xref: /openbmc/linux/drivers/net/virtio_net.c (revision 9f4d26d0f3016cf8813977d624751b94465fa317)
1 /* A simple network driver using virtio.
2  *
3  * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
4  *
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.
9  *
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.
14  *
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
18  */
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>
27 
28 static int napi_weight = 128;
29 module_param(napi_weight, int, 0444);
30 
31 static int csum = 1, gso = 1;
32 module_param(csum, bool, 0444);
33 module_param(gso, bool, 0444);
34 
35 /* FIXME: MTU in config. */
36 #define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN)
37 #define GOOD_COPY_LEN	128
38 
39 struct virtnet_info
40 {
41 	struct virtio_device *vdev;
42 	struct virtqueue *rvq, *svq;
43 	struct net_device *dev;
44 	struct napi_struct napi;
45 	unsigned int status;
46 
47 	/* The skb we couldn't send because buffers were full. */
48 	struct sk_buff *last_xmit_skb;
49 
50 	/* If we need to free in a timer, this is it. */
51 	struct timer_list xmit_free_timer;
52 
53 	/* Number of input buffers, and max we've ever had. */
54 	unsigned int num, max;
55 
56 	/* For cleaning up after transmission. */
57 	struct tasklet_struct tasklet;
58 	bool free_in_tasklet;
59 
60 	/* I like... big packets and I cannot lie! */
61 	bool big_packets;
62 
63 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
64 	bool mergeable_rx_bufs;
65 
66 	/* Receive & send queues. */
67 	struct sk_buff_head recv;
68 	struct sk_buff_head send;
69 
70 	/* Chain pages by the private ptr. */
71 	struct page *pages;
72 };
73 
74 static inline void *skb_vnet_hdr(struct sk_buff *skb)
75 {
76 	return (struct virtio_net_hdr *)skb->cb;
77 }
78 
79 static void give_a_page(struct virtnet_info *vi, struct page *page)
80 {
81 	page->private = (unsigned long)vi->pages;
82 	vi->pages = page;
83 }
84 
85 static void trim_pages(struct virtnet_info *vi, struct sk_buff *skb)
86 {
87 	unsigned int i;
88 
89 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
90 		give_a_page(vi, skb_shinfo(skb)->frags[i].page);
91 	skb_shinfo(skb)->nr_frags = 0;
92 	skb->data_len = 0;
93 }
94 
95 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
96 {
97 	struct page *p = vi->pages;
98 
99 	if (p)
100 		vi->pages = (struct page *)p->private;
101 	else
102 		p = alloc_page(gfp_mask);
103 	return p;
104 }
105 
106 static void skb_xmit_done(struct virtqueue *svq)
107 {
108 	struct virtnet_info *vi = svq->vdev->priv;
109 
110 	/* Suppress further interrupts. */
111 	svq->vq_ops->disable_cb(svq);
112 
113 	/* We were probably waiting for more output buffers. */
114 	netif_wake_queue(vi->dev);
115 
116 	/* Make sure we re-xmit last_xmit_skb: if there are no more packets
117 	 * queued, start_xmit won't be called. */
118 	tasklet_schedule(&vi->tasklet);
119 }
120 
121 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
122 			unsigned len)
123 {
124 	struct virtnet_info *vi = netdev_priv(dev);
125 	struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
126 	int err;
127 	int i;
128 
129 	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
130 		pr_debug("%s: short packet %i\n", dev->name, len);
131 		dev->stats.rx_length_errors++;
132 		goto drop;
133 	}
134 
135 	if (vi->mergeable_rx_bufs) {
136 		struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
137 		unsigned int copy;
138 		char *p = page_address(skb_shinfo(skb)->frags[0].page);
139 
140 		if (len > PAGE_SIZE)
141 			len = PAGE_SIZE;
142 		len -= sizeof(struct virtio_net_hdr_mrg_rxbuf);
143 
144 		memcpy(hdr, p, sizeof(*mhdr));
145 		p += sizeof(*mhdr);
146 
147 		copy = len;
148 		if (copy > skb_tailroom(skb))
149 			copy = skb_tailroom(skb);
150 
151 		memcpy(skb_put(skb, copy), p, copy);
152 
153 		len -= copy;
154 
155 		if (!len) {
156 			give_a_page(vi, skb_shinfo(skb)->frags[0].page);
157 			skb_shinfo(skb)->nr_frags--;
158 		} else {
159 			skb_shinfo(skb)->frags[0].page_offset +=
160 				sizeof(*mhdr) + copy;
161 			skb_shinfo(skb)->frags[0].size = len;
162 			skb->data_len += len;
163 			skb->len += len;
164 		}
165 
166 		while (--mhdr->num_buffers) {
167 			struct sk_buff *nskb;
168 
169 			i = skb_shinfo(skb)->nr_frags;
170 			if (i >= MAX_SKB_FRAGS) {
171 				pr_debug("%s: packet too long %d\n", dev->name,
172 					 len);
173 				dev->stats.rx_length_errors++;
174 				goto drop;
175 			}
176 
177 			nskb = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
178 			if (!nskb) {
179 				pr_debug("%s: rx error: %d buffers missing\n",
180 					 dev->name, mhdr->num_buffers);
181 				dev->stats.rx_length_errors++;
182 				goto drop;
183 			}
184 
185 			__skb_unlink(nskb, &vi->recv);
186 			vi->num--;
187 
188 			skb_shinfo(skb)->frags[i] = skb_shinfo(nskb)->frags[0];
189 			skb_shinfo(nskb)->nr_frags = 0;
190 			kfree_skb(nskb);
191 
192 			if (len > PAGE_SIZE)
193 				len = PAGE_SIZE;
194 
195 			skb_shinfo(skb)->frags[i].size = len;
196 			skb_shinfo(skb)->nr_frags++;
197 			skb->data_len += len;
198 			skb->len += len;
199 		}
200 	} else {
201 		len -= sizeof(struct virtio_net_hdr);
202 
203 		if (len <= MAX_PACKET_LEN)
204 			trim_pages(vi, skb);
205 
206 		err = pskb_trim(skb, len);
207 		if (err) {
208 			pr_debug("%s: pskb_trim failed %i %d\n", dev->name,
209 				 len, err);
210 			dev->stats.rx_dropped++;
211 			goto drop;
212 		}
213 	}
214 
215 	skb->truesize += skb->data_len;
216 	dev->stats.rx_bytes += skb->len;
217 	dev->stats.rx_packets++;
218 
219 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
220 		pr_debug("Needs csum!\n");
221 		if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
222 			goto frame_err;
223 	}
224 
225 	skb->protocol = eth_type_trans(skb, dev);
226 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
227 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
228 
229 	if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
230 		pr_debug("GSO!\n");
231 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
232 		case VIRTIO_NET_HDR_GSO_TCPV4:
233 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
234 			break;
235 		case VIRTIO_NET_HDR_GSO_UDP:
236 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
237 			break;
238 		case VIRTIO_NET_HDR_GSO_TCPV6:
239 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
240 			break;
241 		default:
242 			if (net_ratelimit())
243 				printk(KERN_WARNING "%s: bad gso type %u.\n",
244 				       dev->name, hdr->gso_type);
245 			goto frame_err;
246 		}
247 
248 		if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
249 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
250 
251 		skb_shinfo(skb)->gso_size = hdr->gso_size;
252 		if (skb_shinfo(skb)->gso_size == 0) {
253 			if (net_ratelimit())
254 				printk(KERN_WARNING "%s: zero gso size.\n",
255 				       dev->name);
256 			goto frame_err;
257 		}
258 
259 		/* Header must be checked, and gso_segs computed. */
260 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
261 		skb_shinfo(skb)->gso_segs = 0;
262 	}
263 
264 	netif_receive_skb(skb);
265 	return;
266 
267 frame_err:
268 	dev->stats.rx_frame_errors++;
269 drop:
270 	dev_kfree_skb(skb);
271 }
272 
273 static void try_fill_recv_maxbufs(struct virtnet_info *vi)
274 {
275 	struct sk_buff *skb;
276 	struct scatterlist sg[2+MAX_SKB_FRAGS];
277 	int num, err, i;
278 
279 	sg_init_table(sg, 2+MAX_SKB_FRAGS);
280 	for (;;) {
281 		struct virtio_net_hdr *hdr;
282 
283 		skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
284 		if (unlikely(!skb))
285 			break;
286 
287 		skb_put(skb, MAX_PACKET_LEN);
288 
289 		hdr = skb_vnet_hdr(skb);
290 		sg_init_one(sg, hdr, sizeof(*hdr));
291 
292 		if (vi->big_packets) {
293 			for (i = 0; i < MAX_SKB_FRAGS; i++) {
294 				skb_frag_t *f = &skb_shinfo(skb)->frags[i];
295 				f->page = get_a_page(vi, GFP_ATOMIC);
296 				if (!f->page)
297 					break;
298 
299 				f->page_offset = 0;
300 				f->size = PAGE_SIZE;
301 
302 				skb->data_len += PAGE_SIZE;
303 				skb->len += PAGE_SIZE;
304 
305 				skb_shinfo(skb)->nr_frags++;
306 			}
307 		}
308 
309 		num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
310 		skb_queue_head(&vi->recv, skb);
311 
312 		err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
313 		if (err) {
314 			skb_unlink(skb, &vi->recv);
315 			trim_pages(vi, skb);
316 			kfree_skb(skb);
317 			break;
318 		}
319 		vi->num++;
320 	}
321 	if (unlikely(vi->num > vi->max))
322 		vi->max = vi->num;
323 	vi->rvq->vq_ops->kick(vi->rvq);
324 }
325 
326 static void try_fill_recv(struct virtnet_info *vi)
327 {
328 	struct sk_buff *skb;
329 	struct scatterlist sg[1];
330 	int err;
331 
332 	if (!vi->mergeable_rx_bufs) {
333 		try_fill_recv_maxbufs(vi);
334 		return;
335 	}
336 
337 	for (;;) {
338 		skb_frag_t *f;
339 
340 		skb = netdev_alloc_skb(vi->dev, GOOD_COPY_LEN + NET_IP_ALIGN);
341 		if (unlikely(!skb))
342 			break;
343 
344 		skb_reserve(skb, NET_IP_ALIGN);
345 
346 		f = &skb_shinfo(skb)->frags[0];
347 		f->page = get_a_page(vi, GFP_ATOMIC);
348 		if (!f->page) {
349 			kfree_skb(skb);
350 			break;
351 		}
352 
353 		f->page_offset = 0;
354 		f->size = PAGE_SIZE;
355 
356 		skb_shinfo(skb)->nr_frags++;
357 
358 		sg_init_one(sg, page_address(f->page), PAGE_SIZE);
359 		skb_queue_head(&vi->recv, skb);
360 
361 		err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
362 		if (err) {
363 			skb_unlink(skb, &vi->recv);
364 			kfree_skb(skb);
365 			break;
366 		}
367 		vi->num++;
368 	}
369 	if (unlikely(vi->num > vi->max))
370 		vi->max = vi->num;
371 	vi->rvq->vq_ops->kick(vi->rvq);
372 }
373 
374 static void skb_recv_done(struct virtqueue *rvq)
375 {
376 	struct virtnet_info *vi = rvq->vdev->priv;
377 	/* Schedule NAPI, Suppress further interrupts if successful. */
378 	if (napi_schedule_prep(&vi->napi)) {
379 		rvq->vq_ops->disable_cb(rvq);
380 		__napi_schedule(&vi->napi);
381 	}
382 }
383 
384 static int virtnet_poll(struct napi_struct *napi, int budget)
385 {
386 	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
387 	struct sk_buff *skb = NULL;
388 	unsigned int len, received = 0;
389 
390 again:
391 	while (received < budget &&
392 	       (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
393 		__skb_unlink(skb, &vi->recv);
394 		receive_skb(vi->dev, skb, len);
395 		vi->num--;
396 		received++;
397 	}
398 
399 	/* FIXME: If we oom and completely run out of inbufs, we need
400 	 * to start a timer trying to fill more. */
401 	if (vi->num < vi->max / 2)
402 		try_fill_recv(vi);
403 
404 	/* Out of packets? */
405 	if (received < budget) {
406 		napi_complete(napi);
407 		if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
408 		    && napi_schedule_prep(napi)) {
409 			vi->rvq->vq_ops->disable_cb(vi->rvq);
410 			__napi_schedule(napi);
411 			goto again;
412 		}
413 	}
414 
415 	return received;
416 }
417 
418 static void free_old_xmit_skbs(struct virtnet_info *vi)
419 {
420 	struct sk_buff *skb;
421 	unsigned int len;
422 
423 	while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
424 		pr_debug("Sent skb %p\n", skb);
425 		__skb_unlink(skb, &vi->send);
426 		vi->dev->stats.tx_bytes += skb->len;
427 		vi->dev->stats.tx_packets++;
428 		kfree_skb(skb);
429 	}
430 }
431 
432 /* If the virtio transport doesn't always notify us when all in-flight packets
433  * are consumed, we fall back to using this function on a timer to free them. */
434 static void xmit_free(unsigned long data)
435 {
436 	struct virtnet_info *vi = (void *)data;
437 
438 	netif_tx_lock(vi->dev);
439 
440 	free_old_xmit_skbs(vi);
441 
442 	if (!skb_queue_empty(&vi->send))
443 		mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
444 
445 	netif_tx_unlock(vi->dev);
446 }
447 
448 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
449 {
450 	int num, err;
451 	struct scatterlist sg[2+MAX_SKB_FRAGS];
452 	struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
453 	struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
454 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
455 
456 	sg_init_table(sg, 2+MAX_SKB_FRAGS);
457 
458 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
459 
460 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
461 		hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
462 		hdr->csum_start = skb->csum_start - skb_headroom(skb);
463 		hdr->csum_offset = skb->csum_offset;
464 	} else {
465 		hdr->flags = 0;
466 		hdr->csum_offset = hdr->csum_start = 0;
467 	}
468 
469 	if (skb_is_gso(skb)) {
470 		hdr->hdr_len = skb_transport_header(skb) - skb->data;
471 		hdr->gso_size = skb_shinfo(skb)->gso_size;
472 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
473 			hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
474 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
475 			hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
476 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
477 			hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
478 		else
479 			BUG();
480 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
481 			hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
482 	} else {
483 		hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
484 		hdr->gso_size = hdr->hdr_len = 0;
485 	}
486 
487 	mhdr->num_buffers = 0;
488 
489 	/* Encode metadata header at front. */
490 	if (vi->mergeable_rx_bufs)
491 		sg_init_one(sg, mhdr, sizeof(*mhdr));
492 	else
493 		sg_init_one(sg, hdr, sizeof(*hdr));
494 
495 	num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
496 
497 	err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
498 	if (!err && !vi->free_in_tasklet)
499 		mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
500 
501 	return err;
502 }
503 
504 static void xmit_tasklet(unsigned long data)
505 {
506 	struct virtnet_info *vi = (void *)data;
507 
508 	netif_tx_lock_bh(vi->dev);
509 	if (vi->last_xmit_skb && xmit_skb(vi, vi->last_xmit_skb) == 0) {
510 		vi->svq->vq_ops->kick(vi->svq);
511 		vi->last_xmit_skb = NULL;
512 	}
513 	if (vi->free_in_tasklet)
514 		free_old_xmit_skbs(vi);
515 	netif_tx_unlock_bh(vi->dev);
516 }
517 
518 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
519 {
520 	struct virtnet_info *vi = netdev_priv(dev);
521 
522 again:
523 	/* Free up any pending old buffers before queueing new ones. */
524 	free_old_xmit_skbs(vi);
525 
526 	/* If we has a buffer left over from last time, send it now. */
527 	if (unlikely(vi->last_xmit_skb) &&
528 	    xmit_skb(vi, vi->last_xmit_skb) != 0)
529 		goto stop_queue;
530 
531 	vi->last_xmit_skb = NULL;
532 
533 	/* Put new one in send queue and do transmit */
534 	if (likely(skb)) {
535 		__skb_queue_head(&vi->send, skb);
536 		if (xmit_skb(vi, skb) != 0) {
537 			vi->last_xmit_skb = skb;
538 			skb = NULL;
539 			goto stop_queue;
540 		}
541 	}
542 done:
543 	vi->svq->vq_ops->kick(vi->svq);
544 	return NETDEV_TX_OK;
545 
546 stop_queue:
547 	pr_debug("%s: virtio not prepared to send\n", dev->name);
548 	netif_stop_queue(dev);
549 
550 	/* Activate callback for using skbs: if this returns false it
551 	 * means some were used in the meantime. */
552 	if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
553 		vi->svq->vq_ops->disable_cb(vi->svq);
554 		netif_start_queue(dev);
555 		goto again;
556 	}
557 	if (skb) {
558 		/* Drop this skb: we only queue one. */
559 		vi->dev->stats.tx_dropped++;
560 		kfree_skb(skb);
561 	}
562 	goto done;
563 }
564 
565 #ifdef CONFIG_NET_POLL_CONTROLLER
566 static void virtnet_netpoll(struct net_device *dev)
567 {
568 	struct virtnet_info *vi = netdev_priv(dev);
569 
570 	napi_schedule(&vi->napi);
571 }
572 #endif
573 
574 static int virtnet_open(struct net_device *dev)
575 {
576 	struct virtnet_info *vi = netdev_priv(dev);
577 
578 	napi_enable(&vi->napi);
579 
580 	/* If all buffers were filled by other side before we napi_enabled, we
581 	 * won't get another interrupt, so process any outstanding packets
582 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
583 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
584 	if (napi_schedule_prep(&vi->napi)) {
585 		vi->rvq->vq_ops->disable_cb(vi->rvq);
586 		__napi_schedule(&vi->napi);
587 	}
588 	return 0;
589 }
590 
591 static int virtnet_close(struct net_device *dev)
592 {
593 	struct virtnet_info *vi = netdev_priv(dev);
594 
595 	napi_disable(&vi->napi);
596 
597 	return 0;
598 }
599 
600 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
601 {
602 	struct virtnet_info *vi = netdev_priv(dev);
603 	struct virtio_device *vdev = vi->vdev;
604 
605 	if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
606 		return -ENOSYS;
607 
608 	return ethtool_op_set_tx_hw_csum(dev, data);
609 }
610 
611 static struct ethtool_ops virtnet_ethtool_ops = {
612 	.set_tx_csum = virtnet_set_tx_csum,
613 	.set_sg = ethtool_op_set_sg,
614 	.set_tso = ethtool_op_set_tso,
615 	.get_link = ethtool_op_get_link,
616 };
617 
618 #define MIN_MTU 68
619 #define MAX_MTU 65535
620 
621 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
622 {
623 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
624 		return -EINVAL;
625 	dev->mtu = new_mtu;
626 	return 0;
627 }
628 
629 static const struct net_device_ops virtnet_netdev = {
630 	.ndo_open            = virtnet_open,
631 	.ndo_stop   	     = virtnet_close,
632 	.ndo_start_xmit      = start_xmit,
633 	.ndo_validate_addr   = eth_validate_addr,
634 	.ndo_set_mac_address = eth_mac_addr,
635 	.ndo_change_mtu	     = virtnet_change_mtu,
636 #ifdef CONFIG_NET_POLL_CONTROLLER
637 	.ndo_poll_controller = virtnet_netpoll,
638 #endif
639 };
640 
641 static void virtnet_update_status(struct virtnet_info *vi)
642 {
643 	u16 v;
644 
645 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
646 		return;
647 
648 	vi->vdev->config->get(vi->vdev,
649 			      offsetof(struct virtio_net_config, status),
650 			      &v, sizeof(v));
651 
652 	/* Ignore unknown (future) status bits */
653 	v &= VIRTIO_NET_S_LINK_UP;
654 
655 	if (vi->status == v)
656 		return;
657 
658 	vi->status = v;
659 
660 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
661 		netif_carrier_on(vi->dev);
662 		netif_wake_queue(vi->dev);
663 	} else {
664 		netif_carrier_off(vi->dev);
665 		netif_stop_queue(vi->dev);
666 	}
667 }
668 
669 static void virtnet_config_changed(struct virtio_device *vdev)
670 {
671 	struct virtnet_info *vi = vdev->priv;
672 
673 	virtnet_update_status(vi);
674 }
675 
676 static int virtnet_probe(struct virtio_device *vdev)
677 {
678 	int err;
679 	struct net_device *dev;
680 	struct virtnet_info *vi;
681 
682 	/* Allocate ourselves a network device with room for our info */
683 	dev = alloc_etherdev(sizeof(struct virtnet_info));
684 	if (!dev)
685 		return -ENOMEM;
686 
687 	/* Set up network device as normal. */
688 	dev->netdev_ops = &virtnet_netdev;
689 	dev->features = NETIF_F_HIGHDMA;
690 	SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
691 	SET_NETDEV_DEV(dev, &vdev->dev);
692 
693 	/* Do we support "hardware" checksums? */
694 	if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
695 		/* This opens up the world of extra features. */
696 		dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
697 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
698 			dev->features |= NETIF_F_TSO | NETIF_F_UFO
699 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
700 		}
701 		/* Individual feature bits: what can host handle? */
702 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
703 			dev->features |= NETIF_F_TSO;
704 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
705 			dev->features |= NETIF_F_TSO6;
706 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
707 			dev->features |= NETIF_F_TSO_ECN;
708 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
709 			dev->features |= NETIF_F_UFO;
710 	}
711 
712 	/* Configuration may specify what MAC to use.  Otherwise random. */
713 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
714 		vdev->config->get(vdev,
715 				  offsetof(struct virtio_net_config, mac),
716 				  dev->dev_addr, dev->addr_len);
717 	} else
718 		random_ether_addr(dev->dev_addr);
719 
720 	/* Set up our device-specific information */
721 	vi = netdev_priv(dev);
722 	netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
723 	vi->dev = dev;
724 	vi->vdev = vdev;
725 	vdev->priv = vi;
726 	vi->pages = NULL;
727 
728 	/* If they give us a callback when all buffers are done, we don't need
729 	 * the timer. */
730 	vi->free_in_tasklet = virtio_has_feature(vdev,VIRTIO_F_NOTIFY_ON_EMPTY);
731 
732 	/* If we can receive ANY GSO packets, we must allocate large ones. */
733 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
734 	    || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
735 	    || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
736 		vi->big_packets = true;
737 
738 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
739 		vi->mergeable_rx_bufs = true;
740 
741 	/* We expect two virtqueues, receive then send. */
742 	vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
743 	if (IS_ERR(vi->rvq)) {
744 		err = PTR_ERR(vi->rvq);
745 		goto free;
746 	}
747 
748 	vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
749 	if (IS_ERR(vi->svq)) {
750 		err = PTR_ERR(vi->svq);
751 		goto free_recv;
752 	}
753 
754 	/* Initialize our empty receive and send queues. */
755 	skb_queue_head_init(&vi->recv);
756 	skb_queue_head_init(&vi->send);
757 
758 	tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
759 
760 	if (!vi->free_in_tasklet)
761 		setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
762 
763 	err = register_netdev(dev);
764 	if (err) {
765 		pr_debug("virtio_net: registering device failed\n");
766 		goto free_send;
767 	}
768 
769 	/* Last of all, set up some receive buffers. */
770 	try_fill_recv(vi);
771 
772 	/* If we didn't even get one input buffer, we're useless. */
773 	if (vi->num == 0) {
774 		err = -ENOMEM;
775 		goto unregister;
776 	}
777 
778 	vi->status = VIRTIO_NET_S_LINK_UP;
779 	virtnet_update_status(vi);
780 
781 	pr_debug("virtnet: registered device %s\n", dev->name);
782 	return 0;
783 
784 unregister:
785 	unregister_netdev(dev);
786 free_send:
787 	vdev->config->del_vq(vi->svq);
788 free_recv:
789 	vdev->config->del_vq(vi->rvq);
790 free:
791 	free_netdev(dev);
792 	return err;
793 }
794 
795 static void virtnet_remove(struct virtio_device *vdev)
796 {
797 	struct virtnet_info *vi = vdev->priv;
798 	struct sk_buff *skb;
799 
800 	/* Stop all the virtqueues. */
801 	vdev->config->reset(vdev);
802 
803 	if (!vi->free_in_tasklet)
804 		del_timer_sync(&vi->xmit_free_timer);
805 
806 	/* Free our skbs in send and recv queues, if any. */
807 	while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
808 		kfree_skb(skb);
809 		vi->num--;
810 	}
811 	__skb_queue_purge(&vi->send);
812 
813 	BUG_ON(vi->num != 0);
814 
815 	vdev->config->del_vq(vi->svq);
816 	vdev->config->del_vq(vi->rvq);
817 	unregister_netdev(vi->dev);
818 
819 	while (vi->pages)
820 		__free_pages(get_a_page(vi, GFP_KERNEL), 0);
821 
822 	free_netdev(vi->dev);
823 }
824 
825 static struct virtio_device_id id_table[] = {
826 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
827 	{ 0 },
828 };
829 
830 static unsigned int features[] = {
831 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
832 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
833 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
834 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
835 	VIRTIO_NET_F_GUEST_ECN, /* We don't yet handle UFO input. */
836 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS,
837 	VIRTIO_F_NOTIFY_ON_EMPTY,
838 };
839 
840 static struct virtio_driver virtio_net = {
841 	.feature_table = features,
842 	.feature_table_size = ARRAY_SIZE(features),
843 	.driver.name =	KBUILD_MODNAME,
844 	.driver.owner =	THIS_MODULE,
845 	.id_table =	id_table,
846 	.probe =	virtnet_probe,
847 	.remove =	__devexit_p(virtnet_remove),
848 	.config_changed = virtnet_config_changed,
849 };
850 
851 static int __init init(void)
852 {
853 	return register_virtio_driver(&virtio_net);
854 }
855 
856 static void __exit fini(void)
857 {
858 	unregister_virtio_driver(&virtio_net);
859 }
860 module_init(init);
861 module_exit(fini);
862 
863 MODULE_DEVICE_TABLE(virtio, id_table);
864 MODULE_DESCRIPTION("Virtio network driver");
865 MODULE_LICENSE("GPL");
866