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/ethtool.h>
23 #include <linux/module.h>
24 #include <linux/virtio.h>
25 #include <linux/virtio_net.h>
26 #include <linux/scatterlist.h>
28 static int napi_weight = 128;
29 module_param(napi_weight, int, 0444);
31 static int csum = 1, gso = 1;
32 module_param(csum, bool, 0444);
33 module_param(gso, bool, 0444);
35 /* FIXME: MTU in config. */
36 #define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN)
40 struct virtio_device *vdev;
41 struct virtqueue *rvq, *svq;
42 struct net_device *dev;
43 struct napi_struct napi;
45 /* The skb we couldn't send because buffers were full. */
46 struct sk_buff *last_xmit_skb;
48 /* If we need to free in a timer, this is it. */
49 struct timer_list xmit_free_timer;
51 /* Number of input buffers, and max we've ever had. */
52 unsigned int num, max;
54 /* For cleaning up after transmission. */
55 struct tasklet_struct tasklet;
58 /* I like... big packets and I cannot lie! */
61 /* Receive & send queues. */
62 struct sk_buff_head recv;
63 struct sk_buff_head send;
65 /* Chain pages by the private ptr. */
69 static inline struct virtio_net_hdr *skb_vnet_hdr(struct sk_buff *skb)
71 return (struct virtio_net_hdr *)skb->cb;
74 static inline void vnet_hdr_to_sg(struct scatterlist *sg, struct sk_buff *skb)
76 sg_init_one(sg, skb_vnet_hdr(skb), sizeof(struct virtio_net_hdr));
79 static void give_a_page(struct virtnet_info *vi, struct page *page)
81 page->private = (unsigned long)vi->pages;
85 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
87 struct page *p = vi->pages;
90 vi->pages = (struct page *)p->private;
92 p = alloc_page(gfp_mask);
96 static void skb_xmit_done(struct virtqueue *svq)
98 struct virtnet_info *vi = svq->vdev->priv;
100 /* Suppress further interrupts. */
101 svq->vq_ops->disable_cb(svq);
103 /* We were probably waiting for more output buffers. */
104 netif_wake_queue(vi->dev);
106 /* Make sure we re-xmit last_xmit_skb: if there are no more packets
107 * queued, start_xmit won't be called. */
108 tasklet_schedule(&vi->tasklet);
111 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
114 struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
117 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
118 pr_debug("%s: short packet %i\n", dev->name, len);
119 dev->stats.rx_length_errors++;
122 len -= sizeof(struct virtio_net_hdr);
124 if (len <= MAX_PACKET_LEN) {
127 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
128 give_a_page(netdev_priv(dev),
129 skb_shinfo(skb)->frags[i].page);
131 skb_shinfo(skb)->nr_frags = 0;
134 err = pskb_trim(skb, len);
136 pr_debug("%s: pskb_trim failed %i %d\n", dev->name, len, err);
137 dev->stats.rx_dropped++;
140 skb->truesize += skb->data_len;
141 dev->stats.rx_bytes += skb->len;
142 dev->stats.rx_packets++;
144 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
145 pr_debug("Needs csum!\n");
146 if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
150 skb->protocol = eth_type_trans(skb, dev);
151 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
152 ntohs(skb->protocol), skb->len, skb->pkt_type);
154 if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
156 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
157 case VIRTIO_NET_HDR_GSO_TCPV4:
158 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
160 case VIRTIO_NET_HDR_GSO_UDP:
161 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
163 case VIRTIO_NET_HDR_GSO_TCPV6:
164 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
168 printk(KERN_WARNING "%s: bad gso type %u.\n",
169 dev->name, hdr->gso_type);
173 if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
174 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
176 skb_shinfo(skb)->gso_size = hdr->gso_size;
177 if (skb_shinfo(skb)->gso_size == 0) {
179 printk(KERN_WARNING "%s: zero gso size.\n",
184 /* Header must be checked, and gso_segs computed. */
185 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
186 skb_shinfo(skb)->gso_segs = 0;
189 netif_receive_skb(skb);
193 dev->stats.rx_frame_errors++;
198 static void try_fill_recv(struct virtnet_info *vi)
201 struct scatterlist sg[2+MAX_SKB_FRAGS];
204 sg_init_table(sg, 2+MAX_SKB_FRAGS);
206 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
210 skb_put(skb, MAX_PACKET_LEN);
211 vnet_hdr_to_sg(sg, skb);
213 if (vi->big_packets) {
214 for (i = 0; i < MAX_SKB_FRAGS; i++) {
215 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
216 f->page = get_a_page(vi, GFP_ATOMIC);
223 skb->data_len += PAGE_SIZE;
224 skb->len += PAGE_SIZE;
226 skb_shinfo(skb)->nr_frags++;
230 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
231 skb_queue_head(&vi->recv, skb);
233 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
235 skb_unlink(skb, &vi->recv);
241 if (unlikely(vi->num > vi->max))
243 vi->rvq->vq_ops->kick(vi->rvq);
246 static void skb_recv_done(struct virtqueue *rvq)
248 struct virtnet_info *vi = rvq->vdev->priv;
249 /* Schedule NAPI, Suppress further interrupts if successful. */
250 if (netif_rx_schedule_prep(vi->dev, &vi->napi)) {
251 rvq->vq_ops->disable_cb(rvq);
252 __netif_rx_schedule(vi->dev, &vi->napi);
256 static int virtnet_poll(struct napi_struct *napi, int budget)
258 struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
259 struct sk_buff *skb = NULL;
260 unsigned int len, received = 0;
263 while (received < budget &&
264 (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
265 __skb_unlink(skb, &vi->recv);
266 receive_skb(vi->dev, skb, len);
271 /* FIXME: If we oom and completely run out of inbufs, we need
272 * to start a timer trying to fill more. */
273 if (vi->num < vi->max / 2)
276 /* Out of packets? */
277 if (received < budget) {
278 netif_rx_complete(vi->dev, napi);
279 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
280 && napi_schedule_prep(napi)) {
281 vi->rvq->vq_ops->disable_cb(vi->rvq);
282 __netif_rx_schedule(vi->dev, napi);
290 static void free_old_xmit_skbs(struct virtnet_info *vi)
295 while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
296 pr_debug("Sent skb %p\n", skb);
297 __skb_unlink(skb, &vi->send);
298 vi->dev->stats.tx_bytes += skb->len;
299 vi->dev->stats.tx_packets++;
304 /* If the virtio transport doesn't always notify us when all in-flight packets
305 * are consumed, we fall back to using this function on a timer to free them. */
306 static void xmit_free(unsigned long data)
308 struct virtnet_info *vi = (void *)data;
310 netif_tx_lock(vi->dev);
312 free_old_xmit_skbs(vi);
314 if (!skb_queue_empty(&vi->send))
315 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
317 netif_tx_unlock(vi->dev);
320 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
323 struct scatterlist sg[2+MAX_SKB_FRAGS];
324 struct virtio_net_hdr *hdr;
325 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
327 sg_init_table(sg, 2+MAX_SKB_FRAGS);
329 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
331 /* Encode metadata header at front. */
332 hdr = skb_vnet_hdr(skb);
333 if (skb->ip_summed == CHECKSUM_PARTIAL) {
334 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
335 hdr->csum_start = skb->csum_start - skb_headroom(skb);
336 hdr->csum_offset = skb->csum_offset;
339 hdr->csum_offset = hdr->csum_start = 0;
342 if (skb_is_gso(skb)) {
343 hdr->hdr_len = skb_transport_header(skb) - skb->data;
344 hdr->gso_size = skb_shinfo(skb)->gso_size;
345 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
346 hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
347 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
348 hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
349 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
350 hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
353 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
354 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
356 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
357 hdr->gso_size = hdr->hdr_len = 0;
360 vnet_hdr_to_sg(sg, skb);
361 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
363 err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
364 if (!err && !vi->free_in_tasklet)
365 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
370 static void xmit_tasklet(unsigned long data)
372 struct virtnet_info *vi = (void *)data;
374 netif_tx_lock_bh(vi->dev);
375 if (vi->last_xmit_skb && xmit_skb(vi, vi->last_xmit_skb) == 0) {
376 vi->svq->vq_ops->kick(vi->svq);
377 vi->last_xmit_skb = NULL;
379 if (vi->free_in_tasklet)
380 free_old_xmit_skbs(vi);
381 netif_tx_unlock_bh(vi->dev);
384 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
386 struct virtnet_info *vi = netdev_priv(dev);
389 /* Free up any pending old buffers before queueing new ones. */
390 free_old_xmit_skbs(vi);
392 /* If we has a buffer left over from last time, send it now. */
393 if (unlikely(vi->last_xmit_skb) &&
394 xmit_skb(vi, vi->last_xmit_skb) != 0)
397 vi->last_xmit_skb = NULL;
399 /* Put new one in send queue and do transmit */
401 __skb_queue_head(&vi->send, skb);
402 if (xmit_skb(vi, skb) != 0) {
403 vi->last_xmit_skb = skb;
409 vi->svq->vq_ops->kick(vi->svq);
413 pr_debug("%s: virtio not prepared to send\n", dev->name);
414 netif_stop_queue(dev);
416 /* Activate callback for using skbs: if this returns false it
417 * means some were used in the meantime. */
418 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
419 vi->svq->vq_ops->disable_cb(vi->svq);
420 netif_start_queue(dev);
424 /* Drop this skb: we only queue one. */
425 vi->dev->stats.tx_dropped++;
431 #ifdef CONFIG_NET_POLL_CONTROLLER
432 static void virtnet_netpoll(struct net_device *dev)
434 struct virtnet_info *vi = netdev_priv(dev);
436 napi_schedule(&vi->napi);
440 static int virtnet_open(struct net_device *dev)
442 struct virtnet_info *vi = netdev_priv(dev);
444 napi_enable(&vi->napi);
446 /* If all buffers were filled by other side before we napi_enabled, we
447 * won't get another interrupt, so process any outstanding packets
448 * now. virtnet_poll wants re-enable the queue, so we disable here.
449 * We synchronize against interrupts via NAPI_STATE_SCHED */
450 if (netif_rx_schedule_prep(dev, &vi->napi)) {
451 vi->rvq->vq_ops->disable_cb(vi->rvq);
452 __netif_rx_schedule(dev, &vi->napi);
457 static int virtnet_close(struct net_device *dev)
459 struct virtnet_info *vi = netdev_priv(dev);
461 napi_disable(&vi->napi);
466 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
468 struct virtnet_info *vi = netdev_priv(dev);
469 struct virtio_device *vdev = vi->vdev;
471 if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
474 return ethtool_op_set_tx_hw_csum(dev, data);
477 static struct ethtool_ops virtnet_ethtool_ops = {
478 .set_tx_csum = virtnet_set_tx_csum,
479 .set_sg = ethtool_op_set_sg,
482 static int virtnet_probe(struct virtio_device *vdev)
485 struct net_device *dev;
486 struct virtnet_info *vi;
488 /* Allocate ourselves a network device with room for our info */
489 dev = alloc_etherdev(sizeof(struct virtnet_info));
493 /* Set up network device as normal. */
494 dev->open = virtnet_open;
495 dev->stop = virtnet_close;
496 dev->hard_start_xmit = start_xmit;
497 dev->features = NETIF_F_HIGHDMA;
498 #ifdef CONFIG_NET_POLL_CONTROLLER
499 dev->poll_controller = virtnet_netpoll;
501 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
502 SET_NETDEV_DEV(dev, &vdev->dev);
504 /* Do we support "hardware" checksums? */
505 if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
506 /* This opens up the world of extra features. */
507 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
508 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
509 dev->features |= NETIF_F_TSO | NETIF_F_UFO
510 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
512 /* Individual feature bits: what can host handle? */
513 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
514 dev->features |= NETIF_F_TSO;
515 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
516 dev->features |= NETIF_F_TSO6;
517 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
518 dev->features |= NETIF_F_TSO_ECN;
519 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
520 dev->features |= NETIF_F_UFO;
523 /* Configuration may specify what MAC to use. Otherwise random. */
524 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
525 vdev->config->get(vdev,
526 offsetof(struct virtio_net_config, mac),
527 dev->dev_addr, dev->addr_len);
529 random_ether_addr(dev->dev_addr);
531 /* Set up our device-specific information */
532 vi = netdev_priv(dev);
533 netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
539 /* If they give us a callback when all buffers are done, we don't need
541 vi->free_in_tasklet = virtio_has_feature(vdev,VIRTIO_F_NOTIFY_ON_EMPTY);
543 /* If we can receive ANY GSO packets, we must allocate large ones. */
544 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
545 || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
546 || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
547 vi->big_packets = true;
549 /* We expect two virtqueues, receive then send. */
550 vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
551 if (IS_ERR(vi->rvq)) {
552 err = PTR_ERR(vi->rvq);
556 vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
557 if (IS_ERR(vi->svq)) {
558 err = PTR_ERR(vi->svq);
562 /* Initialize our empty receive and send queues. */
563 skb_queue_head_init(&vi->recv);
564 skb_queue_head_init(&vi->send);
566 tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
568 if (!vi->free_in_tasklet)
569 setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
571 err = register_netdev(dev);
573 pr_debug("virtio_net: registering device failed\n");
577 /* Last of all, set up some receive buffers. */
580 /* If we didn't even get one input buffer, we're useless. */
586 pr_debug("virtnet: registered device %s\n", dev->name);
590 unregister_netdev(dev);
592 vdev->config->del_vq(vi->svq);
594 vdev->config->del_vq(vi->rvq);
600 static void virtnet_remove(struct virtio_device *vdev)
602 struct virtnet_info *vi = vdev->priv;
605 /* Stop all the virtqueues. */
606 vdev->config->reset(vdev);
608 if (!vi->free_in_tasklet)
609 del_timer_sync(&vi->xmit_free_timer);
611 /* Free our skbs in send and recv queues, if any. */
612 while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
616 __skb_queue_purge(&vi->send);
618 BUG_ON(vi->num != 0);
620 vdev->config->del_vq(vi->svq);
621 vdev->config->del_vq(vi->rvq);
622 unregister_netdev(vi->dev);
625 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
627 free_netdev(vi->dev);
630 static struct virtio_device_id id_table[] = {
631 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
635 static unsigned int features[] = {
636 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
637 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
638 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
639 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
640 VIRTIO_NET_F_GUEST_ECN, /* We don't yet handle UFO input. */
641 VIRTIO_F_NOTIFY_ON_EMPTY,
644 static struct virtio_driver virtio_net = {
645 .feature_table = features,
646 .feature_table_size = ARRAY_SIZE(features),
647 .driver.name = KBUILD_MODNAME,
648 .driver.owner = THIS_MODULE,
649 .id_table = id_table,
650 .probe = virtnet_probe,
651 .remove = __devexit_p(virtnet_remove),
654 static int __init init(void)
656 return register_virtio_driver(&virtio_net);
659 static void __exit fini(void)
661 unregister_virtio_driver(&virtio_net);
666 MODULE_DEVICE_TABLE(virtio, id_table);
667 MODULE_DESCRIPTION("Virtio network driver");
668 MODULE_LICENSE("GPL");