1 /* A simple network driver using virtio.
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/module.h>
23 #include <linux/virtio.h>
24 #include <linux/virtio_net.h>
25 #include <linux/scatterlist.h>
27 static int napi_weight
= 128;
28 module_param(napi_weight
, int, 0444);
30 static int csum
= 1, gso
= 1;
31 module_param(csum
, bool, 0444);
32 module_param(gso
, bool, 0444);
34 /* FIXME: MTU in config. */
35 #define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN)
39 struct virtio_device
*vdev
;
40 struct virtqueue
*rvq
, *svq
;
41 struct net_device
*dev
;
42 struct napi_struct napi
;
44 /* Number of input buffers, and max we've ever had. */
45 unsigned int num
, max
;
47 /* Receive & send queues. */
48 struct sk_buff_head recv
;
49 struct sk_buff_head send
;
52 static inline struct virtio_net_hdr
*skb_vnet_hdr(struct sk_buff
*skb
)
54 return (struct virtio_net_hdr
*)skb
->cb
;
57 static inline void vnet_hdr_to_sg(struct scatterlist
*sg
, struct sk_buff
*skb
)
59 sg_init_one(sg
, skb_vnet_hdr(skb
), sizeof(struct virtio_net_hdr
));
62 static void skb_xmit_done(struct virtqueue
*svq
)
64 struct virtnet_info
*vi
= svq
->vdev
->priv
;
66 /* Suppress further interrupts. */
67 svq
->vq_ops
->disable_cb(svq
);
68 /* We were waiting for more output buffers. */
69 netif_wake_queue(vi
->dev
);
72 static void receive_skb(struct net_device
*dev
, struct sk_buff
*skb
,
75 struct virtio_net_hdr
*hdr
= skb_vnet_hdr(skb
);
77 if (unlikely(len
< sizeof(struct virtio_net_hdr
) + ETH_HLEN
)) {
78 pr_debug("%s: short packet %i\n", dev
->name
, len
);
79 dev
->stats
.rx_length_errors
++;
82 len
-= sizeof(struct virtio_net_hdr
);
83 BUG_ON(len
> MAX_PACKET_LEN
);
86 skb
->protocol
= eth_type_trans(skb
, dev
);
87 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
88 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
89 dev
->stats
.rx_bytes
+= skb
->len
;
90 dev
->stats
.rx_packets
++;
92 if (hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
93 pr_debug("Needs csum!\n");
94 if (!skb_partial_csum_set(skb
,hdr
->csum_start
,hdr
->csum_offset
))
98 if (hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
100 switch (hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
101 case VIRTIO_NET_HDR_GSO_TCPV4
:
102 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
104 case VIRTIO_NET_HDR_GSO_UDP
:
105 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
107 case VIRTIO_NET_HDR_GSO_TCPV6
:
108 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
112 printk(KERN_WARNING
"%s: bad gso type %u.\n",
113 dev
->name
, hdr
->gso_type
);
117 if (hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
118 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
120 skb_shinfo(skb
)->gso_size
= hdr
->gso_size
;
121 if (skb_shinfo(skb
)->gso_size
== 0) {
123 printk(KERN_WARNING
"%s: zero gso size.\n",
128 /* Header must be checked, and gso_segs computed. */
129 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
130 skb_shinfo(skb
)->gso_segs
= 0;
133 netif_receive_skb(skb
);
137 dev
->stats
.rx_frame_errors
++;
142 static void try_fill_recv(struct virtnet_info
*vi
)
145 struct scatterlist sg
[1+MAX_SKB_FRAGS
];
148 sg_init_table(sg
, 1+MAX_SKB_FRAGS
);
150 skb
= netdev_alloc_skb(vi
->dev
, MAX_PACKET_LEN
);
154 skb_put(skb
, MAX_PACKET_LEN
);
155 vnet_hdr_to_sg(sg
, skb
);
156 num
= skb_to_sgvec(skb
, sg
+1, 0, skb
->len
) + 1;
157 skb_queue_head(&vi
->recv
, skb
);
159 err
= vi
->rvq
->vq_ops
->add_buf(vi
->rvq
, sg
, 0, num
, skb
);
161 skb_unlink(skb
, &vi
->recv
);
167 if (unlikely(vi
->num
> vi
->max
))
169 vi
->rvq
->vq_ops
->kick(vi
->rvq
);
172 static void skb_recv_done(struct virtqueue
*rvq
)
174 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
175 /* Schedule NAPI, Suppress further interrupts if successful. */
176 if (netif_rx_schedule_prep(vi
->dev
, &vi
->napi
)) {
177 rvq
->vq_ops
->disable_cb(rvq
);
178 __netif_rx_schedule(vi
->dev
, &vi
->napi
);
182 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
184 struct virtnet_info
*vi
= container_of(napi
, struct virtnet_info
, napi
);
185 struct sk_buff
*skb
= NULL
;
186 unsigned int len
, received
= 0;
189 while (received
< budget
&&
190 (skb
= vi
->rvq
->vq_ops
->get_buf(vi
->rvq
, &len
)) != NULL
) {
191 __skb_unlink(skb
, &vi
->recv
);
192 receive_skb(vi
->dev
, skb
, len
);
197 /* FIXME: If we oom and completely run out of inbufs, we need
198 * to start a timer trying to fill more. */
199 if (vi
->num
< vi
->max
/ 2)
202 /* Out of packets? */
203 if (received
< budget
) {
204 netif_rx_complete(vi
->dev
, napi
);
205 if (unlikely(!vi
->rvq
->vq_ops
->enable_cb(vi
->rvq
))
206 && napi_schedule_prep(napi
)) {
207 vi
->rvq
->vq_ops
->disable_cb(vi
->rvq
);
208 __netif_rx_schedule(vi
->dev
, napi
);
216 static void free_old_xmit_skbs(struct virtnet_info
*vi
)
221 while ((skb
= vi
->svq
->vq_ops
->get_buf(vi
->svq
, &len
)) != NULL
) {
222 pr_debug("Sent skb %p\n", skb
);
223 __skb_unlink(skb
, &vi
->send
);
224 vi
->dev
->stats
.tx_bytes
+= len
;
225 vi
->dev
->stats
.tx_packets
++;
230 static int start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
232 struct virtnet_info
*vi
= netdev_priv(dev
);
234 struct scatterlist sg
[1+MAX_SKB_FRAGS
];
235 struct virtio_net_hdr
*hdr
;
236 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
237 DECLARE_MAC_BUF(mac
);
239 sg_init_table(sg
, 1+MAX_SKB_FRAGS
);
241 pr_debug("%s: xmit %p %s\n", dev
->name
, skb
, print_mac(mac
, dest
));
243 /* Encode metadata header at front. */
244 hdr
= skb_vnet_hdr(skb
);
245 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
246 hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
247 hdr
->csum_start
= skb
->csum_start
- skb_headroom(skb
);
248 hdr
->csum_offset
= skb
->csum_offset
;
251 hdr
->csum_offset
= hdr
->csum_start
= 0;
254 if (skb_is_gso(skb
)) {
255 hdr
->hdr_len
= skb_transport_header(skb
) - skb
->data
;
256 hdr
->gso_size
= skb_shinfo(skb
)->gso_size
;
257 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
258 hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
259 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
260 hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
261 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
262 hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
265 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
266 hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
268 hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
269 hdr
->gso_size
= hdr
->hdr_len
= 0;
272 vnet_hdr_to_sg(sg
, skb
);
273 num
= skb_to_sgvec(skb
, sg
+1, 0, skb
->len
) + 1;
274 __skb_queue_head(&vi
->send
, skb
);
277 /* Free up any pending old buffers before queueing new ones. */
278 free_old_xmit_skbs(vi
);
279 err
= vi
->svq
->vq_ops
->add_buf(vi
->svq
, sg
, num
, 0, skb
);
281 pr_debug("%s: virtio not prepared to send\n", dev
->name
);
282 netif_stop_queue(dev
);
284 /* Activate callback for using skbs: if this returns false it
285 * means some were used in the meantime. */
286 if (unlikely(!vi
->svq
->vq_ops
->enable_cb(vi
->svq
))) {
287 printk("Unlikely: restart svq race\n");
288 vi
->svq
->vq_ops
->disable_cb(vi
->svq
);
289 netif_start_queue(dev
);
292 __skb_unlink(skb
, &vi
->send
);
294 return NETDEV_TX_BUSY
;
296 vi
->svq
->vq_ops
->kick(vi
->svq
);
301 #ifdef CONFIG_NET_POLL_CONTROLLER
302 static void virtnet_netpoll(struct net_device
*dev
)
304 struct virtnet_info
*vi
= netdev_priv(dev
);
306 napi_schedule(&vi
->napi
);
310 static int virtnet_open(struct net_device
*dev
)
312 struct virtnet_info
*vi
= netdev_priv(dev
);
314 napi_enable(&vi
->napi
);
316 /* If all buffers were filled by other side before we napi_enabled, we
317 * won't get another interrupt, so process any outstanding packets
318 * now. virtnet_poll wants re-enable the queue, so we disable here.
319 * We synchronize against interrupts via NAPI_STATE_SCHED */
320 if (netif_rx_schedule_prep(dev
, &vi
->napi
)) {
321 vi
->rvq
->vq_ops
->disable_cb(vi
->rvq
);
322 __netif_rx_schedule(dev
, &vi
->napi
);
327 static int virtnet_close(struct net_device
*dev
)
329 struct virtnet_info
*vi
= netdev_priv(dev
);
331 napi_disable(&vi
->napi
);
336 static int virtnet_probe(struct virtio_device
*vdev
)
339 struct net_device
*dev
;
340 struct virtnet_info
*vi
;
342 /* Allocate ourselves a network device with room for our info */
343 dev
= alloc_etherdev(sizeof(struct virtnet_info
));
347 /* Set up network device as normal. */
348 dev
->open
= virtnet_open
;
349 dev
->stop
= virtnet_close
;
350 dev
->hard_start_xmit
= start_xmit
;
351 dev
->features
= NETIF_F_HIGHDMA
;
352 #ifdef CONFIG_NET_POLL_CONTROLLER
353 dev
->poll_controller
= virtnet_netpoll
;
355 SET_NETDEV_DEV(dev
, &vdev
->dev
);
357 /* Do we support "hardware" checksums? */
358 if (csum
&& vdev
->config
->feature(vdev
, VIRTIO_NET_F_CSUM
)) {
359 /* This opens up the world of extra features. */
360 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
361 if (gso
&& vdev
->config
->feature(vdev
, VIRTIO_NET_F_GSO
)) {
362 dev
->features
|= NETIF_F_TSO
| NETIF_F_UFO
363 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
367 /* Configuration may specify what MAC to use. Otherwise random. */
368 if (vdev
->config
->feature(vdev
, VIRTIO_NET_F_MAC
)) {
369 vdev
->config
->get(vdev
,
370 offsetof(struct virtio_net_config
, mac
),
371 dev
->dev_addr
, dev
->addr_len
);
373 random_ether_addr(dev
->dev_addr
);
375 /* Set up our device-specific information */
376 vi
= netdev_priv(dev
);
377 netif_napi_add(dev
, &vi
->napi
, virtnet_poll
, napi_weight
);
382 /* We expect two virtqueues, receive then send. */
383 vi
->rvq
= vdev
->config
->find_vq(vdev
, 0, skb_recv_done
);
384 if (IS_ERR(vi
->rvq
)) {
385 err
= PTR_ERR(vi
->rvq
);
389 vi
->svq
= vdev
->config
->find_vq(vdev
, 1, skb_xmit_done
);
390 if (IS_ERR(vi
->svq
)) {
391 err
= PTR_ERR(vi
->svq
);
395 /* Initialize our empty receive and send queues. */
396 skb_queue_head_init(&vi
->recv
);
397 skb_queue_head_init(&vi
->send
);
399 err
= register_netdev(dev
);
401 pr_debug("virtio_net: registering device failed\n");
405 /* Last of all, set up some receive buffers. */
408 /* If we didn't even get one input buffer, we're useless. */
414 pr_debug("virtnet: registered device %s\n", dev
->name
);
418 unregister_netdev(dev
);
420 vdev
->config
->del_vq(vi
->svq
);
422 vdev
->config
->del_vq(vi
->rvq
);
428 static void virtnet_remove(struct virtio_device
*vdev
)
430 struct virtnet_info
*vi
= vdev
->priv
;
433 /* Stop all the virtqueues. */
434 vdev
->config
->reset(vdev
);
436 /* Free our skbs in send and recv queues, if any. */
437 while ((skb
= __skb_dequeue(&vi
->recv
)) != NULL
) {
441 while ((skb
= __skb_dequeue(&vi
->send
)) != NULL
)
444 BUG_ON(vi
->num
!= 0);
446 vdev
->config
->del_vq(vi
->svq
);
447 vdev
->config
->del_vq(vi
->rvq
);
448 unregister_netdev(vi
->dev
);
449 free_netdev(vi
->dev
);
452 static struct virtio_device_id id_table
[] = {
453 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
457 static struct virtio_driver virtio_net
= {
458 .driver
.name
= KBUILD_MODNAME
,
459 .driver
.owner
= THIS_MODULE
,
460 .id_table
= id_table
,
461 .probe
= virtnet_probe
,
462 .remove
= __devexit_p(virtnet_remove
),
465 static int __init
init(void)
467 return register_virtio_driver(&virtio_net
);
470 static void __exit
fini(void)
472 unregister_virtio_driver(&virtio_net
);
477 MODULE_DEVICE_TABLE(virtio
, id_table
);
478 MODULE_DESCRIPTION("Virtio network driver");
479 MODULE_LICENSE("GPL");