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 /* The skb we couldn't send because buffers were full. */
45 struct sk_buff *last_xmit_skb;
47 struct timer_list xmit_free_timer;
49 /* Number of input buffers, and max we've ever had. */
50 unsigned int num, max;
52 /* For cleaning up after transmission. */
53 struct tasklet_struct tasklet;
55 /* Receive & send queues. */
56 struct sk_buff_head recv;
57 struct sk_buff_head send;
60 static inline struct virtio_net_hdr *skb_vnet_hdr(struct sk_buff *skb)
62 return (struct virtio_net_hdr *)skb->cb;
65 static inline void vnet_hdr_to_sg(struct scatterlist *sg, struct sk_buff *skb)
67 sg_init_one(sg, skb_vnet_hdr(skb), sizeof(struct virtio_net_hdr));
70 static void skb_xmit_done(struct virtqueue *svq)
72 struct virtnet_info *vi = svq->vdev->priv;
74 /* Suppress further interrupts. */
75 svq->vq_ops->disable_cb(svq);
77 /* We were waiting for more output buffers. */
78 netif_wake_queue(vi->dev);
80 /* Make sure we re-xmit last_xmit_skb: if there are no more packets
81 * queued, start_xmit won't be called. */
82 tasklet_schedule(&vi->tasklet);
85 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
88 struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
90 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
91 pr_debug("%s: short packet %i\n", dev->name, len);
92 dev->stats.rx_length_errors++;
95 len -= sizeof(struct virtio_net_hdr);
96 BUG_ON(len > MAX_PACKET_LEN);
100 dev->stats.rx_bytes += skb->len;
101 dev->stats.rx_packets++;
103 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
104 pr_debug("Needs csum!\n");
105 if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
109 skb->protocol = eth_type_trans(skb, dev);
110 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
111 ntohs(skb->protocol), skb->len, skb->pkt_type);
113 if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
115 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
116 case VIRTIO_NET_HDR_GSO_TCPV4:
117 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
119 case VIRTIO_NET_HDR_GSO_UDP:
120 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
122 case VIRTIO_NET_HDR_GSO_TCPV6:
123 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
127 printk(KERN_WARNING "%s: bad gso type %u.\n",
128 dev->name, hdr->gso_type);
132 if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
133 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
135 skb_shinfo(skb)->gso_size = hdr->gso_size;
136 if (skb_shinfo(skb)->gso_size == 0) {
138 printk(KERN_WARNING "%s: zero gso size.\n",
143 /* Header must be checked, and gso_segs computed. */
144 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
145 skb_shinfo(skb)->gso_segs = 0;
148 netif_receive_skb(skb);
152 dev->stats.rx_frame_errors++;
157 static void try_fill_recv(struct virtnet_info *vi)
160 struct scatterlist sg[2+MAX_SKB_FRAGS];
163 sg_init_table(sg, 2+MAX_SKB_FRAGS);
165 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
169 skb_put(skb, MAX_PACKET_LEN);
170 vnet_hdr_to_sg(sg, skb);
171 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
172 skb_queue_head(&vi->recv, skb);
174 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
176 skb_unlink(skb, &vi->recv);
182 if (unlikely(vi->num > vi->max))
184 vi->rvq->vq_ops->kick(vi->rvq);
187 static void skb_recv_done(struct virtqueue *rvq)
189 struct virtnet_info *vi = rvq->vdev->priv;
190 /* Schedule NAPI, Suppress further interrupts if successful. */
191 if (netif_rx_schedule_prep(vi->dev, &vi->napi)) {
192 rvq->vq_ops->disable_cb(rvq);
193 __netif_rx_schedule(vi->dev, &vi->napi);
197 static int virtnet_poll(struct napi_struct *napi, int budget)
199 struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
200 struct sk_buff *skb = NULL;
201 unsigned int len, received = 0;
204 while (received < budget &&
205 (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
206 __skb_unlink(skb, &vi->recv);
207 receive_skb(vi->dev, skb, len);
212 /* FIXME: If we oom and completely run out of inbufs, we need
213 * to start a timer trying to fill more. */
214 if (vi->num < vi->max / 2)
217 /* Out of packets? */
218 if (received < budget) {
219 netif_rx_complete(vi->dev, napi);
220 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
221 && napi_schedule_prep(napi)) {
222 vi->rvq->vq_ops->disable_cb(vi->rvq);
223 __netif_rx_schedule(vi->dev, napi);
231 static void free_old_xmit_skbs(struct virtnet_info *vi)
236 while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
237 pr_debug("Sent skb %p\n", skb);
238 __skb_unlink(skb, &vi->send);
239 vi->dev->stats.tx_bytes += skb->len;
240 vi->dev->stats.tx_packets++;
245 static void xmit_free(unsigned long data)
247 struct virtnet_info *vi = (void *)data;
249 netif_tx_lock(vi->dev);
251 free_old_xmit_skbs(vi);
253 if (!skb_queue_empty(&vi->send))
254 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
256 netif_tx_unlock(vi->dev);
259 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
262 struct scatterlist sg[2+MAX_SKB_FRAGS];
263 struct virtio_net_hdr *hdr;
264 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
266 sg_init_table(sg, 2+MAX_SKB_FRAGS);
268 pr_debug("%s: xmit %p " MAC_FMT "\n", vi->dev->name, skb,
269 dest[0], dest[1], dest[2],
270 dest[3], dest[4], dest[5]);
272 /* Encode metadata header at front. */
273 hdr = skb_vnet_hdr(skb);
274 if (skb->ip_summed == CHECKSUM_PARTIAL) {
275 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
276 hdr->csum_start = skb->csum_start - skb_headroom(skb);
277 hdr->csum_offset = skb->csum_offset;
280 hdr->csum_offset = hdr->csum_start = 0;
283 if (skb_is_gso(skb)) {
284 hdr->hdr_len = skb_transport_header(skb) - skb->data;
285 hdr->gso_size = skb_shinfo(skb)->gso_size;
286 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
287 hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
288 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
289 hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
290 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
291 hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
294 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
295 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
297 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
298 hdr->gso_size = hdr->hdr_len = 0;
301 vnet_hdr_to_sg(sg, skb);
302 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
304 err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
306 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
311 static void xmit_tasklet(unsigned long data)
313 struct virtnet_info *vi = (void *)data;
315 netif_tx_lock_bh(vi->dev);
316 if (vi->last_xmit_skb && xmit_skb(vi, vi->last_xmit_skb) == 0) {
317 vi->svq->vq_ops->kick(vi->svq);
318 vi->last_xmit_skb = NULL;
320 netif_tx_unlock_bh(vi->dev);
323 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
325 struct virtnet_info *vi = netdev_priv(dev);
328 /* Free up any pending old buffers before queueing new ones. */
329 free_old_xmit_skbs(vi);
331 /* If we has a buffer left over from last time, send it now. */
332 if (unlikely(vi->last_xmit_skb)) {
333 if (xmit_skb(vi, vi->last_xmit_skb) != 0) {
334 /* Drop this skb: we only queue one. */
335 vi->dev->stats.tx_dropped++;
340 vi->last_xmit_skb = NULL;
343 /* Put new one in send queue and do transmit */
345 __skb_queue_head(&vi->send, skb);
346 if (xmit_skb(vi, skb) != 0) {
347 vi->last_xmit_skb = skb;
353 vi->svq->vq_ops->kick(vi->svq);
357 pr_debug("%s: virtio not prepared to send\n", dev->name);
358 netif_stop_queue(dev);
360 /* Activate callback for using skbs: if this returns false it
361 * means some were used in the meantime. */
362 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
363 vi->svq->vq_ops->disable_cb(vi->svq);
364 netif_start_queue(dev);
370 #ifdef CONFIG_NET_POLL_CONTROLLER
371 static void virtnet_netpoll(struct net_device *dev)
373 struct virtnet_info *vi = netdev_priv(dev);
375 napi_schedule(&vi->napi);
379 static int virtnet_open(struct net_device *dev)
381 struct virtnet_info *vi = netdev_priv(dev);
383 napi_enable(&vi->napi);
385 /* If all buffers were filled by other side before we napi_enabled, we
386 * won't get another interrupt, so process any outstanding packets
387 * now. virtnet_poll wants re-enable the queue, so we disable here.
388 * We synchronize against interrupts via NAPI_STATE_SCHED */
389 if (netif_rx_schedule_prep(dev, &vi->napi)) {
390 vi->rvq->vq_ops->disable_cb(vi->rvq);
391 __netif_rx_schedule(dev, &vi->napi);
396 static int virtnet_close(struct net_device *dev)
398 struct virtnet_info *vi = netdev_priv(dev);
400 napi_disable(&vi->napi);
405 static int virtnet_probe(struct virtio_device *vdev)
408 struct net_device *dev;
409 struct virtnet_info *vi;
411 /* Allocate ourselves a network device with room for our info */
412 dev = alloc_etherdev(sizeof(struct virtnet_info));
416 /* Set up network device as normal. */
417 dev->open = virtnet_open;
418 dev->stop = virtnet_close;
419 dev->hard_start_xmit = start_xmit;
420 dev->features = NETIF_F_HIGHDMA;
421 #ifdef CONFIG_NET_POLL_CONTROLLER
422 dev->poll_controller = virtnet_netpoll;
424 SET_NETDEV_DEV(dev, &vdev->dev);
426 /* Do we support "hardware" checksums? */
427 if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
428 /* This opens up the world of extra features. */
429 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
430 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
431 dev->features |= NETIF_F_TSO | NETIF_F_UFO
432 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
434 /* Individual feature bits: what can host handle? */
435 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
436 dev->features |= NETIF_F_TSO;
437 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
438 dev->features |= NETIF_F_TSO6;
439 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
440 dev->features |= NETIF_F_TSO_ECN;
441 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
442 dev->features |= NETIF_F_UFO;
445 /* Configuration may specify what MAC to use. Otherwise random. */
446 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
447 vdev->config->get(vdev,
448 offsetof(struct virtio_net_config, mac),
449 dev->dev_addr, dev->addr_len);
451 random_ether_addr(dev->dev_addr);
453 /* Set up our device-specific information */
454 vi = netdev_priv(dev);
455 netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
460 /* We expect two virtqueues, receive then send. */
461 vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
462 if (IS_ERR(vi->rvq)) {
463 err = PTR_ERR(vi->rvq);
467 vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
468 if (IS_ERR(vi->svq)) {
469 err = PTR_ERR(vi->svq);
473 /* Initialize our empty receive and send queues. */
474 skb_queue_head_init(&vi->recv);
475 skb_queue_head_init(&vi->send);
477 tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
479 setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
481 err = register_netdev(dev);
483 pr_debug("virtio_net: registering device failed\n");
487 /* Last of all, set up some receive buffers. */
490 /* If we didn't even get one input buffer, we're useless. */
496 pr_debug("virtnet: registered device %s\n", dev->name);
500 unregister_netdev(dev);
502 vdev->config->del_vq(vi->svq);
504 vdev->config->del_vq(vi->rvq);
510 static void virtnet_remove(struct virtio_device *vdev)
512 struct virtnet_info *vi = vdev->priv;
515 /* Stop all the virtqueues. */
516 vdev->config->reset(vdev);
518 del_timer_sync(&vi->xmit_free_timer);
520 /* Free our skbs in send and recv queues, if any. */
521 while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
525 __skb_queue_purge(&vi->send);
527 BUG_ON(vi->num != 0);
529 vdev->config->del_vq(vi->svq);
530 vdev->config->del_vq(vi->rvq);
531 unregister_netdev(vi->dev);
532 free_netdev(vi->dev);
535 static struct virtio_device_id id_table[] = {
536 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
540 static unsigned int features[] = {
541 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
542 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
543 VIRTIO_NET_F_HOST_ECN,
546 static struct virtio_driver virtio_net = {
547 .feature_table = features,
548 .feature_table_size = ARRAY_SIZE(features),
549 .driver.name = KBUILD_MODNAME,
550 .driver.owner = THIS_MODULE,
551 .id_table = id_table,
552 .probe = virtnet_probe,
553 .remove = __devexit_p(virtnet_remove),
556 static int __init init(void)
558 return register_virtio_driver(&virtio_net);
561 static void __exit fini(void)
563 unregister_virtio_driver(&virtio_net);
568 MODULE_DEVICE_TABLE(virtio, id_table);
569 MODULE_DESCRIPTION("Virtio network driver");
570 MODULE_LICENSE("GPL");