Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / caif / caif_virtio.c
diff --git a/kernel/drivers/net/caif/caif_virtio.c b/kernel/drivers/net/caif/caif_virtio.c
new file mode 100644 (file)
index 0000000..b306210
--- /dev/null
@@ -0,0 +1,789 @@
+/*
+ * Copyright (C) ST-Ericsson AB 2013
+ * Authors: Vicram Arv
+ *         Dmitry Tarnyagin <dmitry.tarnyagin@lockless.no>
+ *         Sjur Brendeland
+ * License terms: GNU General Public License (GPL) version 2
+ */
+#include <linux/module.h>
+#include <linux/if_arp.h>
+#include <linux/virtio.h>
+#include <linux/vringh.h>
+#include <linux/debugfs.h>
+#include <linux/spinlock.h>
+#include <linux/genalloc.h>
+#include <linux/interrupt.h>
+#include <linux/netdevice.h>
+#include <linux/rtnetlink.h>
+#include <linux/virtio_ids.h>
+#include <linux/virtio_caif.h>
+#include <linux/virtio_ring.h>
+#include <linux/dma-mapping.h>
+#include <net/caif/caif_dev.h>
+#include <linux/virtio_config.h>
+
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Vicram Arv");
+MODULE_AUTHOR("Sjur Brendeland");
+MODULE_DESCRIPTION("Virtio CAIF Driver");
+
+/* NAPI schedule quota */
+#define CFV_DEFAULT_QUOTA 32
+
+/* Defaults used if virtio config space is unavailable */
+#define CFV_DEF_MTU_SIZE 4096
+#define CFV_DEF_HEADROOM 32
+#define CFV_DEF_TAILROOM 32
+
+/* Required IP header alignment */
+#define IP_HDR_ALIGN 4
+
+/* struct cfv_napi_contxt - NAPI context info
+ * @riov: IOV holding data read from the ring. Note that riov may
+ *       still hold data when cfv_rx_poll() returns.
+ * @head: Last descriptor ID we received from vringh_getdesc_kern.
+ *       We use this to put descriptor back on the used ring. USHRT_MAX is
+ *       used to indicate invalid head-id.
+ */
+struct cfv_napi_context {
+       struct vringh_kiov riov;
+       unsigned short head;
+};
+
+/* struct cfv_stats - statistics for debugfs
+ * @rx_napi_complete:  Number of NAPI completions (RX)
+ * @rx_napi_resched:   Number of calls where the full quota was used (RX)
+ * @rx_nomem:          Number of SKB alloc failures (RX)
+ * @rx_kicks:          Number of RX kicks
+ * @tx_full_ring:      Number times TX ring was full
+ * @tx_no_mem:         Number of times TX went out of memory
+ * @tx_flow_on:                Number of flow on (TX)
+ * @tx_kicks:          Number of TX kicks
+ */
+struct cfv_stats {
+       u32 rx_napi_complete;
+       u32 rx_napi_resched;
+       u32 rx_nomem;
+       u32 rx_kicks;
+       u32 tx_full_ring;
+       u32 tx_no_mem;
+       u32 tx_flow_on;
+       u32 tx_kicks;
+};
+
+/* struct cfv_info - Caif Virtio control structure
+ * @cfdev:     caif common header
+ * @vdev:      Associated virtio device
+ * @vr_rx:     rx/downlink host vring
+ * @vq_tx:     tx/uplink virtqueue
+ * @ndev:      CAIF link layer device
+ * @watermark_tx: indicates number of free descriptors we need
+ *             to reopen the tx-queues after overload.
+ * @tx_lock:   protects vq_tx from concurrent use
+ * @tx_release_tasklet: Tasklet for freeing consumed TX buffers
+ * @napi:       Napi context used in cfv_rx_poll()
+ * @ctx:        Context data used in cfv_rx_poll()
+ * @tx_hr:     transmit headroom
+ * @rx_hr:     receive headroom
+ * @tx_tr:     transmit tail room
+ * @rx_tr:     receive tail room
+ * @mtu:       transmit max size
+ * @mru:       receive max size
+ * @allocsz:    size of dma memory reserved for TX buffers
+ * @alloc_addr: virtual address to dma memory for TX buffers
+ * @alloc_dma:  dma address to dma memory for TX buffers
+ * @genpool:    Gen Pool used for allocating TX buffers
+ * @reserved_mem: Pointer to memory reserve allocated from genpool
+ * @reserved_size: Size of memory reserve allocated from genpool
+ * @stats:       Statistics exposed in sysfs
+ * @debugfs:    Debugfs dentry for statistic counters
+ */
+struct cfv_info {
+       struct caif_dev_common cfdev;
+       struct virtio_device *vdev;
+       struct vringh *vr_rx;
+       struct virtqueue *vq_tx;
+       struct net_device *ndev;
+       unsigned int watermark_tx;
+       /* Protect access to vq_tx */
+       spinlock_t tx_lock;
+       struct tasklet_struct tx_release_tasklet;
+       struct napi_struct napi;
+       struct cfv_napi_context ctx;
+       u16 tx_hr;
+       u16 rx_hr;
+       u16 tx_tr;
+       u16 rx_tr;
+       u32 mtu;
+       u32 mru;
+       size_t allocsz;
+       void *alloc_addr;
+       dma_addr_t alloc_dma;
+       struct gen_pool *genpool;
+       unsigned long reserved_mem;
+       size_t reserved_size;
+       struct cfv_stats stats;
+       struct dentry *debugfs;
+};
+
+/* struct buf_info - maintains transmit buffer data handle
+ * @size:      size of transmit buffer
+ * @dma_handle: handle to allocated dma device memory area
+ * @vaddr:     virtual address mapping to allocated memory area
+ */
+struct buf_info {
+       size_t size;
+       u8 *vaddr;
+};
+
+/* Called from virtio device, in IRQ context */
+static void cfv_release_cb(struct virtqueue *vq_tx)
+{
+       struct cfv_info *cfv = vq_tx->vdev->priv;
+
+       ++cfv->stats.tx_kicks;
+       tasklet_schedule(&cfv->tx_release_tasklet);
+}
+
+static void free_buf_info(struct cfv_info *cfv, struct buf_info *buf_info)
+{
+       if (!buf_info)
+               return;
+       gen_pool_free(cfv->genpool, (unsigned long) buf_info->vaddr,
+                     buf_info->size);
+       kfree(buf_info);
+}
+
+/* This is invoked whenever the remote processor completed processing
+ * a TX msg we just sent, and the buffer is put back to the used ring.
+ */
+static void cfv_release_used_buf(struct virtqueue *vq_tx)
+{
+       struct cfv_info *cfv = vq_tx->vdev->priv;
+       unsigned long flags;
+
+       BUG_ON(vq_tx != cfv->vq_tx);
+
+       for (;;) {
+               unsigned int len;
+               struct buf_info *buf_info;
+
+               /* Get used buffer from used ring to recycle used descriptors */
+               spin_lock_irqsave(&cfv->tx_lock, flags);
+               buf_info = virtqueue_get_buf(vq_tx, &len);
+               spin_unlock_irqrestore(&cfv->tx_lock, flags);
+
+               /* Stop looping if there are no more buffers to free */
+               if (!buf_info)
+                       break;
+
+               free_buf_info(cfv, buf_info);
+
+               /* watermark_tx indicates if we previously stopped the tx
+                * queues. If we have enough free stots in the virtio ring,
+                * re-establish memory reserved and open up tx queues.
+                */
+               if (cfv->vq_tx->num_free <= cfv->watermark_tx)
+                       continue;
+
+               /* Re-establish memory reserve */
+               if (cfv->reserved_mem == 0 && cfv->genpool)
+                       cfv->reserved_mem =
+                               gen_pool_alloc(cfv->genpool,
+                                              cfv->reserved_size);
+
+               /* Open up the tx queues */
+               if (cfv->reserved_mem) {
+                       cfv->watermark_tx =
+                               virtqueue_get_vring_size(cfv->vq_tx);
+                       netif_tx_wake_all_queues(cfv->ndev);
+                       /* Buffers are recycled in cfv_netdev_tx, so
+                        * disable notifications when queues are opened.
+                        */
+                       virtqueue_disable_cb(cfv->vq_tx);
+                       ++cfv->stats.tx_flow_on;
+               } else {
+                       /* if no memory reserve, wait for more free slots */
+                       WARN_ON(cfv->watermark_tx >
+                              virtqueue_get_vring_size(cfv->vq_tx));
+                       cfv->watermark_tx +=
+                               virtqueue_get_vring_size(cfv->vq_tx) / 4;
+               }
+       }
+}
+
+/* Allocate a SKB and copy packet data to it */
+static struct sk_buff *cfv_alloc_and_copy_skb(int *err,
+                                             struct cfv_info *cfv,
+                                             u8 *frm, u32 frm_len)
+{
+       struct sk_buff *skb;
+       u32 cfpkt_len, pad_len;
+
+       *err = 0;
+       /* Verify that packet size with down-link header and mtu size */
+       if (frm_len > cfv->mru || frm_len <= cfv->rx_hr + cfv->rx_tr) {
+               netdev_err(cfv->ndev,
+                          "Invalid frmlen:%u  mtu:%u hr:%d tr:%d\n",
+                          frm_len, cfv->mru,  cfv->rx_hr,
+                          cfv->rx_tr);
+               *err = -EPROTO;
+               return NULL;
+       }
+
+       cfpkt_len = frm_len - (cfv->rx_hr + cfv->rx_tr);
+       pad_len = (unsigned long)(frm + cfv->rx_hr) & (IP_HDR_ALIGN - 1);
+
+       skb = netdev_alloc_skb(cfv->ndev, frm_len + pad_len);
+       if (!skb) {
+               *err = -ENOMEM;
+               return NULL;
+       }
+
+       skb_reserve(skb, cfv->rx_hr + pad_len);
+
+       memcpy(skb_put(skb, cfpkt_len), frm + cfv->rx_hr, cfpkt_len);
+       return skb;
+}
+
+/* Get packets from the host vring */
+static int cfv_rx_poll(struct napi_struct *napi, int quota)
+{
+       struct cfv_info *cfv = container_of(napi, struct cfv_info, napi);
+       int rxcnt = 0;
+       int err = 0;
+       void *buf;
+       struct sk_buff *skb;
+       struct vringh_kiov *riov = &cfv->ctx.riov;
+       unsigned int skb_len;
+
+       do {
+               skb = NULL;
+
+               /* Put the previous iovec back on the used ring and
+                * fetch a new iovec if we have processed all elements.
+                */
+               if (riov->i == riov->used) {
+                       if (cfv->ctx.head != USHRT_MAX) {
+                               vringh_complete_kern(cfv->vr_rx,
+                                                    cfv->ctx.head,
+                                                    0);
+                               cfv->ctx.head = USHRT_MAX;
+                       }
+
+                       err = vringh_getdesc_kern(
+                               cfv->vr_rx,
+                               riov,
+                               NULL,
+                               &cfv->ctx.head,
+                               GFP_ATOMIC);
+
+                       if (err <= 0)
+                               goto exit;
+               }
+
+               buf = phys_to_virt((unsigned long) riov->iov[riov->i].iov_base);
+               /* TODO: Add check on valid buffer address */
+
+               skb = cfv_alloc_and_copy_skb(&err, cfv, buf,
+                                            riov->iov[riov->i].iov_len);
+               if (unlikely(err))
+                       goto exit;
+
+               /* Push received packet up the stack. */
+               skb_len = skb->len;
+               skb->protocol = htons(ETH_P_CAIF);
+               skb_reset_mac_header(skb);
+               skb->dev = cfv->ndev;
+               err = netif_receive_skb(skb);
+               if (unlikely(err)) {
+                       ++cfv->ndev->stats.rx_dropped;
+               } else {
+                       ++cfv->ndev->stats.rx_packets;
+                       cfv->ndev->stats.rx_bytes += skb_len;
+               }
+
+               ++riov->i;
+               ++rxcnt;
+       } while (rxcnt < quota);
+
+       ++cfv->stats.rx_napi_resched;
+       goto out;
+
+exit:
+       switch (err) {
+       case 0:
+               ++cfv->stats.rx_napi_complete;
+
+               /* Really out of patckets? (stolen from virtio_net)*/
+               napi_complete(napi);
+               if (unlikely(!vringh_notify_enable_kern(cfv->vr_rx)) &&
+                   napi_schedule_prep(napi)) {
+                       vringh_notify_disable_kern(cfv->vr_rx);
+                       __napi_schedule(napi);
+               }
+               break;
+
+       case -ENOMEM:
+               ++cfv->stats.rx_nomem;
+               dev_kfree_skb(skb);
+               /* Stop NAPI poll on OOM, we hope to be polled later */
+               napi_complete(napi);
+               vringh_notify_enable_kern(cfv->vr_rx);
+               break;
+
+       default:
+               /* We're doomed, any modem fault is fatal */
+               netdev_warn(cfv->ndev, "Bad ring, disable device\n");
+               cfv->ndev->stats.rx_dropped = riov->used - riov->i;
+               napi_complete(napi);
+               vringh_notify_disable_kern(cfv->vr_rx);
+               netif_carrier_off(cfv->ndev);
+               break;
+       }
+out:
+       if (rxcnt && vringh_need_notify_kern(cfv->vr_rx) > 0)
+               vringh_notify(cfv->vr_rx);
+       return rxcnt;
+}
+
+static void cfv_recv(struct virtio_device *vdev, struct vringh *vr_rx)
+{
+       struct cfv_info *cfv = vdev->priv;
+
+       ++cfv->stats.rx_kicks;
+       vringh_notify_disable_kern(cfv->vr_rx);
+       napi_schedule(&cfv->napi);
+}
+
+static void cfv_destroy_genpool(struct cfv_info *cfv)
+{
+       if (cfv->alloc_addr)
+               dma_free_coherent(cfv->vdev->dev.parent->parent,
+                                 cfv->allocsz, cfv->alloc_addr,
+                                 cfv->alloc_dma);
+
+       if (!cfv->genpool)
+               return;
+       gen_pool_free(cfv->genpool,  cfv->reserved_mem,
+                     cfv->reserved_size);
+       gen_pool_destroy(cfv->genpool);
+       cfv->genpool = NULL;
+}
+
+static int cfv_create_genpool(struct cfv_info *cfv)
+{
+       int err;
+
+       /* dma_alloc can only allocate whole pages, and we need a more
+        * fine graned allocation so we use genpool. We ask for space needed
+        * by IP and a full ring. If the dma allcoation fails we retry with a
+        * smaller allocation size.
+        */
+       err = -ENOMEM;
+       cfv->allocsz = (virtqueue_get_vring_size(cfv->vq_tx) *
+                       (ETH_DATA_LEN + cfv->tx_hr + cfv->tx_tr) * 11)/10;
+       if (cfv->allocsz <= (num_possible_cpus() + 1) * cfv->ndev->mtu)
+               return -EINVAL;
+
+       for (;;) {
+               if (cfv->allocsz <= num_possible_cpus() * cfv->ndev->mtu) {
+                       netdev_info(cfv->ndev, "Not enough device memory\n");
+                       return -ENOMEM;
+               }
+
+               cfv->alloc_addr = dma_alloc_coherent(
+                                               cfv->vdev->dev.parent->parent,
+                                               cfv->allocsz, &cfv->alloc_dma,
+                                               GFP_ATOMIC);
+               if (cfv->alloc_addr)
+                       break;
+
+               cfv->allocsz = (cfv->allocsz * 3) >> 2;
+       }
+
+       netdev_dbg(cfv->ndev, "Allocated %zd bytes from dma-memory\n",
+                  cfv->allocsz);
+
+       /* Allocate on 128 bytes boundaries (1 << 7)*/
+       cfv->genpool = gen_pool_create(7, -1);
+       if (!cfv->genpool)
+               goto err;
+
+       err = gen_pool_add_virt(cfv->genpool, (unsigned long)cfv->alloc_addr,
+                               (phys_addr_t)virt_to_phys(cfv->alloc_addr),
+                               cfv->allocsz, -1);
+       if (err)
+               goto err;
+
+       /* Reserve some memory for low memory situations. If we hit the roof
+        * in the memory pool, we stop TX flow and release the reserve.
+        */
+       cfv->reserved_size = num_possible_cpus() * cfv->ndev->mtu;
+       cfv->reserved_mem = gen_pool_alloc(cfv->genpool,
+                                          cfv->reserved_size);
+       if (!cfv->reserved_mem) {
+               err = -ENOMEM;
+               goto err;
+       }
+
+       cfv->watermark_tx = virtqueue_get_vring_size(cfv->vq_tx);
+       return 0;
+err:
+       cfv_destroy_genpool(cfv);
+       return err;
+}
+
+/* Enable the CAIF interface and allocate the memory-pool */
+static int cfv_netdev_open(struct net_device *netdev)
+{
+       struct cfv_info *cfv = netdev_priv(netdev);
+
+       if (cfv_create_genpool(cfv))
+               return -ENOMEM;
+
+       netif_carrier_on(netdev);
+       napi_enable(&cfv->napi);
+
+       /* Schedule NAPI to read any pending packets */
+       napi_schedule(&cfv->napi);
+       return 0;
+}
+
+/* Disable the CAIF interface and free the memory-pool */
+static int cfv_netdev_close(struct net_device *netdev)
+{
+       struct cfv_info *cfv = netdev_priv(netdev);
+       unsigned long flags;
+       struct buf_info *buf_info;
+
+       /* Disable interrupts, queues and NAPI polling */
+       netif_carrier_off(netdev);
+       virtqueue_disable_cb(cfv->vq_tx);
+       vringh_notify_disable_kern(cfv->vr_rx);
+       napi_disable(&cfv->napi);
+
+       /* Release any TX buffers on both used and avilable rings */
+       cfv_release_used_buf(cfv->vq_tx);
+       spin_lock_irqsave(&cfv->tx_lock, flags);
+       while ((buf_info = virtqueue_detach_unused_buf(cfv->vq_tx)))
+               free_buf_info(cfv, buf_info);
+       spin_unlock_irqrestore(&cfv->tx_lock, flags);
+
+       /* Release all dma allocated memory and destroy the pool */
+       cfv_destroy_genpool(cfv);
+       return 0;
+}
+
+/* Allocate a buffer in dma-memory and copy skb to it */
+static struct buf_info *cfv_alloc_and_copy_to_shm(struct cfv_info *cfv,
+                                                      struct sk_buff *skb,
+                                                      struct scatterlist *sg)
+{
+       struct caif_payload_info *info = (void *)&skb->cb;
+       struct buf_info *buf_info = NULL;
+       u8 pad_len, hdr_ofs;
+
+       if (!cfv->genpool)
+               goto err;
+
+       if (unlikely(cfv->tx_hr + skb->len + cfv->tx_tr > cfv->mtu)) {
+               netdev_warn(cfv->ndev, "Invalid packet len (%d > %d)\n",
+                           cfv->tx_hr + skb->len + cfv->tx_tr, cfv->mtu);
+               goto err;
+       }
+
+       buf_info = kmalloc(sizeof(struct buf_info), GFP_ATOMIC);
+       if (unlikely(!buf_info))
+               goto err;
+
+       /* Make the IP header aligned in tbe buffer */
+       hdr_ofs = cfv->tx_hr + info->hdr_len;
+       pad_len = hdr_ofs & (IP_HDR_ALIGN - 1);
+       buf_info->size = cfv->tx_hr + skb->len + cfv->tx_tr + pad_len;
+
+       /* allocate dma memory buffer */
+       buf_info->vaddr = (void *)gen_pool_alloc(cfv->genpool, buf_info->size);
+       if (unlikely(!buf_info->vaddr))
+               goto err;
+
+       /* copy skbuf contents to send buffer */
+       skb_copy_bits(skb, 0, buf_info->vaddr + cfv->tx_hr + pad_len, skb->len);
+       sg_init_one(sg, buf_info->vaddr + pad_len,
+                   skb->len + cfv->tx_hr + cfv->rx_hr);
+
+       return buf_info;
+err:
+       kfree(buf_info);
+       return NULL;
+}
+
+/* Put the CAIF packet on the virtio ring and kick the receiver */
+static int cfv_netdev_tx(struct sk_buff *skb, struct net_device *netdev)
+{
+       struct cfv_info *cfv = netdev_priv(netdev);
+       struct buf_info *buf_info;
+       struct scatterlist sg;
+       unsigned long flags;
+       bool flow_off = false;
+       int ret;
+
+       /* garbage collect released buffers */
+       cfv_release_used_buf(cfv->vq_tx);
+       spin_lock_irqsave(&cfv->tx_lock, flags);
+
+       /* Flow-off check takes into account number of cpus to make sure
+        * virtqueue will not be overfilled in any possible smp conditions.
+        *
+        * Flow-on is triggered when sufficient buffers are freed
+        */
+       if (unlikely(cfv->vq_tx->num_free <= num_present_cpus())) {
+               flow_off = true;
+               cfv->stats.tx_full_ring++;
+       }
+
+       /* If we run out of memory, we release the memory reserve and retry
+        * allocation.
+        */
+       buf_info = cfv_alloc_and_copy_to_shm(cfv, skb, &sg);
+       if (unlikely(!buf_info)) {
+               cfv->stats.tx_no_mem++;
+               flow_off = true;
+
+               if (cfv->reserved_mem && cfv->genpool) {
+                       gen_pool_free(cfv->genpool,  cfv->reserved_mem,
+                                     cfv->reserved_size);
+                       cfv->reserved_mem = 0;
+                       buf_info = cfv_alloc_and_copy_to_shm(cfv, skb, &sg);
+               }
+       }
+
+       if (unlikely(flow_off)) {
+               /* Turn flow on when a 1/4 of the descriptors are released */
+               cfv->watermark_tx = virtqueue_get_vring_size(cfv->vq_tx) / 4;
+               /* Enable notifications of recycled TX buffers */
+               virtqueue_enable_cb(cfv->vq_tx);
+               netif_tx_stop_all_queues(netdev);
+       }
+
+       if (unlikely(!buf_info)) {
+               /* If the memory reserve does it's job, this shouldn't happen */
+               netdev_warn(cfv->ndev, "Out of gen_pool memory\n");
+               goto err;
+       }
+
+       ret = virtqueue_add_outbuf(cfv->vq_tx, &sg, 1, buf_info, GFP_ATOMIC);
+       if (unlikely((ret < 0))) {
+               /* If flow control works, this shouldn't happen */
+               netdev_warn(cfv->ndev, "Failed adding buffer to TX vring:%d\n",
+                           ret);
+               goto err;
+       }
+
+       /* update netdev statistics */
+       cfv->ndev->stats.tx_packets++;
+       cfv->ndev->stats.tx_bytes += skb->len;
+       spin_unlock_irqrestore(&cfv->tx_lock, flags);
+
+       /* tell the remote processor it has a pending message to read */
+       virtqueue_kick(cfv->vq_tx);
+
+       dev_kfree_skb(skb);
+       return NETDEV_TX_OK;
+err:
+       spin_unlock_irqrestore(&cfv->tx_lock, flags);
+       cfv->ndev->stats.tx_dropped++;
+       free_buf_info(cfv, buf_info);
+       dev_kfree_skb(skb);
+       return NETDEV_TX_OK;
+}
+
+static void cfv_tx_release_tasklet(unsigned long drv)
+{
+       struct cfv_info *cfv = (struct cfv_info *)drv;
+       cfv_release_used_buf(cfv->vq_tx);
+}
+
+static const struct net_device_ops cfv_netdev_ops = {
+       .ndo_open = cfv_netdev_open,
+       .ndo_stop = cfv_netdev_close,
+       .ndo_start_xmit = cfv_netdev_tx,
+};
+
+static void cfv_netdev_setup(struct net_device *netdev)
+{
+       netdev->netdev_ops = &cfv_netdev_ops;
+       netdev->type = ARPHRD_CAIF;
+       netdev->tx_queue_len = 100;
+       netdev->flags = IFF_POINTOPOINT | IFF_NOARP;
+       netdev->mtu = CFV_DEF_MTU_SIZE;
+       netdev->destructor = free_netdev;
+}
+
+/* Create debugfs counters for the device */
+static inline void debugfs_init(struct cfv_info *cfv)
+{
+       cfv->debugfs =
+               debugfs_create_dir(netdev_name(cfv->ndev), NULL);
+
+       if (IS_ERR(cfv->debugfs))
+               return;
+
+       debugfs_create_u32("rx-napi-complete", S_IRUSR, cfv->debugfs,
+                          &cfv->stats.rx_napi_complete);
+       debugfs_create_u32("rx-napi-resched", S_IRUSR, cfv->debugfs,
+                          &cfv->stats.rx_napi_resched);
+       debugfs_create_u32("rx-nomem", S_IRUSR, cfv->debugfs,
+                          &cfv->stats.rx_nomem);
+       debugfs_create_u32("rx-kicks", S_IRUSR, cfv->debugfs,
+                          &cfv->stats.rx_kicks);
+       debugfs_create_u32("tx-full-ring", S_IRUSR, cfv->debugfs,
+                          &cfv->stats.tx_full_ring);
+       debugfs_create_u32("tx-no-mem", S_IRUSR, cfv->debugfs,
+                          &cfv->stats.tx_no_mem);
+       debugfs_create_u32("tx-kicks", S_IRUSR, cfv->debugfs,
+                          &cfv->stats.tx_kicks);
+       debugfs_create_u32("tx-flow-on", S_IRUSR, cfv->debugfs,
+                          &cfv->stats.tx_flow_on);
+}
+
+/* Setup CAIF for the a virtio device */
+static int cfv_probe(struct virtio_device *vdev)
+{
+       vq_callback_t *vq_cbs = cfv_release_cb;
+       vrh_callback_t *vrh_cbs = cfv_recv;
+       const char *names =  "output";
+       const char *cfv_netdev_name = "cfvrt";
+       struct net_device *netdev;
+       struct cfv_info *cfv;
+       int err = -EINVAL;
+
+       netdev = alloc_netdev(sizeof(struct cfv_info), cfv_netdev_name,
+                             NET_NAME_UNKNOWN, cfv_netdev_setup);
+       if (!netdev)
+               return -ENOMEM;
+
+       cfv = netdev_priv(netdev);
+       cfv->vdev = vdev;
+       cfv->ndev = netdev;
+
+       spin_lock_init(&cfv->tx_lock);
+
+       /* Get the RX virtio ring. This is a "host side vring". */
+       err = -ENODEV;
+       if (!vdev->vringh_config || !vdev->vringh_config->find_vrhs)
+               goto err;
+
+       err = vdev->vringh_config->find_vrhs(vdev, 1, &cfv->vr_rx, &vrh_cbs);
+       if (err)
+               goto err;
+
+       /* Get the TX virtio ring. This is a "guest side vring". */
+       err = vdev->config->find_vqs(vdev, 1, &cfv->vq_tx, &vq_cbs, &names);
+       if (err)
+               goto err;
+
+       /* Get the CAIF configuration from virtio config space, if available */
+       if (vdev->config->get) {
+               virtio_cread(vdev, struct virtio_caif_transf_config, headroom,
+                            &cfv->tx_hr);
+               virtio_cread(vdev, struct virtio_caif_transf_config, headroom,
+                            &cfv->rx_hr);
+               virtio_cread(vdev, struct virtio_caif_transf_config, tailroom,
+                            &cfv->tx_tr);
+               virtio_cread(vdev, struct virtio_caif_transf_config, tailroom,
+                            &cfv->rx_tr);
+               virtio_cread(vdev, struct virtio_caif_transf_config, mtu,
+                            &cfv->mtu);
+               virtio_cread(vdev, struct virtio_caif_transf_config, mtu,
+                            &cfv->mru);
+       } else {
+               cfv->tx_hr = CFV_DEF_HEADROOM;
+               cfv->rx_hr = CFV_DEF_HEADROOM;
+               cfv->tx_tr = CFV_DEF_TAILROOM;
+               cfv->rx_tr = CFV_DEF_TAILROOM;
+               cfv->mtu = CFV_DEF_MTU_SIZE;
+               cfv->mru = CFV_DEF_MTU_SIZE;
+       }
+
+       netdev->needed_headroom = cfv->tx_hr;
+       netdev->needed_tailroom = cfv->tx_tr;
+
+       /* Disable buffer release interrupts unless we have stopped TX queues */
+       virtqueue_disable_cb(cfv->vq_tx);
+
+       netdev->mtu = cfv->mtu - cfv->tx_tr;
+       vdev->priv = cfv;
+
+       /* Initialize NAPI poll context data */
+       vringh_kiov_init(&cfv->ctx.riov, NULL, 0);
+       cfv->ctx.head = USHRT_MAX;
+       netif_napi_add(netdev, &cfv->napi, cfv_rx_poll, CFV_DEFAULT_QUOTA);
+
+       tasklet_init(&cfv->tx_release_tasklet,
+                    cfv_tx_release_tasklet,
+                    (unsigned long)cfv);
+
+       /* Carrier is off until netdevice is opened */
+       netif_carrier_off(netdev);
+
+       /* register Netdev */
+       err = register_netdev(netdev);
+       if (err) {
+               dev_err(&vdev->dev, "Unable to register netdev (%d)\n", err);
+               goto err;
+       }
+
+       debugfs_init(cfv);
+
+       return 0;
+err:
+       netdev_warn(cfv->ndev, "CAIF Virtio probe failed:%d\n", err);
+
+       if (cfv->vr_rx)
+               vdev->vringh_config->del_vrhs(cfv->vdev);
+       if (cfv->vdev)
+               vdev->config->del_vqs(cfv->vdev);
+       free_netdev(netdev);
+       return err;
+}
+
+static void cfv_remove(struct virtio_device *vdev)
+{
+       struct cfv_info *cfv = vdev->priv;
+
+       rtnl_lock();
+       dev_close(cfv->ndev);
+       rtnl_unlock();
+
+       tasklet_kill(&cfv->tx_release_tasklet);
+       debugfs_remove_recursive(cfv->debugfs);
+
+       vringh_kiov_cleanup(&cfv->ctx.riov);
+       vdev->config->reset(vdev);
+       vdev->vringh_config->del_vrhs(cfv->vdev);
+       cfv->vr_rx = NULL;
+       vdev->config->del_vqs(cfv->vdev);
+       unregister_netdev(cfv->ndev);
+}
+
+static struct virtio_device_id id_table[] = {
+       { VIRTIO_ID_CAIF, VIRTIO_DEV_ANY_ID },
+       { 0 },
+};
+
+static unsigned int features[] = {
+};
+
+static struct virtio_driver caif_virtio_driver = {
+       .feature_table          = features,
+       .feature_table_size     = ARRAY_SIZE(features),
+       .driver.name            = KBUILD_MODNAME,
+       .driver.owner           = THIS_MODULE,
+       .id_table               = id_table,
+       .probe                  = cfv_probe,
+       .remove                 = cfv_remove,
+};
+
+module_virtio_driver(caif_virtio_driver);
+MODULE_DEVICE_TABLE(virtio, id_table);