allocated.  Some hardware doesn't deal well with short
+        * reads (e.g. DMA must be N*maxpacket), so for now don't trim a
+        * byte off the end (to force hardware errors on overflow).
+        *
+        * RNDIS uses internal framing, and explicitly allows senders to
+        * pad to end-of-packet.  That's potentially nice for speed, but
+        * means receivers can't recover lost synch on their own (because
+        * new packets don't only start after a short RX).
+        */
+       size += sizeof(struct ethhdr) + dev->net->mtu + RX_EXTRA;
+       size += dev->port_usb->header_len;
+       size += out->maxpacket - 1;
+       size -= size % out->maxpacket;
+
+       if (dev->port_usb->is_fixed)
+               size = max_t(size_t, size, dev->port_usb->fixed_out_len);
+
+       skb = alloc_skb(size + NET_IP_ALIGN, gfp_flags);
+       if (skb == NULL) {
+               DBG(dev, "no rx skb\n");
+               goto enomem;
+       }
+
+       /* Some platforms perform better when IP packets are aligned,
+        * but on at least one, checksumming fails otherwise.  Note:
+        * RNDIS headers involve variable numbers of LE32 values.
+        */
+       skb_reserve(skb, NET_IP_ALIGN);
+
+       req->buf = skb->data;
+       req->length = size;
+       req->complete = rx_complete;
+       req->context = skb;
+
+       retval = usb_ep_queue(out, req, gfp_flags);
+       if (retval == -ENOMEM)
+enomem:
+               defer_kevent(dev, WORK_RX_MEMORY);
+       if (retval) {
+               DBG(dev, "rx submit --> %d\n", retval);
+               if (skb)
+                       dev_kfree_skb_any(skb);
+               spin_lock_irqsave(&dev->req_lock, flags);
+               list_add(&req->list, &dev->rx_reqs);
+               spin_unlock_irqrestore(&dev->req_lock, flags);
+       }
+       return retval;
+}
+
+static void rx_complete(struct usb_ep *ep, struct usb_request *req)
+{
+       struct sk_buff  *skb = req->context, *skb2;
+       struct eth_dev  *dev = ep->driver_data;
+       int             status = req->status;
+
+       switch (status) {
+
+       /* normal completion */
+       case 0:
+               skb_put(skb, req->actual);
+
+               if (dev->unwrap) {
+                       unsigned long   flags;
+
+                       spin_lock_irqsave(&dev->lock, flags);
+                       if (dev->port_usb) {
+                               status = dev->unwrap(dev->port_usb,
+                                                       skb,
+                                                       &dev->rx_frames);
+                       } else {
+                               dev_kfree_skb_any(skb);
+                               status = -ENOTCONN;
+                       }
+                       spin_unlock_irqrestore(&dev->lock, flags);
+               } else {
+                       skb_queue_tail(&dev->rx_frames, skb);
+               }
+               skb = NULL;
+
+               skb2 = skb_dequeue(&dev->rx_frames);
+               while (skb2) {
+                       if (status < 0
+                                       || ETH_HLEN > skb2->len
+                                       || skb2->len > VLAN_ETH_FRAME_LEN) {
+                               dev->net->stats.rx_errors++;
+                               dev->net->stats.rx_length_errors++;
+                               DBG(dev, "rx length %d\n", skb2->len);
+                               dev_kfree_skb_any(skb2);
+                               goto next_frame;
+                       }
+                       skb2->protocol = eth_type_trans(skb2, dev->net);
+                       dev->net->stats.rx_packets++;
+                       dev->net->stats.rx_bytes += skb2->len;
+
+                       /* no buffer copies needed, unless hardware can't
+                        * use skb buffers.
+                        */
+                       status = netif_rx(skb2);
+next_frame:
+                       skb2 = skb_dequeue(&dev->rx_frames);
+               }
+               break;
+
+       /* software-driven interface shutdown */
+       case -ECONNRESET:               /* unlink */
+       case -ESHUTDOWN:                /* disconnect etc */
+               VDBG(dev, "rx shutdown, code %d\n", status);
+               goto quiesce;
+
+       /* for hardware automagic (such as pxa) */
+       case -ECONNABORTED:             /* endpoint reset */
+               DBG(dev, "rx %s reset\n", ep->name);
+               defer_kevent(dev, WORK_RX_MEMORY);
+quiesce:
+               dev_kfree_skb_any(skb);
+               goto clean;
+
+       /* data overrun */
+       case -EOVERFLOW:
+               dev->net->stats.rx_over_errors++;
+               /* FALLTHROUGH */
+
+       default:
+               dev->net->stats.rx_errors++;
+               DBG(dev, "rx status %d\n", status);
+               break;
+       }
+
+       if (skb)
+               dev_kfree_skb_any(skb);
+       if (!netif_running(dev->net)) {
+clean:
+               spin_lock(&dev->req_lock);
+               list_add(&req->list, &dev->rx_reqs);
+               spin_unlock(&dev->req_lock);
+               req = NULL;
+       }
+       if (req)
+               rx_submit(dev, req, GFP_ATOMIC);
+}
+
+static int prealloc(struct list_head *list, struct usb_ep *ep, unsigned n)
+{
+       unsigned                i;
+       struct usb_request      *req;
+
+       if (!n)
+               return -ENOMEM;
+
+       /* queue/recycle up to N requests */
+       i = n;
+       list_for_each_entry(req, list, list) {
+               if (i-- == 0)
+                       goto extra;
+       }
+       while (i--) {
+               req = usb_ep_alloc_request(ep, GFP_ATOMIC);
+               if (!req)
+                       return list_empty(list) ? -ENOMEM : 0;
+               list_add(&req->list, list);
+       }
+       return 0;
+
+extra:
+       /* free extras */
+       for (;;) {
+               struct list_head        *next;
+
+               next = req->list.next;
+               list_del(&req->list);
+               usb_ep_free_request(ep, req);
+
+               if (next == list)
+                       break;
+
+               req = container_of(next, struct usb_request, list);
+       }
+       return 0;
+}
+
+static int alloc_requests(struct eth_dev *dev, struct gether *link, unsigned n)
+{
+       int     status;
+
+       spin_lock(&dev->req_lock);
+       status = prealloc(&dev->tx_reqs, link->in_ep, n);
+       if (status < 0)
+               goto fail;
+       status = prealloc(&dev->rx_reqs, link->out_ep, n);
+       if (status < 0)
+               goto fail;
+       goto done;
+fail:
+       DBG(dev, "can't alloc requests\n");
+done:
+       spin_unlock(&dev->req_lock);
+       return status;
+}
+
+static void rx_fill(struct eth_dev *dev, gfp_t gfp_flags)
+{
+       struct usb_request      *req;
+       unsigned long           flags;
+
+       /* fill unused rxq slots with some skb */
+       spin_lock_irqsave(&dev->req_lock, flags);
+       while (!list_empty(&dev->rx_reqs)) {
+               req = container_of(dev->rx_reqs.next,
+                               struct usb_request, list);
+               list_del_init(&req->list);
+               spin_unlock_irqrestore(&dev->req_lock, flags);
+
+               if (rx_submit(dev, req, gfp_flags) < 0) {
+                       defer_kevent(dev, WORK_RX_MEMORY);
+                       return;
+               }
+
+               spin_lock_irqsave(&dev->req_lock, flags);
+       }
+       spin_unlock_irqrestore(&dev->req_lock, flags);
+}
+
+static void eth_work(struct work_struct *work)
+{
+       struct eth_dev  *dev = container_of(work, struct eth_dev, work);
+
+       if (test_and_clear_bit(WORK_RX_MEMORY, &dev->todo)) {
+               if (netif_running(dev->net))
+                       rx_fill(dev, GFP_KERNEL);
+       }
+
+       if (dev->todo)
+               DBG(dev, "work done, flags = 0x%lx\n", dev->todo);
+}
+
+static void tx_complete(struct usb_ep *ep, struct usb_request *req)
+{
+       struct sk_buff  *skb = req->context;
+       struct eth_dev  *dev = ep->driver_data;
+
+       switch (req->status) {
+       default:
+               dev->net->stats.tx_errors++;
+               VDBG(dev, "tx err %d\n", req->status);
+               /* FALLTHROUGH */
+       case -ECONNRESET:               /* unlink */
+       case -ESHUTDOWN:                /* disconnect etc */
+               break;
+       case 0:
+               dev->net->stats.tx_bytes += skb->len;
+       }
+       dev->net->stats.tx_packets++;
+
+       spin_lock(&dev->req_lock);
+       list_add(&req->list, &dev->tx_reqs);
+       spin_unlock(&dev->req_lock);
+       dev_kfree_skb_any(skb);
+
+       atomic_dec(&dev->tx_qlen);
+       if (netif_carrier_ok(dev->net))
+               netif_wake_queue(dev->net);
+}
+
+static inline int is_promisc(u16 cdc_filter)
+{
+       return cdc_filter & USB_CDC_PACKET_TYPE_PROMISCUOUS;
+}
+
+static netdev_tx_t eth_start_xmit(struct sk_buff *skb,
+                                       struct net_device *net)
+{
+       struct eth_dev          *dev = netdev_priv(net);
+       int                     length = 0;
+       int                     retval;
+       struct usb_request      *req = NULL;
+       unsigned long           flags;
+       struct usb_ep           *in;
+       u16                     cdc_filter;
+
+       spin_lock_irqsave(&dev->lock, flags);
+       if (dev->port_usb) {
+               in = dev->port_usb->in_ep;
+               cdc_filter = dev->port_usb->cdc_filter;
+       } else {
+               in = NULL;
+               cdc_filter = 0;
+       }
+       spin_unlock_irqrestore(&dev->lock, flags);
+
+       if (skb && !in) {
+               dev_kfree_skb_any(skb);
+               return NETDEV_TX_OK;
+       }
+
+       /* apply outgoing CDC or RNDIS filters */
+       if (skb && !is_promisc(cdc_filter)) {
+               u8              *dest = skb->data;
+
+               if (is_multicast_ether_addr(dest)) {
+                       u16     type;
+
+                       /* ignores USB_CDC_PACKET_TYPE_MULTICAST and host
+                        * SET_ETHERNET_MULTICAST_FILTERS requests
+                        */
+                       if (is_broadcast_ether_addr(dest))
+                               type = USB_CDC_PACKET_TYPE_BROADCAST;
+                       else
+                               type = USB_CDC_PACKET_TYPE_ALL_MULTICAST;
+                       if (!(cdc_filter & type)) {
+                               dev_kfree_skb_any(skb);
+                               return NETDEV_TX_OK;
+                       }
+               }
+               /* ignores USB_CDC_PACKET_TYPE_DIRECTED */
+       }
+
+       spin_lock_irqsave(&dev->req_lock, flags);
+       /*
+        * this freelist can be empty if an interrupt triggered disconnect()
+        * and reconfigured the gadget (shutting down this queue) after the
+        * network stack decided to xmit but before we got the spinlock.
+        */
+       if (list_empty(&dev->tx_reqs)) {
+               spin_unlock_irqrestore(&dev->req_lock, flags);
+               return NETDEV_TX_BUSY;
+       }
+
+       req = container_of(dev->tx_reqs.next, struct usb_request, list);
+       list_del(&req->list);
+
+       /* temporarily stop TX queue when the freelist empties */
+       if (list_empty(&dev->tx_reqs))
+               netif_stop_queue(net);
+       spin_unlock_irqrestore(&dev->req_lock, flags);
+
+       /* no buffer copies needed, unless the network stack did it
+        * or the hardware can't use skb buffers.
+        * or there's not enough space for extra headers we need
+        */
+       if (dev->wrap) {
+               unsigned long   flags;
+
+               spin_lock_irqsave(&dev->lock, flags);
+               if (dev->port_usb)
+                       skb = dev->wrap(dev->port_usb, skb);
+               spin_unlock_irqrestore(&dev->lock, flags);
+               if (!skb) {
+                       /* Multi frame CDC protocols may store the frame for
+                        * later which is not a dropped frame.
+                        */
+                       if (dev->port_usb->supports_multi_frame)
+                               goto multiframe;
+                       goto drop;
+               }
+       }
+
+       length = skb->len;
+       req->buf = skb->data;
+       req->context = skb;
+       req->complete = tx_complete;
+
+       /* NCM requires no zlp if transfer is dwNtbInMaxSize */
+       if (dev->port_usb->is_fixed &&
+           length == dev->port_usb->fixed_in_len &&
+           (length % in->maxpacket) == 0)
+               req->zero = 0;
+       else
+               req->zero = 1;
+
+       /* use zlp framing on tx for strict CDC-Ether conformance,
+        * though any robust network rx path ignores extra padding.
+        * and some hardware doesn't like to write zlps.
+        */
+       if (req->zero && !dev->zlp && (length % in->maxpacket) == 0)
+               length++;
+
+       req->length = length;
+
+       /* throttle high/super speed IRQ rate back slightly */
+       if (gadget_is_dualspeed(dev->gadget))
+               req->no_interrupt = (dev->gadget->speed == USB_SPEED_HIGH ||
+                                    dev->gadget->speed == USB_SPEED_SUPER)
+                       ? ((atomic_read(&dev->tx_qlen) % dev->qmult) != 0)
+                       : 0;
+
+       retval = usb_ep_queue(in, req, GFP_ATOMIC);
+       switch (retval) {
+       default:
+               DBG(dev, "tx queue err %d\n", retval);
+               break;
+       case 0:
+               net->trans_start = jiffies;
+               atomic_inc(&dev->tx_qlen);
+       }
+
+       if (retval) {
+               dev_kfree_skb_any(skb);
+drop:
+               dev->net->stats.tx_dropped++;
+multiframe:
+               spin_lock_irqsave(&dev->req_lock, flags);
+               if (list_empty(&dev->tx_reqs))
+                       netif_start_queue(net);
+               list_add(&req->list, &dev->tx_reqs);
+               spin_unlock_irqrestore(&dev->req_lock, flags);
+       }
+       return NETDEV_TX_OK;
+}
+
+/*-------------------------------------------------------------------------*/
+
+static void eth_start(struct eth_dev *dev, gfp_t gfp_flags)
+{
+       DBG(dev, "%s\n", __func__);
+
+       /* fill the rx queue */
+       rx_fill(dev, gfp_flags);
+
+       /* and open the tx floodgates */
+       atomic_set(&dev->tx_qlen, 0);
+       netif_wake_queue(dev->net);
+}
+
+static int eth_open(struct net_device *net)
+{
+       struct eth_dev  *dev = netdev_priv(net);
+       struct gether   *link;
+
+       DBG(dev, "%s\n", __func__);
+       if (netif_carrier_ok(dev->net))
+               eth_start(dev, GFP_KERNEL);
+
+       spin_lock_irq(&dev->lock);
+       link = dev->port_usb;
+       if (link && link->open)
+               link->open(link);
+       spin_unlock_irq(&dev->lock);
+
+       return 0;
+}
+
+static int eth_stop(struct net_device *net)
+{
+       struct eth_dev  *dev = netdev_priv(net);
+       unsigned long   flags;
+
+       VDBG(dev, "%s\n", __func__);
+       netif_stop_queue(net);
+
+       DBG(dev, "stop stats: rx/tx %ld/%ld, errs %ld/%ld\n",
+               dev->net->stats.rx_packets, dev->net->stats.tx_packets,
+               dev->net->stats.rx_errors, dev->net->stats.tx_errors
+               );
+
+       /* ensure there are no more active requests */
+       spin_lock_irqsave(&dev->lock, flags);
+       if (dev->port_usb) {
+               struct gether   *link = dev->port_usb;
+               const struct usb_endpoint_descriptor *in;
+               const struct usb_endpoint_descriptor *out;
+
+               if (link->close)
+                       link->close(link);
+
+               /* NOTE:  we have no abort-queue primitive we could use
+                * to cancel all pending I/O.  Instead, we disable then
+                * reenable the endpoints ... this idiom may leave toggle
+                * wrong, but that's a self-correcting error.
+                *
+                * REVISIT:  we *COULD* just let the transfers complete at
+                * their own pace; the network stack can handle old packets.
+                * For the moment we leave this here, since it works.
+                */
+               in = link->in_ep->desc;
+               out = link->out_ep->desc;
+               usb_ep_disable(link->in_ep);
+               usb_ep_disable(link->out_ep);
+               if (netif_carrier_ok(net)) {
+                       DBG(dev, "host still using in/out endpoints\n");
+                       link->in_ep->desc = in;
+                       link->out_ep->desc = out;
+                       usb_ep_enable(link->in_ep);
+                       usb_ep_enable(link->out_ep);
+               }
+       }
+       spin_unlock_irqrestore(&dev->lock, flags);
+
+       return 0;
+}
+
+/*-------------------------------------------------------------------------*/
+
+static int get_ether_addr(const char *str, u8 *dev_addr)
+{
+       if (str) {
+               unsigned        i;
+
+               for (i = 0; i < 6; i++) {
+                       unsigned char num;
+
+                       if ((*str == '.') || (*str == ':'))
+                               str++;
+                       num = hex_to_bin(*str++) << 4;
+                       num |= hex_to_bin(*str++);
+                       dev_addr [i] = num;
+               }
+               if (is_valid_ether_addr(dev_addr))
+                       return 0;
+       }
+       eth_random_addr(dev_addr);
+       return 1;
+}
+
+static int get_ether_addr_str(u8 dev_addr[ETH_ALEN], char *str, int len)
+{
+       if (len < 18)
+               return -EINVAL;
+
+       snprintf(str, len, "%pM", dev_addr);
+       return 18;
+}
+
+static const struct net_device_ops eth_netdev_ops = {
+       .ndo_open               = eth_open,
+       .ndo_stop               = eth_stop,
+       .ndo_start_xmit         = eth_start_xmit,
+       .ndo_change_mtu         = ueth_change_mtu,
+       .ndo_set_mac_address    = eth_mac_addr,
+       .ndo_validate_addr      = eth_validate_addr,
+};
+
+static struct device_type gadget_type = {
+       .name   = "gadget",
+};
+
+/**
+ * gether_setup_name - initialize one ethernet-over-usb link
+ * @g: gadget to associated with these links
+ * @ethaddr: NULL, or a buffer in which the ethernet address of the
+ *     host side of the link is recorded
+ * @netname: name for network device (for example, "usb")
+ * Context: may sleep
+ *
+ * This sets up the single network link that may be exported by a
+ * gadget driver using this framework.  The link layer addresses are
+ * set up using module parameters.
+ *
+ * Returns an eth_dev pointer on success, or an ERR_PTR on failure.
+ */
+struct eth_dev *gether_setup_name(struct usb_gadget *g,
+               const char *dev_addr, const char *host_addr,
+               u8 ethaddr[ETH_ALEN], unsigned qmult, const char *netname)
+{
+       struct eth_dev          *dev;
+       struct net_device       *net;
+       int                     status;
+
+       net = alloc_etherdev(sizeof *dev);
+       if (!net)
+               return ERR_PTR(-ENOMEM);
+
+       dev = netdev_priv(net);
+       spin_lock_init(&dev->lock);
+       spin_lock_init(&dev->req_lock);
+       INIT_WORK(&dev->work, eth_work);
+       INIT_LIST_HEAD(&dev->tx_reqs);
+       INIT_LIST_HEAD(&dev->rx_reqs);
+
+       skb_queue_head_init(&dev->rx_frames);
+
+       /* network device setup */
+       dev->net = net;
+       dev->qmult = qmult;
+       snprintf(net->name, sizeof(net->name), "%s%%d", netname);
+
+       if (get_ether_addr(dev_addr, net->dev_addr))
+               dev_warn(&g->dev,
+                       "using random %s ethernet address\n", "self");
+       if (get_ether_addr(host_addr, dev->host_mac))
+               dev_warn(&g->dev,
+                       "using random %s ethernet address\n", "host");
+
+       if (ethaddr)
+               memcpy(ethaddr, dev->host_mac, ETH_ALEN);
+
+       net->netdev_ops = &eth_netdev_ops;
+
+       net->ethtool_ops = &ops;
+
+       dev->gadget = g;
+       SET_NETDEV_DEV(net, &g->dev);
+       SET_NETDEV_DEVTYPE(net, &gadget_type);
+
+       status = register_netdev(net);
+       if (status < 0) {
+               dev_dbg(&g->dev, "register_netdev failed, %d\n", status);
+               free_netdev(net);
+               dev = ERR_PTR(status);
+       } else {
+               INFO(dev, "MAC %pM\n", net->dev_addr);
+               INFO(dev, "HOST MAC %pM\n", dev->host_mac);
+
+               /*
+                * two kinds of host-initiated state changes:
+                *  - iff DATA transfer is active, carrier is "on"
+                *  - tx queueing enabled if open *and* carrier is "on"
+                */
+               netif_carrier_off(net);
+       }
+
+       return dev;
+}
+EXPORT_SYMBOL_GPL(gether_setup_name);
+
+struct net_device *gether_setup_name_default(const char *netname)
+{
+       struct net_device       *net;
+       struct eth_dev          *dev;
+
+       net = alloc_etherdev(sizeof(*dev));
+       if (!net)
+               return ERR_PTR(-ENOMEM);
+
+       dev = netdev_priv(net);
+       spin_lock_init(&dev->lock);
+       spin_lock_init(&dev->req_lock);
+       INIT_WORK(&dev->work, eth_work);
+       INIT_LIST_HEAD(&dev->tx_reqs);
+       INIT_LIST_HEAD(&dev->rx_reqs);
+
+       skb_queue_head_init(&dev->rx_frames);
+
+       /* network device setup */
+       dev->net = net;
+       dev->qmult = QMULT_DEFAULT;
+       snprintf(net->name, sizeof(net->name), "%s%%d", netname);
+
+       eth_random_addr(dev->dev_mac);
+       pr_warn("using random %s ethernet address\n", "self");
+       eth_random_addr(dev->host_mac);
+       pr_warn("using random %s ethernet address\n", "host");
+
+       net->netdev_ops = &eth_netdev_ops;
+
+       net->ethtool_ops = &ops;
+       SET_NETDEV_DEVTYPE(net, &gadget_type);
+
+       return net;
+}
+EXPORT_SYMBOL_GPL(gether_setup_name_default);
+
+int gether_register_netdev(struct net_device *net)
+{
+       struct eth_dev *dev;
+       struct usb_gadget *g;
+       struct sockaddr sa;
+       int status;
+
+       if (!net->dev.parent)
+               return -EINVAL;
+       dev = netdev_priv(net);
+       g = dev->gadget;
+       status = register_netdev(net);
+       if (status < 0) {
+               dev_dbg(&g->dev, "register_netdev failed, %d\n", status);
+               return status;
+       } else {
+               INFO(dev, "HOST MAC %pM\n", dev->host_mac);
+
+               /* two kinds of host-initiated state changes:
+                *  - iff DATA transfer is active, carrier is "on"
+                *  - tx queueing enabled if open *and* carrier is "on"
+                */
+               netif_carrier_off(net);
+       }
+       sa.sa_family = net->type;
+       memcpy(sa.sa_data, dev->dev_mac, ETH_ALEN);
+       rtnl_lock();
+       status = dev_set_mac_address(net, &sa);
+       rtnl_unlock();
+       if (status)
+               pr_warn("cannot set self ethernet address: %d\n", status);
+       else
+               INFO(dev, "MAC %pM\n", dev->dev_mac);
+
+       return status;
+}
+EXPORT_SYMBOL_GPL(gether_register_netdev);
+
+void gether_set_gadget(struct net_device *net, struct usb_gadget *g)
+{
+       struct eth_dev *dev;
+
+       dev = netdev_priv(net);
+       dev->gadget = g;
+       SET_NETDEV_DEV(net, &g->dev);
+}
+EXPORT_SYMBOL_GPL(gether_set_gadget);
+
+int gether_set_dev_addr(struct net_device *net, const char *dev_addr)
+{
+       struct eth_dev *dev;
+       u8 new_addr[ETH_ALEN];
+
+       dev = netdev_priv(net);
+       if (get_ether_addr(dev_addr, new_addr))
+               return -EINVAL;
+       memcpy(dev->dev_mac, new_addr, ETH_ALEN);
+       return 0;
+}
+EXPORT_SYMBOL_GPL(gether_set_dev_addr);
+
+int gether_get_dev_addr(struct net_device *net, char *dev_addr, int len)
+{
+       struct eth_dev *dev;
+
+       dev = netdev_priv(net);
+       return get_ether_addr_str(dev->dev_mac, dev_addr, len);
+}
+EXPORT_SYMBOL_GPL(gether_get_dev_addr);
+
+int gether_set_host_addr(struct net_device *net, const char *host_addr)
+{
+       struct eth_dev *dev;
+       u8 new_addr[ETH_ALEN];
+
+       dev = netdev_priv(net);
+       if (get_ether_addr(host_addr, new_addr))
+               return -EINVAL;
+       memcpy(dev->host_mac, new_addr, ETH_ALEN);
+       return 0;
+}
+EXPORT_SYMBOL_GPL(gether_set_host_addr);
+
+int gether_get_host_addr(struct net_device *net, char *host_addr, int len)
+{
+       struct eth_dev *dev;
+
+       dev = netdev_priv(net);
+       return get_ether_addr_str(dev->host_mac, host_addr, len);
+}
+EXPORT_SYMBOL_GPL(gether_get_host_addr);
+
+int gether_get_host_addr_cdc(struct net_device *net, char *host_addr, int len)
+{
+       struct eth_dev *dev;
+
+       if (len < 13)
+               return -EINVAL;
+
+       dev = netdev_priv(net);
+       snprintf(host_addr, len, "%pm", dev->host_mac);
+
+       return strlen(host_addr);
+}
+EXPORT_SYMBOL_GPL(gether_get_host_addr_cdc);
+
+void gether_get_host_addr_u8(struct net_device *net, u8 host_mac[ETH_ALEN])
+{
+       struct eth_dev *dev;
+
+       dev = netdev_priv(net);
+       memcpy(host_mac, dev->host_mac, ETH_ALEN);
+}
+EXPORT_SYMBOL_GPL(gether_get_host_addr_u8);
+
+void gether_set_qmult(struct net_device *net, unsigned qmult)
+{
+       struct eth_dev *dev;
+
+       dev = netdev_priv(net);
+       dev->qmult = qmult;
+}
+EXPORT_SYMBOL_GPL(gether_set_qmult);
+
+unsigned gether_get_qmult(struct net_device *net)
+{
+       struct eth_dev *dev;
+
+       dev = netdev_priv(net);
+       return dev->qmult;
+}
+EXPORT_SYMBOL_GPL(gether_get_qmult);
+
+int gether_get_ifname(struct net_device *net, char *name, int len)
+{
+       rtnl_lock();
+       strlcpy(name, netdev_name(net), len);
+       rtnl_unlock();
+       return strlen(name);
+}
+EXPORT_SYMBOL_GPL(gether_get_ifname);
+
+/**
+ * gether_cleanup - remove Ethernet-over-USB device
+ * Context: may sleep
+ *
+ * This is called to free all resources allocated by @gether_setup().
+ */
+void gether_cleanup(struct eth_dev *dev)
+{
+       if (!dev)
+               return;
+
+       unregister_netdev(dev->net);
+       flush_work(&dev->work);
+       free_netdev(dev->net);
+}
+EXPORT_SYMBOL_GPL(gether_cleanup);
+
+/**
+ * gether_connect - notify network layer that USB link is active
+ * @link: the USB link, set up with endpoints, descriptors matching
+ *     current device speed, and any framing wrapper(s) set up.
+ * Context: irqs blocked
+ *
+ * This is called to activate endpoints and let the network layer know
+ * the connection is active ("carrier detect").  It may cause the I/O
+ * queues to open and start letting network packets flow, but will in
+ * any case activate the endpoints so that they respond properly to the
+ * USB host.
+ *
+ * Verify net_device pointer returned using IS_ERR().  If it doesn't
+ * indicate some error code (negative errno), ep->driver_data values
+ * have been overwritten.
+ */
+struct net_device *gether_connect(struct gether *link)
+{
+       struct eth_dev          *dev = link->ioport;
+       int                     result = 0;
+
+       if (!dev)
+               return ERR_PTR(-EINVAL);
+
+       link->in_ep->driver_data = dev;
+       result = usb_ep_enable(link->in_ep);
+       if (result != 0) {
+               DBG(dev, "enable %s --> %d\n",
+                       link->in_ep->name, result);
+               goto fail0;
+       }
+
+       link->out_ep->driver_data = dev;
+       result = usb_ep_enable(link->out_ep);
+       if (result != 0) {
+               DBG(dev, "enable %s --> %d\n",
+                       link->out_ep->name, result);
+               goto fail1;
+       }
+
+       if (result == 0)
+               result = alloc_requests(dev, link, qlen(dev->gadget,
+                                       dev->qmult));
+
+       if (result == 0) {
+               dev->zlp = link->is_zlp_ok;
+               DBG(dev, "qlen %d\n", qlen(dev->gadget, dev->qmult));
+
+               dev->header_len = link->header_len;
+               dev->unwrap = link->unwrap;
+               dev->wrap = link->wrap;
+
+               spin_lock(&dev->lock);
+               dev->port_usb = link;
+               if (netif_running(dev->net)) {
+                       if (link->open)
+                               link->open(link);
+               } else {
+                       if (link->close)
+                               link->close(link);
+               }
+               spin_unlock(&dev->lock);
+
+               netif_carrier_on(dev->net);
+               if (netif_running(dev->net))
+                       eth_start(dev, GFP_ATOMIC);
+
+       /* on error, disable any endpoints  */
+       } else {
+               (void) usb_ep_disable(link->out_ep);
+fail1:
+               (void) usb_ep_disable(link->in_ep);
+       }
+fail0:
+       /* caller is responsible for cleanup on error */
+       if (result < 0)
+               return ERR_PTR(result);
+       return dev->net;
+}
+EXPORT_SYMBOL_GPL(gether_connect);
+
+/**
+ * gether_disconnect - notify network layer that USB link is inactive
+ * @link: the USB link, on which gether_connect() was called
+ * Context: irqs blocked
+ *
+ * This is called to deactivate endpoints and let the network layer know
+ * the connection went inactive ("no carrier").
+ *
+ * On return, the state is as if gether_connect() had never been called.
+ * The endpoints are inactive, and accordingly without active USB I/O.
+ * Pointers to endpoint descriptors and endpoint private data are nulled.
+ */
+void gether_disconnect(struct gether *link)
+{
+       struct eth_dev          *dev = link->ioport;
+       struct usb_request      *req;
+
+       WARN_ON(!dev);
+       if (!dev)
+               return;
+
+       DBG(dev, "%s\n", __func__);
+
+       netif_stop_queue(dev->net);
+       netif_carrier_off(dev->net);
+
+       /* disable endpoints, forcing (synchronous) completion
+        * of all pending i/o.  then free the request objects
+        * and forget about the endpoints.
+        */
+       usb_ep_disable(link->in_ep);
+       spin_lock(&dev->req_lock);
+       while (!list_empty(&dev->tx_reqs)) {
+               req = container_of(dev->tx_reqs.next,
+                                       struct usb_request, list);
+               list_del(&req->list);
+
+               spin_unlock(&dev->req_lock);
+               usb_ep_free_request(link->in_ep, req);
+               spin_lock(&dev->req_lock);
+       }
+       spin_unlock(&dev->req_lock);
+       link->in_ep->driver_data = NULL;
+       link->in_ep->desc = NULL;
+
+       usb_ep_disable(link->out_ep);
+       spin_lock(&dev->req_lock);
+       while (!list_empty(&dev->rx_reqs)) {
+               req = container_of(dev->rx_reqs.next,
+                                       struct usb_request, list);
+               list_del(&req->list);
+
+               spin_unlock(&dev->req_lock);
+               usb_ep_free_request(link->out_ep, req);
+               spin_lock(&dev->req_lock);
+       }
+       spin_unlock(&dev->req_lock);
+       link->out_ep->driver_data = NULL;
+       link->out_ep->desc = NULL;
+
+       /* finish forgetting about this USB link episode */
+       dev->header_len = 0;
+       dev->unwrap = NULL;
+       dev->wrap = NULL;
+
+       spin_lock(&dev->lock);
+       dev->port_usb = NULL;
+       spin_unlock(&dev->lock);
+}
+EXPORT_SYMBOL_GPL(gether_disconnect);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("David Brownell");