Merge git://git.infradead.org/mtd-2.6
[linux-2.6/zen-sources.git] / drivers / net / virtio_net.c
blob0196a0df90210995ec6a5abe228eb23bb22d494c
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
19 //#define DEBUG
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)
38 struct virtnet_info
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;
56 bool free_in_tasklet;
58 /* I like... big packets and I cannot lie! */
59 bool big_packets;
61 /* Receive & send queues. */
62 struct sk_buff_head recv;
63 struct sk_buff_head send;
65 /* Chain pages by the private ptr. */
66 struct page *pages;
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;
82 vi->pages = page;
85 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
87 struct page *p = vi->pages;
89 if (p)
90 vi->pages = (struct page *)p->private;
91 else
92 p = alloc_page(gfp_mask);
93 return p;
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,
112 unsigned len)
114 struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
115 int err;
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++;
120 goto drop;
122 len -= sizeof(struct virtio_net_hdr);
124 if (len <= MAX_PACKET_LEN) {
125 unsigned int i;
127 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
128 give_a_page(dev->priv, skb_shinfo(skb)->frags[i].page);
129 skb->data_len = 0;
130 skb_shinfo(skb)->nr_frags = 0;
133 err = pskb_trim(skb, len);
134 if (err) {
135 pr_debug("%s: pskb_trim failed %i %d\n", dev->name, len, err);
136 dev->stats.rx_dropped++;
137 goto drop;
139 skb->truesize += skb->data_len;
140 dev->stats.rx_bytes += skb->len;
141 dev->stats.rx_packets++;
143 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
144 pr_debug("Needs csum!\n");
145 if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
146 goto frame_err;
149 skb->protocol = eth_type_trans(skb, dev);
150 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
151 ntohs(skb->protocol), skb->len, skb->pkt_type);
153 if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
154 pr_debug("GSO!\n");
155 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
156 case VIRTIO_NET_HDR_GSO_TCPV4:
157 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
158 break;
159 case VIRTIO_NET_HDR_GSO_UDP:
160 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
161 break;
162 case VIRTIO_NET_HDR_GSO_TCPV6:
163 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
164 break;
165 default:
166 if (net_ratelimit())
167 printk(KERN_WARNING "%s: bad gso type %u.\n",
168 dev->name, hdr->gso_type);
169 goto frame_err;
172 if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
173 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
175 skb_shinfo(skb)->gso_size = hdr->gso_size;
176 if (skb_shinfo(skb)->gso_size == 0) {
177 if (net_ratelimit())
178 printk(KERN_WARNING "%s: zero gso size.\n",
179 dev->name);
180 goto frame_err;
183 /* Header must be checked, and gso_segs computed. */
184 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
185 skb_shinfo(skb)->gso_segs = 0;
188 netif_receive_skb(skb);
189 return;
191 frame_err:
192 dev->stats.rx_frame_errors++;
193 drop:
194 dev_kfree_skb(skb);
197 static void try_fill_recv(struct virtnet_info *vi)
199 struct sk_buff *skb;
200 struct scatterlist sg[2+MAX_SKB_FRAGS];
201 int num, err, i;
203 sg_init_table(sg, 2+MAX_SKB_FRAGS);
204 for (;;) {
205 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
206 if (unlikely(!skb))
207 break;
209 skb_put(skb, MAX_PACKET_LEN);
210 vnet_hdr_to_sg(sg, skb);
212 if (vi->big_packets) {
213 for (i = 0; i < MAX_SKB_FRAGS; i++) {
214 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
215 f->page = get_a_page(vi, GFP_ATOMIC);
216 if (!f->page)
217 break;
219 f->page_offset = 0;
220 f->size = PAGE_SIZE;
222 skb->data_len += PAGE_SIZE;
223 skb->len += PAGE_SIZE;
225 skb_shinfo(skb)->nr_frags++;
229 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
230 skb_queue_head(&vi->recv, skb);
232 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
233 if (err) {
234 skb_unlink(skb, &vi->recv);
235 kfree_skb(skb);
236 break;
238 vi->num++;
240 if (unlikely(vi->num > vi->max))
241 vi->max = vi->num;
242 vi->rvq->vq_ops->kick(vi->rvq);
245 static void skb_recv_done(struct virtqueue *rvq)
247 struct virtnet_info *vi = rvq->vdev->priv;
248 /* Schedule NAPI, Suppress further interrupts if successful. */
249 if (netif_rx_schedule_prep(vi->dev, &vi->napi)) {
250 rvq->vq_ops->disable_cb(rvq);
251 __netif_rx_schedule(vi->dev, &vi->napi);
255 static int virtnet_poll(struct napi_struct *napi, int budget)
257 struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
258 struct sk_buff *skb = NULL;
259 unsigned int len, received = 0;
261 again:
262 while (received < budget &&
263 (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
264 __skb_unlink(skb, &vi->recv);
265 receive_skb(vi->dev, skb, len);
266 vi->num--;
267 received++;
270 /* FIXME: If we oom and completely run out of inbufs, we need
271 * to start a timer trying to fill more. */
272 if (vi->num < vi->max / 2)
273 try_fill_recv(vi);
275 /* Out of packets? */
276 if (received < budget) {
277 netif_rx_complete(vi->dev, napi);
278 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
279 && napi_schedule_prep(napi)) {
280 vi->rvq->vq_ops->disable_cb(vi->rvq);
281 __netif_rx_schedule(vi->dev, napi);
282 goto again;
286 return received;
289 static void free_old_xmit_skbs(struct virtnet_info *vi)
291 struct sk_buff *skb;
292 unsigned int len;
294 while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
295 pr_debug("Sent skb %p\n", skb);
296 __skb_unlink(skb, &vi->send);
297 vi->dev->stats.tx_bytes += skb->len;
298 vi->dev->stats.tx_packets++;
299 kfree_skb(skb);
303 /* If the virtio transport doesn't always notify us when all in-flight packets
304 * are consumed, we fall back to using this function on a timer to free them. */
305 static void xmit_free(unsigned long data)
307 struct virtnet_info *vi = (void *)data;
309 netif_tx_lock(vi->dev);
311 free_old_xmit_skbs(vi);
313 if (!skb_queue_empty(&vi->send))
314 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
316 netif_tx_unlock(vi->dev);
319 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
321 int num, err;
322 struct scatterlist sg[2+MAX_SKB_FRAGS];
323 struct virtio_net_hdr *hdr;
324 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
326 sg_init_table(sg, 2+MAX_SKB_FRAGS);
328 pr_debug("%s: xmit %p " MAC_FMT "\n", vi->dev->name, skb,
329 dest[0], dest[1], dest[2],
330 dest[3], dest[4], dest[5]);
332 /* Encode metadata header at front. */
333 hdr = skb_vnet_hdr(skb);
334 if (skb->ip_summed == CHECKSUM_PARTIAL) {
335 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
336 hdr->csum_start = skb->csum_start - skb_headroom(skb);
337 hdr->csum_offset = skb->csum_offset;
338 } else {
339 hdr->flags = 0;
340 hdr->csum_offset = hdr->csum_start = 0;
343 if (skb_is_gso(skb)) {
344 hdr->hdr_len = skb_transport_header(skb) - skb->data;
345 hdr->gso_size = skb_shinfo(skb)->gso_size;
346 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
347 hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
348 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
349 hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
350 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
351 hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
352 else
353 BUG();
354 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
355 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
356 } else {
357 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
358 hdr->gso_size = hdr->hdr_len = 0;
361 vnet_hdr_to_sg(sg, skb);
362 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
364 err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
365 if (!err && !vi->free_in_tasklet)
366 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
368 return err;
371 static void xmit_tasklet(unsigned long data)
373 struct virtnet_info *vi = (void *)data;
375 netif_tx_lock_bh(vi->dev);
376 if (vi->last_xmit_skb && xmit_skb(vi, vi->last_xmit_skb) == 0) {
377 vi->svq->vq_ops->kick(vi->svq);
378 vi->last_xmit_skb = NULL;
380 if (vi->free_in_tasklet)
381 free_old_xmit_skbs(vi);
382 netif_tx_unlock_bh(vi->dev);
385 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
387 struct virtnet_info *vi = netdev_priv(dev);
389 again:
390 /* Free up any pending old buffers before queueing new ones. */
391 free_old_xmit_skbs(vi);
393 /* If we has a buffer left over from last time, send it now. */
394 if (unlikely(vi->last_xmit_skb) &&
395 xmit_skb(vi, vi->last_xmit_skb) != 0)
396 goto stop_queue;
398 vi->last_xmit_skb = NULL;
400 /* Put new one in send queue and do transmit */
401 if (likely(skb)) {
402 __skb_queue_head(&vi->send, skb);
403 if (xmit_skb(vi, skb) != 0) {
404 vi->last_xmit_skb = skb;
405 skb = NULL;
406 goto stop_queue;
409 done:
410 vi->svq->vq_ops->kick(vi->svq);
411 return NETDEV_TX_OK;
413 stop_queue:
414 pr_debug("%s: virtio not prepared to send\n", dev->name);
415 netif_stop_queue(dev);
417 /* Activate callback for using skbs: if this returns false it
418 * means some were used in the meantime. */
419 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
420 vi->svq->vq_ops->disable_cb(vi->svq);
421 netif_start_queue(dev);
422 goto again;
424 if (skb) {
425 /* Drop this skb: we only queue one. */
426 vi->dev->stats.tx_dropped++;
427 kfree_skb(skb);
429 goto done;
432 #ifdef CONFIG_NET_POLL_CONTROLLER
433 static void virtnet_netpoll(struct net_device *dev)
435 struct virtnet_info *vi = netdev_priv(dev);
437 napi_schedule(&vi->napi);
439 #endif
441 static int virtnet_open(struct net_device *dev)
443 struct virtnet_info *vi = netdev_priv(dev);
445 napi_enable(&vi->napi);
447 /* If all buffers were filled by other side before we napi_enabled, we
448 * won't get another interrupt, so process any outstanding packets
449 * now. virtnet_poll wants re-enable the queue, so we disable here.
450 * We synchronize against interrupts via NAPI_STATE_SCHED */
451 if (netif_rx_schedule_prep(dev, &vi->napi)) {
452 vi->rvq->vq_ops->disable_cb(vi->rvq);
453 __netif_rx_schedule(dev, &vi->napi);
455 return 0;
458 static int virtnet_close(struct net_device *dev)
460 struct virtnet_info *vi = netdev_priv(dev);
462 napi_disable(&vi->napi);
464 return 0;
467 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
469 struct virtnet_info *vi = netdev_priv(dev);
470 struct virtio_device *vdev = vi->vdev;
472 if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
473 return -ENOSYS;
475 return ethtool_op_set_tx_hw_csum(dev, data);
478 static struct ethtool_ops virtnet_ethtool_ops = {
479 .set_tx_csum = virtnet_set_tx_csum,
480 .set_sg = ethtool_op_set_sg,
483 static int virtnet_probe(struct virtio_device *vdev)
485 int err;
486 struct net_device *dev;
487 struct virtnet_info *vi;
489 /* Allocate ourselves a network device with room for our info */
490 dev = alloc_etherdev(sizeof(struct virtnet_info));
491 if (!dev)
492 return -ENOMEM;
494 /* Set up network device as normal. */
495 dev->open = virtnet_open;
496 dev->stop = virtnet_close;
497 dev->hard_start_xmit = start_xmit;
498 dev->features = NETIF_F_HIGHDMA;
499 #ifdef CONFIG_NET_POLL_CONTROLLER
500 dev->poll_controller = virtnet_netpoll;
501 #endif
502 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
503 SET_NETDEV_DEV(dev, &vdev->dev);
505 /* Do we support "hardware" checksums? */
506 if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
507 /* This opens up the world of extra features. */
508 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
509 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
510 dev->features |= NETIF_F_TSO | NETIF_F_UFO
511 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
513 /* Individual feature bits: what can host handle? */
514 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
515 dev->features |= NETIF_F_TSO;
516 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
517 dev->features |= NETIF_F_TSO6;
518 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
519 dev->features |= NETIF_F_TSO_ECN;
520 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
521 dev->features |= NETIF_F_UFO;
524 /* Configuration may specify what MAC to use. Otherwise random. */
525 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
526 vdev->config->get(vdev,
527 offsetof(struct virtio_net_config, mac),
528 dev->dev_addr, dev->addr_len);
529 } else
530 random_ether_addr(dev->dev_addr);
532 /* Set up our device-specific information */
533 vi = netdev_priv(dev);
534 netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
535 vi->dev = dev;
536 vi->vdev = vdev;
537 vdev->priv = vi;
538 vi->pages = NULL;
540 /* If they give us a callback when all buffers are done, we don't need
541 * the timer. */
542 vi->free_in_tasklet = virtio_has_feature(vdev,VIRTIO_F_NOTIFY_ON_EMPTY);
544 /* If we can receive ANY GSO packets, we must allocate large ones. */
545 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
546 || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
547 || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
548 vi->big_packets = true;
550 /* We expect two virtqueues, receive then send. */
551 vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
552 if (IS_ERR(vi->rvq)) {
553 err = PTR_ERR(vi->rvq);
554 goto free;
557 vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
558 if (IS_ERR(vi->svq)) {
559 err = PTR_ERR(vi->svq);
560 goto free_recv;
563 /* Initialize our empty receive and send queues. */
564 skb_queue_head_init(&vi->recv);
565 skb_queue_head_init(&vi->send);
567 tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
569 if (!vi->free_in_tasklet)
570 setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
572 err = register_netdev(dev);
573 if (err) {
574 pr_debug("virtio_net: registering device failed\n");
575 goto free_send;
578 /* Last of all, set up some receive buffers. */
579 try_fill_recv(vi);
581 /* If we didn't even get one input buffer, we're useless. */
582 if (vi->num == 0) {
583 err = -ENOMEM;
584 goto unregister;
587 pr_debug("virtnet: registered device %s\n", dev->name);
588 return 0;
590 unregister:
591 unregister_netdev(dev);
592 free_send:
593 vdev->config->del_vq(vi->svq);
594 free_recv:
595 vdev->config->del_vq(vi->rvq);
596 free:
597 free_netdev(dev);
598 return err;
601 static void virtnet_remove(struct virtio_device *vdev)
603 struct virtnet_info *vi = vdev->priv;
604 struct sk_buff *skb;
606 /* Stop all the virtqueues. */
607 vdev->config->reset(vdev);
609 if (!vi->free_in_tasklet)
610 del_timer_sync(&vi->xmit_free_timer);
612 /* Free our skbs in send and recv queues, if any. */
613 while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
614 kfree_skb(skb);
615 vi->num--;
617 __skb_queue_purge(&vi->send);
619 BUG_ON(vi->num != 0);
621 vdev->config->del_vq(vi->svq);
622 vdev->config->del_vq(vi->rvq);
623 unregister_netdev(vi->dev);
625 while (vi->pages)
626 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
628 free_netdev(vi->dev);
631 static struct virtio_device_id id_table[] = {
632 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
633 { 0 },
636 static unsigned int features[] = {
637 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
638 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
639 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
640 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
641 VIRTIO_NET_F_GUEST_ECN, /* We don't yet handle UFO input. */
642 VIRTIO_F_NOTIFY_ON_EMPTY,
645 static struct virtio_driver virtio_net = {
646 .feature_table = features,
647 .feature_table_size = ARRAY_SIZE(features),
648 .driver.name = KBUILD_MODNAME,
649 .driver.owner = THIS_MODULE,
650 .id_table = id_table,
651 .probe = virtnet_probe,
652 .remove = __devexit_p(virtnet_remove),
655 static int __init init(void)
657 return register_virtio_driver(&virtio_net);
660 static void __exit fini(void)
662 unregister_virtio_driver(&virtio_net);
664 module_init(init);
665 module_exit(fini);
667 MODULE_DEVICE_TABLE(virtio, id_table);
668 MODULE_DESCRIPTION("Virtio network driver");
669 MODULE_LICENSE("GPL");