[virtio] Replace virtio-net with native iPXE driver

This patch adds a native iPXE virtio-net driver and removes the legacy
Etherboot virtio-net driver.  The main reasons for doing this are:

1. Multiple virtio-net NICs are now supported by iPXE.  The legacy
   driver kept global state and caused issues in virtual machines with
   more than one virtio-net device.

2. Faster downloads.  The native iPXE driver downloads 100 MB over
   HTTP in 12s, the legacy Etherboot driver in 37s.  This simple
   benchmark uses KVM with tap networking and the Python
   SimpleHTTPServer both running on the same host.

Changes to core virtio code reduce vring descriptors to 256 (QEMU uses
128 for virtio-blk and 256 for virtio-net) and change the opaque token
from u16 to void*.  Lowering the descriptor count reduces memory
consumption.  The void* opaque token change makes driver code simpler.

Signed-off-by: Stefan Hajnoczi <stefanha@gmail.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>

src/drivers/bus/virtio-ring.c

@@ -57,12 +57,12 @@ void vring_detach(struct vring_virtqueue *vq, unsigned int head)
  *
  */
 
-int vring_get_buf(struct vring_virtqueue *vq, unsigned int *len)
+void *vring_get_buf(struct vring_virtqueue *vq, unsigned int *len)
 {
    struct vring *vr = &vq->vring;
    struct vring_used_elem *elem;
    u32 id;
-   int ret;
+   void *opaque;
 
    BUG_ON(!vring_more_used(vq));
 
@@ -72,19 +72,19 @@ int vring_get_buf(struct vring_virtqueue *vq, unsigned int *len)
    if (len != NULL)
            *len = elem->len;
 
-   ret = vq->vdata[id];
+   opaque = vq->vdata[id];
 
    vring_detach(vq, id);
 
    vq->last_used_idx++;
 
-   return ret;
+   return opaque;
 }
 
 void vring_add_buf(struct vring_virtqueue *vq,
 		   struct vring_list list[],
 		   unsigned int out, unsigned int in,
-		   int index, int num_added)
+		   void *opaque, int num_added)
 {
    struct vring *vr = &vq->vring;
    int i, avail, head, prev;
@@ -113,7 +113,7 @@ void vring_add_buf(struct vring_virtqueue *vq,
 
    vq->free_head = i;
 
-   vq->vdata[head] = index;
+   vq->vdata[head] = opaque;
 
    avail = (vr->avail->idx + num_added) % vr->num;
    vr->avail->ring[avail] = head;

src/drivers/net/virtio-net.c

@@ -1,6 +1,9 @@
-/* virtio-net.c - etherboot driver for virtio network interface
+/*
+ * (c) Copyright 2010 Stefan Hajnoczi <stefanha@gmail.com>
+ *
+ * based on the Etherboot virtio-net driver
  *
- * (c) Copyright 2008 Bull S.A.S.
+ *  (c) Copyright 2008 Bull S.A.S.
  *
  *  Author: Laurent Vivier <Laurent.Vivier@bull.net>
  *
@@ -15,293 +18,397 @@
  *
  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  * See the COPYING file in the top-level directory.
- *
- *
  */
 
-#include "etherboot.h"
-#include "nic.h"
-#include "ipxe/virtio-ring.h"
-#include "ipxe/virtio-pci.h"
+FILE_LICENCE ( GPL2_OR_LATER );
+
+#include <errno.h>
+#include <stdlib.h>
+#include <ipxe/list.h>
+#include <ipxe/iobuf.h>
+#include <ipxe/netdevice.h>
+#include <ipxe/pci.h>
+#include <ipxe/if_ether.h>
+#include <ipxe/ethernet.h>
+#include <ipxe/virtio-ring.h>
+#include <ipxe/virtio-pci.h>
 #include "virtio-net.h"
 
-#define BUG() do { \
-   printf("BUG: failure at %s:%d/%s()!\n", \
-          __FILE__, __LINE__, __FUNCTION__); \
-   while(1); \
-} while (0)
-#define BUG_ON(condition) do { if (condition) BUG(); } while (0)
+/*
+ * Virtio network device driver
+ *
+ * Specification:
+ * http://ozlabs.org/~rusty/virtio-spec/
+ *
+ * The virtio network device is supported by Linux virtualization software
+ * including QEMU/KVM and lguest.  This driver supports the virtio over PCI
+ * transport; virtual machines have one virtio-net PCI adapter per NIC.
+ *
+ * Virtio-net is different from hardware NICs because virtio devices
+ * communicate with the hypervisor via virtqueues, not traditional descriptor
+ * rings.  Virtqueues are unordered queues, they support add_buf() and
+ * get_buf() operations.  To transmit a packet, the driver has to add the
+ * packet buffer onto the virtqueue.  To receive a packet, the driver must
+ * first add an empty buffer to the virtqueue and then get the filled packet
+ * buffer on completion.
+ *
+ * Virtqueues are an abstraction that is commonly implemented using the vring
+ * descriptor ring layout.  The vring is the actual shared memory structure
+ * that allows the virtual machine to communicate buffers with the hypervisor.
+ * Because the vring layout is optimized for flexibility and performance rather
+ * than space, it is heavy-weight and allocated like traditional descriptor
+ * rings in the open() function of the driver and not in probe().
+ *
+ * There is no true interrupt enable/disable.  Virtqueues have callback
+ * enable/disable flags but these are only hints.  The hypervisor may still
+ * raise an interrupt.  Nevertheless, this driver disables callbacks in the
+ * hopes of avoiding interrupts.
+ */
 
-/* Ethernet header */
+/* Driver types are declared here so virtio-net.h can be easily synced with its
+ * Linux source.
+ */
 
-struct eth_hdr {
-   unsigned char dst_addr[ETH_ALEN];
-   unsigned char src_addr[ETH_ALEN];
-   unsigned short type;
+/* Virtqueue indicies */
+enum {
+	RX_INDEX = 0,
+	TX_INDEX,
+	QUEUE_NB
 };
 
-struct eth_frame {
-   struct eth_hdr hdr;
-   unsigned char data[ETH_FRAME_LEN];
-};
+enum {
+	/** Max number of pending rx packets */
+	NUM_RX_BUF = 8,
 
-/* TX: virtio header and eth buffer */
+	/** Max Ethernet frame length, including FCS and VLAN tag */
+	RX_BUF_SIZE = 1522,
+};
 
-static struct virtio_net_hdr tx_virtio_hdr;
-static struct eth_frame tx_eth_frame;
+struct virtnet_nic {
+	/** Base pio register address */
+	unsigned long ioaddr;
 
-/* RX: virtio headers and buffers */
+	/** RX/TX virtqueues */
+	struct vring_virtqueue *virtqueue;
 
-#define RX_BUF_NB  6
-static struct virtio_net_hdr rx_hdr[RX_BUF_NB];
-static unsigned char rx_buffer[RX_BUF_NB][ETH_FRAME_LEN];
+	/** RX packets handed to the NIC waiting to be filled in */
+	struct list_head rx_iobufs;
 
-/* virtio queues and vrings */
+	/** Pending rx packet count */
+	unsigned int rx_num_iobufs;
 
-enum {
-   RX_INDEX = 0,
-   TX_INDEX,
-   QUEUE_NB
+	/** Virtio net packet header, we only need one */
+	struct virtio_net_hdr empty_header;
 };
 
-static struct vring_virtqueue virtqueue[QUEUE_NB];
-
-/*
- * virtnet_disable
+/** Add an iobuf to a virtqueue
  *
- * Turn off ethernet interface
+ * @v netdev		Network device
+ * @v vq_idx		Virtqueue index (RX_INDEX or TX_INDEX)
+ * @v iobuf		I/O buffer
  *
+ * The virtqueue is kicked after the iobuf has been added.
  */
-
-static void virtnet_disable(struct nic *nic)
-{
-   int i;
-
-   for (i = 0; i < QUEUE_NB; i++) {
-           vring_disable_cb(&virtqueue[i]);
-           vp_del_vq(nic->ioaddr, i);
-   }
-   vp_reset(nic->ioaddr);
+static void virtnet_enqueue_iob ( struct net_device *netdev,
+				  int vq_idx, struct io_buffer *iobuf ) {
+	struct virtnet_nic *virtnet = netdev->priv;
+	struct vring_virtqueue *vq = &virtnet->virtqueue[vq_idx];
+	unsigned int out = ( vq_idx == TX_INDEX ) ? 2 : 0;
+	unsigned int in = ( vq_idx == TX_INDEX ) ? 0 : 2;
+	struct vring_list list[] = {
+		{
+			/* Share a single zeroed virtio net header between all
+			 * rx and tx packets.  This works because this driver
+			 * does not use any advanced features so none of the
+			 * header fields get used.
+			 */
+			.addr = ( char* ) &virtnet->empty_header,
+			.length = sizeof ( virtnet->empty_header ),
+		},
+		{
+			.addr = ( char* ) iobuf->data,
+			.length = iob_len ( iobuf ),
+		},
+	};
+
+	DBGC ( virtnet, "VIRTIO-NET %p enqueuing iobuf %p on vq %d\n",
+	       virtnet, iobuf, vq_idx );
+
+	vring_add_buf ( vq, list, out, in, iobuf, 0 );
+	vring_kick ( virtnet->ioaddr, vq, 1 );
 }
 
-/*
- * virtnet_poll
- *
- * Wait for a frame
- *
- * return true if there is a packet ready to read
- *
- * nic->packet should contain data on return
- * nic->packetlen should contain length of data
+/** Try to keep rx virtqueue filled with iobufs
  *
+ * @v netdev		Network device
  */
-static int virtnet_poll(struct nic *nic, int retrieve)
-{
-   unsigned int len;
-   u16 token;
-   struct virtio_net_hdr *hdr;
-   struct vring_list list[2];
-
-   if (!vring_more_used(&virtqueue[RX_INDEX]))
-           return 0;
-
-   if (!retrieve)
-           return 1;
-
-   token = vring_get_buf(&virtqueue[RX_INDEX], &len);
-
-   BUG_ON(len > sizeof(struct virtio_net_hdr) + ETH_FRAME_LEN);
-
-   hdr = &rx_hdr[token];   /* FIXME: check flags */
-   len -= sizeof(struct virtio_net_hdr);
+static void virtnet_refill_rx_virtqueue ( struct net_device *netdev ) {
+	struct virtnet_nic *virtnet = netdev->priv;
 
-   nic->packetlen = len;
-   memcpy(nic->packet, (char *)rx_buffer[token], nic->packetlen);
+	while ( virtnet->rx_num_iobufs < NUM_RX_BUF ) {
+		struct io_buffer *iobuf;
 
-   /* add buffer to desc */
+		/* Try to allocate a buffer, stop for now if out of memory */
+		iobuf = alloc_iob ( RX_BUF_SIZE );
+		if ( ! iobuf )
+			break;
 
-   list[0].addr = (char*)&rx_hdr[token];
-   list[0].length = sizeof(struct virtio_net_hdr);
-   list[1].addr = (char*)&rx_buffer[token];
-   list[1].length = ETH_FRAME_LEN;
+		/* Keep track of iobuf so close() can free it */
+		list_add ( &iobuf->list, &virtnet->rx_iobufs );
 
-   vring_add_buf(&virtqueue[RX_INDEX], list, 0, 2, token, 0);
-   vring_kick(nic->ioaddr, &virtqueue[RX_INDEX], 1);
+		/* Mark packet length until we know the actual size */
+		iob_put ( iobuf, RX_BUF_SIZE );
 
-   return 1;
+		virtnet_enqueue_iob ( netdev, RX_INDEX, iobuf );
+		virtnet->rx_num_iobufs++;
+	}
 }
 
-/*
- *
- * virtnet_transmit
- *
- * Transmit a frame
+/** Open network device
  *
+ * @v netdev	Network device
+ * @ret rc	Return status code
  */
+static int virtnet_open ( struct net_device *netdev ) {
+	struct virtnet_nic *virtnet = netdev->priv;
+	unsigned long ioaddr = virtnet->ioaddr;
+	u32 features;
+	int i;
+
+	/* Reset for sanity */
+	vp_reset ( ioaddr );
+
+	/* Allocate virtqueues */
+	virtnet->virtqueue = zalloc ( QUEUE_NB *
+				      sizeof ( *virtnet->virtqueue ) );
+	if ( ! virtnet->virtqueue )
+		return -ENOMEM;
+
+	/* Initialize rx/tx virtqueues */
+	for ( i = 0; i < QUEUE_NB; i++ ) {
+		if ( vp_find_vq ( ioaddr, i, &virtnet->virtqueue[i] ) == -1 ) {
+			DBGC ( virtnet, "VIRTIO-NET %p cannot register queue %d\n",
+			       virtnet, i );
+			free ( virtnet->virtqueue );
+			virtnet->virtqueue = NULL;
+			return -ENOENT;
+		}
+	}
+
+	/* Initialize rx packets */
+	INIT_LIST_HEAD ( &virtnet->rx_iobufs );
+	virtnet->rx_num_iobufs = 0;
+	virtnet_refill_rx_virtqueue ( netdev );
+
+	/* Disable interrupts before starting */
+	netdev_irq ( netdev, 0 );
+
+	/* Driver is ready */
+	features = vp_get_features ( ioaddr );
+	vp_set_features ( ioaddr, features & ( 1 << VIRTIO_NET_F_MAC ) );
+	vp_set_status ( ioaddr, VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK );
+	return 0;
+}
 
-static void virtnet_transmit(struct nic *nic, const char *destaddr,
-        unsigned int type, unsigned int len, const char *data)
-{
-   struct vring_list list[2];
-
-   /*
-    * from http://www.etherboot.org/wiki/dev/devmanual :
-    *     "You do not need more than one transmit buffer."
-    */
+/** Close network device
+ *
+ * @v netdev	Network device
+ */
+static void virtnet_close ( struct net_device *netdev ) {
+	struct virtnet_nic *virtnet = netdev->priv;
+	struct io_buffer *iobuf;
+	struct io_buffer *next_iobuf;
+
+	vp_reset ( virtnet->ioaddr );
+
+	/* Virtqueues can be freed now that NIC is reset */
+	free ( virtnet->virtqueue );
+	virtnet->virtqueue = NULL;
+
+	/* Free rx iobufs */
+	list_for_each_entry_safe ( iobuf, next_iobuf, &virtnet->rx_iobufs, list ) {
+		free_iob ( iobuf );
+	}
+	INIT_LIST_HEAD ( &virtnet->rx_iobufs );
+	virtnet->rx_num_iobufs = 0;
+}
 
-   /* FIXME: initialize header according to vp_get_features() */
+/** Transmit packet
+ *
+ * @v netdev	Network device
+ * @v iobuf	I/O buffer
+ * @ret rc	Return status code
+ */
+static int virtnet_transmit ( struct net_device *netdev,
+			      struct io_buffer *iobuf ) {
+	virtnet_enqueue_iob ( netdev, TX_INDEX, iobuf );
+	return 0;
+}
 
-   tx_virtio_hdr.flags = 0;
-   tx_virtio_hdr.csum_offset = 0;
-   tx_virtio_hdr.csum_start = 0;
-   tx_virtio_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
-   tx_virtio_hdr.gso_size = 0;
-   tx_virtio_hdr.hdr_len = 0;
+/** Complete packet transmission
+ *
+ * @v netdev	Network device
+ */
+static void virtnet_process_tx_packets ( struct net_device *netdev ) {
+	struct virtnet_nic *virtnet = netdev->priv;
+	struct vring_virtqueue *tx_vq = &virtnet->virtqueue[TX_INDEX];
 
-   /* add ethernet frame into vring */
+	while ( vring_more_used ( tx_vq ) ) {
+		struct io_buffer *iobuf = vring_get_buf ( tx_vq, NULL );
 
-   BUG_ON(len > sizeof(tx_eth_frame.data));
+		DBGC ( virtnet, "VIRTIO-NET %p tx complete iobuf %p\n",
+		       virtnet, iobuf );
 
-   memcpy(tx_eth_frame.hdr.dst_addr, destaddr, ETH_ALEN);
-   memcpy(tx_eth_frame.hdr.src_addr, nic->node_addr, ETH_ALEN);
-   tx_eth_frame.hdr.type = htons(type);
-   memcpy(tx_eth_frame.data, data, len);
+		netdev_tx_complete ( netdev, iobuf );
+	}
+}
 
-   list[0].addr = (char*)&tx_virtio_hdr;
-   list[0].length = sizeof(struct virtio_net_hdr);
-   list[1].addr = (char*)&tx_eth_frame;
-   list[1].length = ETH_FRAME_LEN;
+/** Complete packet reception
+ *
+ * @v netdev	Network device
+ */
+static void virtnet_process_rx_packets ( struct net_device *netdev ) {
+	struct virtnet_nic *virtnet = netdev->priv;
+	struct vring_virtqueue *rx_vq = &virtnet->virtqueue[RX_INDEX];
 
-   vring_add_buf(&virtqueue[TX_INDEX], list, 2, 0, 0, 0);
+	while ( vring_more_used ( rx_vq ) ) {
+		unsigned int len;
+		struct io_buffer *iobuf = vring_get_buf ( rx_vq, &len );
 
-   vring_kick(nic->ioaddr, &virtqueue[TX_INDEX], 1);
+		/* Release ownership of iobuf */
+		list_del ( &iobuf->list );
+		virtnet->rx_num_iobufs--;
 
-   /*
-    * http://www.etherboot.org/wiki/dev/devmanual
-    *
-    *   "You should ensure the packet is fully transmitted
-    *    before returning from this routine"
-    */
+		/* Update iobuf length */
+		iob_unput ( iobuf, RX_BUF_SIZE );
+		iob_put ( iobuf, len - sizeof ( struct virtio_net_hdr ) );
 
-   while (!vring_more_used(&virtqueue[TX_INDEX])) {
-           mb();
-           udelay(10);
-   }
+		DBGC ( virtnet, "VIRTIO-NET %p rx complete iobuf %p len %zd\n",
+		       virtnet, iobuf, iob_len ( iobuf ) );
 
-   /* free desc */
+		/* Pass completed packet to the network stack */
+		netdev_rx ( netdev, iobuf );
+	}
 
-   (void)vring_get_buf(&virtqueue[TX_INDEX], NULL);
+	virtnet_refill_rx_virtqueue ( netdev );
 }
 
-static void virtnet_irq(struct nic *nic __unused, irq_action_t action)
-{
-   switch ( action ) {
-   case DISABLE :
-           vring_disable_cb(&virtqueue[RX_INDEX]);
-           vring_disable_cb(&virtqueue[TX_INDEX]);
-           break;
-   case ENABLE :
-           vring_enable_cb(&virtqueue[RX_INDEX]);
-           vring_enable_cb(&virtqueue[TX_INDEX]);
-           break;
-   case FORCE :
-           break;
-   }
+/** Poll for completed and received packets
+ *
+ * @v netdev	Network device
+ */
+static void virtnet_poll ( struct net_device *netdev ) {
+	struct virtnet_nic *virtnet = netdev->priv;
+
+	/* Acknowledge interrupt.  This is necessary for UNDI operation and
+	 * interrupts that are raised despite VRING_AVAIL_F_NO_INTERRUPT being
+	 * set (that flag is just a hint and the hypervisor not not have to
+	 * honor it).
+	 */
+	vp_get_isr ( virtnet->ioaddr );
+
+	virtnet_process_tx_packets ( netdev );
+	virtnet_process_rx_packets ( netdev );
 }
 
-static void provide_buffers(struct nic *nic)
-{
-   int i;
-   struct vring_list list[2];
-
-   for (i = 0; i < RX_BUF_NB; i++) {
-           list[0].addr = (char*)&rx_hdr[i];
-           list[0].length = sizeof(struct virtio_net_hdr);
-           list[1].addr = (char*)&rx_buffer[i];
-           list[1].length = ETH_FRAME_LEN;
-           vring_add_buf(&virtqueue[RX_INDEX], list, 0, 2, i, i);
-   }
-
-   /* nofify */
-
-   vring_kick(nic->ioaddr, &virtqueue[RX_INDEX], i);
+/** Enable or disable interrupts
+ *
+ * @v netdev	Network device
+ * @v enable	Interrupts should be enabled
+ */
+static void virtnet_irq ( struct net_device *netdev, int enable ) {
+	struct virtnet_nic *virtnet = netdev->priv;
+	int i;
+
+	for ( i = 0; i < QUEUE_NB; i++ ) {
+		if ( enable )
+			vring_enable_cb ( &virtnet->virtqueue[i] );
+		else
+			vring_disable_cb ( &virtnet->virtqueue[i] );
+	}
 }
 
-static struct nic_operations virtnet_operations = {
-	.connect = dummy_connect,
-	.poll = virtnet_poll,
+/** virtio-net device operations */
+static struct net_device_operations virtnet_operations = {
+	.open = virtnet_open,
+	.close = virtnet_close,
 	.transmit = virtnet_transmit,
+	.poll = virtnet_poll,
 	.irq = virtnet_irq,
 };
 
-/*
- * virtnet_probe
- *
- * Look for a virtio network adapter
+/**
+ * Probe PCI device
  *
+ * @v pci	PCI device
+ * @v id	PCI ID
+ * @ret rc	Return status code
  */
+static int virtnet_probe ( struct pci_device *pci,
+			   const struct pci_device_id *id __unused ) {
+	unsigned long ioaddr = pci->ioaddr;
+	struct net_device *netdev;
+	struct virtnet_nic *virtnet;
+	u32 features;
+	int rc;
+
+	/* Allocate and hook up net device */
+	netdev = alloc_etherdev ( sizeof ( *virtnet ) );
+	if ( ! netdev )
+		return -ENOMEM;
+	netdev_init ( netdev, &virtnet_operations );
+	virtnet = netdev->priv;
+	virtnet->ioaddr = ioaddr;
+	pci_set_drvdata ( pci, netdev );
+	netdev->dev = &pci->dev;
+
+	DBGC ( virtnet, "VIRTIO-NET %p busaddr=%s ioaddr=%#lx irq=%d\n",
+	       virtnet, pci->dev.name, ioaddr, pci->irq );
+
+	/* Enable PCI bus master and reset NIC */
+	adjust_pci_device ( pci );
+	vp_reset ( ioaddr );
+
+	/* Load MAC address */
+	features = vp_get_features ( ioaddr );
+	if ( features & ( 1 << VIRTIO_NET_F_MAC ) ) {
+		vp_get ( ioaddr, offsetof ( struct virtio_net_config, mac ),
+			 netdev->hw_addr, ETH_ALEN );
+		DBGC ( virtnet, "VIRTIO-NET %p mac=%s\n", virtnet,
+		       eth_ntoa ( netdev->hw_addr ) );
+	}
+
+	/* Mark link as up, control virtqueue is not used */
+	netdev_link_up ( netdev );
+
+	if ( ( rc = register_netdev ( netdev ) ) != 0 ) {
+		vp_reset ( ioaddr );
+		netdev_nullify ( netdev );
+		netdev_put ( netdev );
+	}
+	return rc;
+}
 
-static int virtnet_probe(struct nic *nic, struct pci_device *pci)
-{
-   u32 features;
-   int i;
-
-   /* Mask the bit that says "this is an io addr" */
-
-   nic->ioaddr = pci->ioaddr & ~3;
-
-   /* Copy IRQ from PCI information */
-
-   nic->irqno = pci->irq;
-
-   printf("I/O address 0x%08x, IRQ #%d\n", nic->ioaddr, nic->irqno);
-
-   adjust_pci_device(pci);
-
-   vp_reset(nic->ioaddr);
-
-   features = vp_get_features(nic->ioaddr);
-   if (features & (1 << VIRTIO_NET_F_MAC)) {
-           vp_get(nic->ioaddr, offsetof(struct virtio_net_config, mac),
-                  nic->node_addr, ETH_ALEN);
-           printf("MAC address ");
-	   for (i = 0; i < ETH_ALEN; i++) {
-                   printf("%02x%c", nic->node_addr[i],
-                          (i == ETH_ALEN - 1) ? '\n' : ':');
-           }
-   }
-
-   /* initialize emit/receive queue */
-
-   for (i = 0; i < QUEUE_NB; i++) {
-           virtqueue[i].free_head = 0;
-           virtqueue[i].last_used_idx = 0;
-           memset((char*)&virtqueue[i].queue, 0, sizeof(virtqueue[i].queue));
-           if (vp_find_vq(nic->ioaddr, i, &virtqueue[i]) == -1)
-                   printf("Cannot register queue #%d\n", i);
-   }
-
-   /* provide some receive buffers */
-
-    provide_buffers(nic);
-
-   /* define NIC interface */
-
-    nic->nic_op = &virtnet_operations;
-
-   /* driver is ready */
-
-   vp_set_features(nic->ioaddr, features & (1 << VIRTIO_NET_F_MAC));
-   vp_set_status(nic->ioaddr, VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK);
+/**
+ * Remove device
+ *
+ * @v pci	PCI device
+ */
+static void virtnet_remove ( struct pci_device *pci ) {
+	struct net_device *netdev = pci_get_drvdata ( pci );
 
-   return 1;
+	unregister_netdev ( netdev );
+	netdev_nullify ( netdev );
+	netdev_put ( netdev );
 }
 
 static struct pci_device_id virtnet_nics[] = {
-PCI_ROM(0x1af4, 0x1000, "virtio-net",              "Virtio Network Interface", 0),
+PCI_ROM(0x1af4, 0x1000, "virtio-net", "Virtio Network Interface", 0),
 };
 
-PCI_DRIVER ( virtnet_driver, virtnet_nics, PCI_NO_CLASS );
-
-DRIVER ( "VIRTIO-NET", nic_driver, pci_driver, virtnet_driver,
-	 virtnet_probe, virtnet_disable );
+struct pci_driver virtnet_driver __pci_driver = {
+	.ids = virtnet_nics,
+	.id_count = ( sizeof ( virtnet_nics ) / sizeof ( virtnet_nics[0] ) ),
+	.probe = virtnet_probe,
+	.remove = virtnet_remove,
+};

src/include/ipxe/errfile.h

@@ -126,6 +126,7 @@ FILE_LICENCE ( GPL2_OR_LATER );
 #define ERRFILE_snpnet		     ( ERRFILE_DRIVER | 0x00590000 )
 #define ERRFILE_snponly		     ( ERRFILE_DRIVER | 0x005a0000 )
 #define ERRFILE_jme		     ( ERRFILE_DRIVER | 0x005b0000 )
+#define ERRFILE_virtio_net	     ( ERRFILE_DRIVER | 0x005c0000 )
 
 #define ERRFILE_scsi		     ( ERRFILE_DRIVER | 0x00700000 )
 #define ERRFILE_arbel		     ( ERRFILE_DRIVER | 0x00710000 )

src/include/ipxe/virtio-pci.h

@@ -69,6 +69,10 @@ static inline void vp_set_status(unsigned int ioaddr, u8 status)
    outb(status, ioaddr + VIRTIO_PCI_STATUS);
 }
 
+static inline u8 vp_get_isr(unsigned int ioaddr)
+{
+   return inb(ioaddr + VIRTIO_PCI_ISR);
+}
 
 static inline void vp_reset(unsigned int ioaddr)
 {

src/include/ipxe/virtio-ring.h

@@ -14,7 +14,7 @@
 /* We've given up on this device. */
 #define VIRTIO_CONFIG_S_FAILED          0x80
 
-#define MAX_QUEUE_NUM      (512)
+#define MAX_QUEUE_NUM      (256)
 
 #define VRING_DESC_F_NEXT  1
 #define VRING_DESC_F_WRITE 2
@@ -71,7 +71,7 @@ struct vring_virtqueue {
    struct vring vring;
    u16 free_head;
    u16 last_used_idx;
-   u16 vdata[MAX_QUEUE_NUM];
+   void *vdata[MAX_QUEUE_NUM];
    /* PCI */
    int queue_index;
 };
@@ -133,10 +133,10 @@ static inline int vring_more_used(struct vring_virtqueue *vq)
 }
 
 void vring_detach(struct vring_virtqueue *vq, unsigned int head);
-int vring_get_buf(struct vring_virtqueue *vq, unsigned int *len);
+void *vring_get_buf(struct vring_virtqueue *vq, unsigned int *len);
 void vring_add_buf(struct vring_virtqueue *vq, struct vring_list list[],
                    unsigned int out, unsigned int in,
-                   int index, int num_added);
+                   void *index, int num_added);
 void vring_kick(unsigned int ioaddr, struct vring_virtqueue *vq, int num_added);
 
 #endif /* _VIRTIO_RING_H_ */